Multimodality_lab_SIIM22_github.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 5,
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.7.10"
    },
    "colab": {
      "name": "Multimodality_lab_SIIM22_github.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "accelerator": "GPU",
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "fe261f7e333c49789718b8da1f40a928": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_83266fabcf184519b6c6171dbb04e4a2",
              "IPY_MODEL_d04f41a2f2564080b93fd0f23d4d8842",
              "IPY_MODEL_925254b1e9ef4d96876d923d20e1ec87"
            ],
            "layout": "IPY_MODEL_4e2481b28d9142da9dfd237255fa35a4"
          }
        },
        "83266fabcf184519b6c6171dbb04e4a2": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_ed12f53ef2344a45a99df0a6da4a6e84",
            "placeholder": "​",
            "style": "IPY_MODEL_034b312ee7ec4d309fd8a8b1f289a942",
            "value": "  0%"
          }
        },
        "d04f41a2f2564080b93fd0f23d4d8842": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_73b2e9137f61433fa22a859b39462fbc",
            "max": 3,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_b8d4249c0d544b30bfe7cb655d703d19",
            "value": 0
          }
        },
        "925254b1e9ef4d96876d923d20e1ec87": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_e5e04d54574d4fb8a595e0e4fafbb482",
            "placeholder": "​",
            "style": "IPY_MODEL_266c41902ec24b5385172180bd11c3b5",
            "value": " 0/3 [00:00&lt;?, ?it/s]"
          }
        },
        "4e2481b28d9142da9dfd237255fa35a4": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "ed12f53ef2344a45a99df0a6da4a6e84": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "034b312ee7ec4d309fd8a8b1f289a942": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "73b2e9137f61433fa22a859b39462fbc": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "b8d4249c0d544b30bfe7cb655d703d19": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "e5e04d54574d4fb8a595e0e4fafbb482": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "266c41902ec24b5385172180bd11c3b5": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "4bdbb02bbb2d43f08c9ca0f57a453031": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_f6c41bb9bcf843eba85dfbbdf725f232",
              "IPY_MODEL_79e204ff103d450991ac815e6b4c11ba",
              "IPY_MODEL_24dce259922b49ba8eaa137728c69b78"
            ],
            "layout": "IPY_MODEL_20d0b9894414426493025ef8307e0bb5"
          }
        },
        "f6c41bb9bcf843eba85dfbbdf725f232": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_8b3bbd669a3042d08685b3eeb8bbbe98",
            "placeholder": "​",
            "style": "IPY_MODEL_9d7dbb7b13034e97b7229a0e31761dee",
            "value": "  8%"
          }
        },
        "79e204ff103d450991ac815e6b4c11ba": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_3c5823b55b834d3398e04d84906656d2",
            "max": 12,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_d8495f189f454b449e5bfb3125eccb8b",
            "value": 1
          }
        },
        "24dce259922b49ba8eaa137728c69b78": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_d93e5b276bd341769d15a687d79ee4d0",
            "placeholder": "​",
            "style": "IPY_MODEL_020470f3eec84edea07c31450fa7a964",
            "value": " 1/12 [01:06&lt;12:16, 66.99s/it]"
          }
        },
        "20d0b9894414426493025ef8307e0bb5": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "8b3bbd669a3042d08685b3eeb8bbbe98": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "9d7dbb7b13034e97b7229a0e31761dee": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "3c5823b55b834d3398e04d84906656d2": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "d8495f189f454b449e5bfb3125eccb8b": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "d93e5b276bd341769d15a687d79ee4d0": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "020470f3eec84edea07c31450fa7a964": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        }
      }
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Mayo-Radiology-Informatics-Lab/MIDeL-Chapters/blob/main/chapters/Section_14_Image_Image_multimodality_registration.ipynb)"
      ],
      "metadata": {
        "id": "ghJw2xL8ZPCY"
      },
      "id": "ghJw2xL8ZPCY"
    },
    {
      "cell_type": "markdown",
      "source": [
        "#Multimodal deep learning using PyTorch\n",
        "##SIIM Github Repo\n",
        "###Task\n",
        "Train a multi modal deep learning model to classify MGMT promoter methylation status using different MRI sequences and age.\n",
        "\n",
        "Requirements\n",
        "Basic understanding of machine learning and deep learning\n",
        "Programming in Python\n",
        "Programming in PyTorch\n",
        "Learning objectives\n",
        "At the end of this activity, you will be able to:\n",
        "\n",
        "- Define multimodality and state its importance\n",
        "- Categorize different types of multimodality\n",
        "- Implement Imaging-Imaging multimodality deep leaning in PyTorch\n",
        "- Implement Imaging-Non imaging multimodality deep leaning in PyTorch\n",
        "- Implement multimodal deep leaning with result ensemble in PyTorch\n",
        "\n",
        "###Acknowledgements\n",
        "This Jupyter Notebook was based on code by Shahriar Faghani, MD (Faghani.shahriar@mayo.edu), Gian Marco Conte, MD, PhD (Gianmarco.conte@mayo.edu) and, Mana Moassefi, MD (Moassefi.mana@mayo.edu) developed at Artificial intelligence lab, Mayo Clinic, Rochester\n"
      ],
      "metadata": {
        "id": "Fl_qISCY5xVb"
      },
      "id": "Fl_qISCY5xVb"
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Please make a copy of this notebook for yourself and work with the copied version of the notebook\n"
      ],
      "metadata": {
        "id": "f3qUJPvrc9il"
      },
      "id": "f3qUJPvrc9il"
    },
    {
      "cell_type": "markdown",
      "source": [
        "#14. Multimodal Deep Learning Models\n",
        "\n"
      ],
      "metadata": {
        "id": "6QgJ7B97hq9J"
      },
      "id": "6QgJ7B97hq9J"
    },
    {
      "cell_type": "markdown",
      "source": [
        "####Similar to human physicians, automated detection and classification systems that use both medical imaging data and clinical data from the EHR -- such as patient demographics, previous diagnoses, and laboratory data -- are likely to produce models with better performance. Recent medical imaging literature also shows a similar trend where both EHR and pixel data are exploited in a \"fusion paradigm\" for solving complex tasks that cannot easily be solved by one modality alone (Huang et al. 2020)."
      ],
      "metadata": {
        "id": "ZkgVnPFohzbt"
      },
      "id": "ZkgVnPFohzbt"
    },
    {
      "cell_type": "markdown",
      "source": [
        "###14-A imaging - imaging models:"
      ],
      "metadata": {
        "id": "4rmjTMpiitfW"
      },
      "id": "4rmjTMpiitfW"
    },
    {
      "cell_type": "markdown",
      "source": [
        "####Contemporary medicine relies on the synthesis of a wide variety of information such as imaging data, laboratory data, unstructured narrative data, 1D signals like EKG, and in some cases, audio or observational data. The clinical context of images often plays a critical role in providing diagnosis decisions in medical image interpretation. For example, it has repeatedly been shown that a lack of access to clinical and laboratory data during image interpretation results in lower performance and decreased clinical utility for the referring provider. In a survey of radiologists, the majority (87%) stated that clinical information had a significant impact on interpretation (Boonn and Langlotz 2009). Radiology is not the only imaging-based medical specialty that relies on context for accurate interpretation of imaging data; pathology, ophthalmology, and dermatology also use clinical context for clinical image interpretation. With accurate and relevant information regarding current symptoms and past medical history, physicians are better able to interpret imaging findings in the appropriate clinical context, resulting in a more relevant differential diagnosis, a more useful report for physicians, and a more favorable outcome for the patient. \n",
        "####Convolutional neural network (CNN) models typically use pixel values as inputs, without contextualizing other clinical information as doctors do in clinical practice, which may limit performance. As an example consider the “simple” task in radiology of identifying pneumonia on a chest x-ray (CXR), something that has been achieved by many investigators training deep learning models for automated detection and classification of pathologies. Such applications, however, may ultimately have a limited impact on clinical practice without clinical context such as chief complaint, history of present illness, past medical history, drug history, family history, physical examination findings, and laboratory values. In spite of having imaging findings that can set pneumonia apart from other diagnoses, CXR is nonspecific, and accurate diagnosis requires clinical and laboratory information. CXR findings that suggest pneumonia would be accurate in patients with fevers and elevated white blood cell counts, but in patients with no similar clinical features or laboratory values, similar images might indicate atelectasis, pulmonary edema, interstitial disease, or even lung cancer. \n",
        "\n",
        "####First, we will discuss different fusion techniques in deep learning and machine learning, and then we will discuss some medical multimodality examples, and finally, we will develop a simple multimodal model.\n"
      ],
      "metadata": {
        "id": "moQxFUkHixWb"
      },
      "id": "moQxFUkHixWb"
    },
    {
      "cell_type": "markdown",
      "source": [
        "###Fusion Techniques"
      ],
      "metadata": {
        "id": "dd8R3-9ui87u"
      },
      "id": "dd8R3-9ui87u"
    },
    {
      "cell_type": "markdown",
      "source": [
        "####Image fusion is a technique for combining the information from several imaging data sources acquired from the same or different modalities. A goal of image fusion, especially in medical imaging, is to enhance or complement each data source's features, so that machine learning algorithms can achieve better performance rather than using only one data channel. The expected output will have a richer feature representation than the individual components. The multimodal fusion technique has achieved remarkable success in a variety of applications like medical image segmentation, image classification, and image reconstruction.\n",
        "####Data fusion requires more than one data source (e.g. could be from one exam like T1 and T2 of MRI or different exams like MRI and CT). The traditional image processing literature refers to these as ‘channels’ and we will also use that nomenclature. Data channels can be heterogeneous, complementary, concordant or discordant, synchronous or asynchronous, and redundant with different scales that must be normalized in order to be combined. Medical images such as MRI/CT/PET/Ultrasound also have differences that must be considered when using them in combination. Therefore, the choice of algorithm or architecture depends primarily on the goal of the application and data channel characteristics. To summarize, the advantages of data fusion (Bellot et al. 2002) are:\n",
        "####1) To provide different types of information to help increase certainty about the diagnosis.\n",
        "####2) To combine source features and extract the new or hidden information with increased certainty.\n",
        "####3) To enhance the abstraction and completeness of data by efficiently combining them.\n",
        "####In recent computer vision publications, people investigate the multimodal fusion strategies in the context of the spatio-temporal convolutional neural network (Karpathy et al. 2014) and broadly categorized into four fusion patterns/classes: Early fusion, Late fusion, Joint fusion, and Slow fusion.    \t\n"
      ],
      "metadata": {
        "id": "u2As3ZN2jAQ4"
      },
      "id": "u2As3ZN2jAQ4"
    },
    {
      "cell_type": "markdown",
      "source": [
        "###Early Fusion:"
      ],
      "metadata": {
        "id": "aWKbcfE2jJpo"
      },
      "id": "aWKbcfE2jJpo"
    },
    {
      "cell_type": "markdown",
      "source": [
        "####Early fusion merges the various data channels at the beginning of the machine learning model training. These individual modality features may be joined in different ways, such as by average pooling, weighted concatenation, or gated fusion techniques (Type I or II in Fig. 1). The unimodal local features are concatenated into a joint representation. Therefore, the unified model must ensure that the data/features are correctly aligned in order to enable joint processing. As a result, converting data sources into a single feature vector is a significant challenge in early data fusion approaches. In reality, the conditionally independent criteria between medical data sources are not always true as multiple modalities may have highly correlated features, possibly because of acquisition setting. Therefore, different data collection modalities are typically spatially aligned (known as image registration) with a standard pixel sampling rate (for images or perhaps time for other modalities) to make a common ground for fusion. If the data is aligned correctly, the cross-correlations between data items may be exploited, which provides an opportunity to increase the system performance (Gadzicki et al. 2020).\n"
      ],
      "metadata": {
        "id": "6zOul7QujN5j"
      },
      "id": "6zOul7QujN5j"
    },
    {
      "cell_type": "markdown",
      "source": [
        "<img src=\"https://i.ibb.co/WchcPN1/fig1.png\"><br>\n",
        "*Figure 1.* **Early Fusion**<br><br>\n"
      ],
      "metadata": {
        "id": "CSEJLXi-jRuH"
      },
      "id": "CSEJLXi-jRuH"
    },
    {
      "cell_type": "markdown",
      "source": [
        "###Late or Ensemble Fusion:\n",
        "####In late fusion, the individual streams fully process each data channel separately and their outputs are merged at a decision-making or prediction stage through a summation, averaging operation, or majority voting. Similar to the ensemble technique, the late fusion process learns the optimal combination of each component that maximizes performance. The learning process handles the errors independently, decorrelates errors, and thus improves performance with respect to each modality. The major drawback lies in the limited potential for the exploitation of cross-correlations between the different unimodal data.\n",
        "<img src=\"https://i.ibb.co/T2JHW4X/fig2.png\"><br>\n",
        "*Figure 2.* **Late Fusion** <br><br>\n",
        "###Intermediate/Joint Fusion\n",
        "####Joint fusion was introduced to fuse the imaging feature representations with the clinical features before feeding them into the model. Because of the differences between the imaging and clinical features in dimensionality and dynamic range, different models were introduced (Haylat 2020) to scale their clinical features before fusion and improve their performance (Fig. 3).\n",
        "<img src=\"https://i.ibb.co/kKxVYfz/fig3.png\"><br>\n",
        "*Figure 3.* **Joint Fusion, Type I (Up) and Type II (down)** <br><br>\n",
        "###Slow Fusion\n",
        "####Slow fusion is widely used in 3D video both the spatial and temporal dimensions (Huang et al. 2020) begin as multiple networks that take consecutive video frames and slowly fuse their temporal features throughout the network such that the higher layers progressively get access to global information(Karpathy et al. 2014). As shown in Fig. 4, the whole architecture is segregated into four layers of actions that progressively share the training parameters. In the first layer, each model takes four temporal sequences (stride of four) that pass through a series of convolutions, producing four feature maps. The second layer processes a series of two temporal features out of four feature maps and passes it to layer three, consequently accessing information across all the input video frames. Finally, layer four performs a pooling operation for global feature extraction and eventually connects higher tasks (classification) with a dense connection. Therefore, the machine learning or CNN models benefit from learning the powerful features with subsequent fusion connections and can be robust to details of the connectivity across time.\n",
        "<img src=\"https://i.ibb.co/St6Ccz9/fig4.png\"><br>\n",
        "\n"
      ],
      "metadata": {
        "id": "voChWu2OoCdj"
      },
      "id": "voChWu2OoCdj"
    },
    {
      "cell_type": "markdown",
      "source": [
        "###Imaging Only Multimodal Learning:\n",
        "\n",
        "####As we discussed before, multimodality can be derived from a variety of sources; this section discusses image-image learning, while sections 14-B and 14-C consider imaging-tabular data learning, and imaging-text learning, respectively.\n",
        "\n",
        "####Multimodal imaging studies usually require co-registration. The process of co-registering images includes geometrically aligning two or more images so that their corresponding pixels (voxels) represent the same object. In general, image co-registration can be classified into two categories based on the reference: one approach registers the images into an existing atlas; the other approach registers the images into one selected image. It is essential that this process is undertaken before any subsequent quantitative image analysis. Various techniques are available for doing image registration, including DL-based registration and traditional registration, which are beyond the scope of this chapter (for further reading see (Haskins, Kruger, and Yan 2020)). \n",
        "####Multimodal imaging has many real-world examples, one of which is the fusion of structural and functional (molecular) imaging such as PET-CT and PET-MR. Since molecular imaging has a limited spatial resolution, it cannot precisely show the location of tumors, and tracers often are not taken up by enough structures to make anatomic location clear. By acquiring CT or MR images within the same physical device, the precise anatomic location can be determined(Fig. 5).\n",
        "<img src=\"https://i.ibb.co/vZf4V1b/fig5.png\"><br>\n",
        "*Figure 5.* **(a) Coregistered PET-CT scan , (b) CT scan,(c) PET scan**<br><br>\n",
        "####CT-MRI fusion for radiation therapy is another example. MRI has superb soft-tissue resolution and can show tumor which is difficult to appreciate on CT. On the other hand, MRI does not reflect radiation attenuation, which is critical to accurate dose delivery. As a result,fusion of MRI images with CT is commonly used in radiosurgery, interventional radiology, and radiotherapy. (Liu et al. 2019) (Fig. 6)\n",
        "<img src=\"https://i.ibb.co/qjyf2Z6/fig6.png\"><br>\n",
        "*Figure 6.* **(a) CT scan, (b) Coregistered CT-MRI scan,(c) MRI scan\t**<br><br>\n",
        "####While we used different modalities in the above examples, there are also times when planes or modes of acquisition of the same modality can yield different types of information that can be utilized. For instance, using both the mediolateral oblique (MLO) and the craniocaudal (CC) mammogram views increases the accuracy of diagnostic imaging for detecting breast cancers, compared to only using one. (Fig. 7)\n"
      ],
      "metadata": {
        "id": "kizWyMnd1GGn"
      },
      "id": "kizWyMnd1GGn"
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jS5x_SCTo0jQ"
      },
      "source": [
        "#### In this notebook we are going to train a model to determine MGMT promoter methylation status within patients with GBM tumor.\n",
        "##### We are going to use T1,T2,T1 post contrast,FLAIR MRI sequences and also mask of tumor."
      ],
      "id": "jS5x_SCTo0jQ"
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "c5e00222-12f4-4db8-8293-823461a3b0f6"
      },
      "source": [
        "### Import Packages"
      ],
      "id": "c5e00222-12f4-4db8-8293-823461a3b0f6"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "X1Sz4Q0MeDPm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "32f61a18-2ee8-4186-9fde-1105a37d2afd"
      },
      "source": [
        "!pip install monai wandb -q"
      ],
      "id": "X1Sz4Q0MeDPm",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[K     |████████████████████████████████| 1.8 MB 16.2 MB/s \n",
            "\u001b[K     |████████████████████████████████| 145 kB 75.9 MB/s \n",
            "\u001b[K     |████████████████████████████████| 181 kB 62.1 MB/s \n",
            "\u001b[K     |████████████████████████████████| 63 kB 1.5 MB/s \n",
            "\u001b[?25h  Building wheel for pathtools (setup.py) ... \u001b[?25l\u001b[?25hdone\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c74533f5"
      },
      "source": [
        "import os\n",
        "\n",
        "import math\n",
        "import shutil\n",
        "import random\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "from tqdm.auto import tqdm\n",
        "from sklearn.model_selection import train_test_split\n",
        "\n",
        "import monai as mn\n",
        "import nibabel as nib\n",
        "\n",
        "import torch\n",
        "from torch.utils.data import Dataset,DataLoader,WeightedRandomSampler\n",
        "import wandb"
      ],
      "id": "c74533f5",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0613f8d5-7d5b-48d0-acd6-8ac9e59a0d9a"
      },
      "source": [
        "### Select one GPU if there are many / Set up WANDB environment"
      ],
      "id": "0613f8d5-7d5b-48d0-acd6-8ac9e59a0d9a"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "337fa6ed"
      },
      "source": [
        "os.environ['CUDA_VISIBLE_DEVICES']='1'\n",
        "os.environ['WANDB_API_KEY']=#Enter your WAND API KEY here.Under setting menu of your wandb profile\n",
        "os.environ['WANDB_SILENT']='true'"
      ],
      "id": "337fa6ed",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "b67fc951-d4fa-4a2b-a036-bc8ee974652a"
      },
      "source": [
        "### Make Training Proccess Deterministic"
      ],
      "id": "b67fc951-d4fa-4a2b-a036-bc8ee974652a"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cb668ba2"
      },
      "source": [
        "def seed_all(seed:int) -> None:\n",
        "    random.seed(seed)\n",
        "    os.environ['PYTHONHASHSEED'] = str(seed)\n",
        "    np.random.seed(seed)\n",
        "    torch.manual_seed(seed)\n",
        "    torch.cuda.manual_seed(seed)\n",
        "    torch.backends.cudnn.deterministic = True\n",
        "    torch.backends.cudnn.benchmark = False\n",
        "    mn.utils.misc.set_determinism(seed=seed)\n",
        "seed_all(123)"
      ],
      "id": "cb668ba2",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "71a7990d-8a26-40d8-9441-11a121b400e2"
      },
      "source": [
        "### Set up some Hyperparameters"
      ],
      "id": "71a7990d-8a26-40d8-9441-11a121b400e2"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "93c68dd2"
      },
      "source": [
        "#Hyperparameter\n",
        "bs=2\n",
        "lr=1e-5\n",
        "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
        "epochs=3"
      ],
      "id": "93c68dd2",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bb3dd95e-cb80-491e-8ecd-198504abab34"
      },
      "source": [
        "### Load Labels(MGMT methylation status) for each patient"
      ],
      "id": "bb3dd95e-cb80-491e-8ecd-198504abab34"
    },
    {
      "cell_type": "markdown",
      "source": [
        "### To download the dataset click on the links below and [make a shortcut](https://support.google.com/drive/answer/9700156?hl=en&co=GENIE.Platform%3DDesktop) of them in your google drive.\n",
        "Link to the NIfTI files:https://drive.google.com/drive/folders/1-vASO92VQ-KLSsoBakgySID0fGbdBRQb?usp=sharing \n",
        "\n",
        "Link to the CSV file:https://drive.google.com/file/d/1-HqrqZKPrJfw0GqrcD55_x-C3BrffTPW/view?usp=sharing\n"
      ],
      "metadata": {
        "id": "ntPI5qa5lSeG"
      },
      "id": "ntPI5qa5lSeG"
    },
    {
      "cell_type": "code",
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PABp4bN9jpIB",
        "outputId": "329f7370-2fe8-4ca2-b208-fa811fec4377"
      },
      "id": "PABp4bN9jpIB",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "f855e6f4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 143
        },
        "outputId": "36ec2e5b-4223-4975-f0e3-0418931847da"
      },
      "source": [
        "df=pd.read_csv('/content/drive/MyDrive/train_labels.csv')# csv path\n",
        "df.head(3)"
      ],
      "id": "f855e6f4",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   BraTS21ID  MGMT_value\n",
              "0          0           1\n",
              "1          2           1\n",
              "2          3           0"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-299223d7-5ecc-453d-8da9-18382acb4f98\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>BraTS21ID</th>\n",
              "      <th>MGMT_value</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>2</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>3</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-299223d7-5ecc-453d-8da9-18382acb4f98')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-299223d7-5ecc-453d-8da9-18382acb4f98 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-299223d7-5ecc-453d-8da9-18382acb4f98');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "98b94e70-5617-4930-ae80-24aac4f8f7c3"
      },
      "source": [
        "### Splitting dataset to train and validation sets at patients level"
      ],
      "id": "98b94e70-5617-4930-ae80-24aac4f8f7c3"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1c9f0b69"
      },
      "source": [
        "df_train,df_val=train_test_split(df, test_size=0.19, train_size=0.81, random_state=123 ,shuffle=True, stratify=df['MGMT_value'])"
      ],
      "id": "1c9f0b69",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "364322f2",
        "outputId": "d5a0618b-8a2b-4a15-cff8-84e78634e4a1",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "print('Number of positive train samples:',len(df_train[df_train['MGMT_value']==1]))\n",
        "print('Number of negative train samples:',len(df_train[df_train['MGMT_value']==0]))\n",
        "print('Number of positive validation samples:',len(df_val[df_val['MGMT_value']==1]))\n",
        "print('Number of negative validation samples:',len(df_val[df_val['MGMT_value']==0]))\n",
        "print('Ratio of positive to negative samples in train set:',len(df_train[df_train['MGMT_value']==1])/len(df_train[df_train['MGMT_value']==0]))\n",
        "print('Ratio of positive to negative samples in validation set:',len(df_val[df_val['MGMT_value']==1])/len(df_val[df_val['MGMT_value']==0]))"
      ],
      "id": "364322f2",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of positive train samples: 248\n",
            "Number of negative train samples: 225\n",
            "Number of positive validation samples: 59\n",
            "Number of negative validation samples: 53\n",
            "Ratio of positive to negative samples in train set: 1.1022222222222222\n",
            "Ratio of positive to negative samples in validation set: 1.1132075471698113\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "d760240b",
        "outputId": "cfa8d8fd-69ed-4bc1-e000-6fa9b66a673b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 302
        }
      },
      "source": [
        "df['MGMT_value'].hist(grid=False,bins=3)\n",
        "len_neg=len(df[df['MGMT_value']==0])\n",
        "len_pos=len(df[df['MGMT_value']==1])\n",
        "print(f'Number of Total positive samples:{len_pos}\\nNumber of Total Negative samples:{len_neg}')"
      ],
      "id": "d760240b",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of Total positive samples:307\n",
            "Number of Total Negative samples:278\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bfdb113f-cff3-48e2-80c5-22edefef8daf"
      },
      "source": [
        "### Retrieve image paths"
      ],
      "id": "bfdb113f-cff3-48e2-80c5-22edefef8daf"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "00dabc5e"
      },
      "source": [
        "list_0=os.listdir('/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData')"
      ],
      "id": "00dabc5e",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "99898295"
      },
      "source": [
        "def path_maker_list(patient_list:list,path:str):\n",
        "    path_list=[]\n",
        "    for item in patient_list:\n",
        "        path_patient=os.path.join(path,item)\n",
        "        path_list.append(path_patient)\n",
        "    return path_list"
      ],
      "id": "99898295",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "47be0a64",
        "outputId": "aeeabb28-03fa-4f03-8b5d-2e65b4e93178",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "path_list_0=path_maker_list(list_0,'/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData')\n",
        "len(path_list_0)"
      ],
      "id": "47be0a64",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1250"
            ]
          },
          "metadata": {},
          "execution_count": 21
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "path_list_0[0]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 36
        },
        "id": "aHTFep0hbhM5",
        "outputId": "206a81e4-f4a7-43a1-d67c-c991cd2dea1c"
      },
      "id": "aHTFep0hbhM5",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "'/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_01487'"
            ],
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            }
          },
          "metadata": {},
          "execution_count": 22
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fea28059-9937-4583-ab1c-eb6ae48d7704"
      },
      "source": [
        "##### Create List of Dictionaries that for each patient contain: Paths of different MRI sequences and Label"
      ],
      "id": "fea28059-9937-4583-ab1c-eb6ae48d7704"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "84c06fd3",
        "outputId": "b0f5e0da-b45b-49dd-8df8-e9c346f45b23",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "train_list=[]\n",
        "val_list=[]\n",
        "for i,idx in enumerate(path_list_0):\n",
        "    data_dict={}\n",
        "    id_num=int(idx.split('/')[-1].split('_')[1])\n",
        "    df_selected=df.loc[lambda df: df['BraTS21ID'] == id_num]\n",
        "    idx_str=idx.split('/')[-1].split('_')[1]\n",
        "    try:\n",
        "      label=list(df_selected['MGMT_value'])[0]\n",
        "      data_dict['img_t1']=f'{idx}/BraTS2021_{idx_str}_t1.nii.gz'\n",
        "      data_dict['img_t2']=f'{idx}/BraTS2021_{idx_str}_t2.nii.gz'\n",
        "      data_dict['img_t1c']=f'{idx}/BraTS2021_{idx_str}_t1ce.nii.gz'\n",
        "      data_dict['img_flair']=f'{idx}/BraTS2021_{idx_str}_flair.nii.gz'\n",
        "      data_dict['seg']=f'{idx}/BraTS2021_{idx_str}_seg.nii.gz'\n",
        "      data_dict['label']=label\n",
        "    except:\n",
        "      pass\n",
        "    if id_num in list(df_train['BraTS21ID']):\n",
        "        train_list.append(data_dict)\n",
        "    elif id_num in list(df_val['BraTS21ID']):\n",
        "        val_list.append(data_dict)\n",
        "print(f'Number of train samples:{len(train_list)}')\n",
        "print(f'Number of validation samples:{len(val_list)}')\n",
        "print(f'Each dictionary would be like:\\n{train_list[0]}')"
      ],
      "id": "84c06fd3",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of train samples:467\n",
            "Number of validation samples:109\n",
            "Each dictionary would be like:\n",
            "{'img_t1': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00395/BraTS2021_00395_t1.nii.gz', 'img_t2': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00395/BraTS2021_00395_t2.nii.gz', 'img_t1c': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00395/BraTS2021_00395_t1ce.nii.gz', 'img_flair': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00395/BraTS2021_00395_flair.nii.gz', 'seg': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00395/BraTS2021_00395_seg.nii.gz', 'label': 0}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "af59ae9a-939e-4df3-a26f-a62740b5eafe"
      },
      "source": [
        "### Transformations"
      ],
      "id": "af59ae9a-939e-4df3-a26f-a62740b5eafe"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "572ce3c3"
      },
      "source": [
        "####In the current notebook we just use post contrast T1 and the T2, but you can use other sequnces as well by modifying 'keys' argument in transformations\n",
        "\n",
        "mri_to_use=[\"img_t1c\",'img_t2']\n",
        "train_transforms=mn.transforms.Compose([\n",
        "    mn.transforms.LoadImageD(keys=mri_to_use),\n",
        "    mn.transforms.AddChannelD(keys=mri_to_use),\n",
        "    mn.transforms.SpacingD(keys=mri_to_use,pixdim=(1,1,1), meta_key_postfix='meta_dict'),\n",
        "    mn.transforms.Resized(keys=mri_to_use, spatial_size=(240,240,155)),\n",
        "\n",
        "    ###QUESTION TIME!###  Complete the next function\n",
        "    mn.transforms.NormalizeIntensityD(keys=mri_to_use,channel_wise=##Your code here##),\n",
        "    #channel_wise (bool) – if True, calculate on each channel separately, otherwise, calculate on the entire image directly. default to False.\n",
        "    #''' https://docs.monai.io/en/stable/transforms.html#normalizeintensityd '''\n",
        "\n",
        "    ###QUESTION TIME AGAIN!###  \n",
        "    mn.transforms.ConcatItemsd(keys=mri_to_use, name=\"img\"),\n",
        "    #'''What would be the shape og ''img'' ? '''\n",
        "\n",
        "    mn.transforms.RandAffineD(\n",
        "        keys=\"img\",\n",
        "        translate_range=(15,15,10),\n",
        "        scale_range=(0.05,0.05,0.05),\n",
        "        rotate_range=(math.pi/12,math.pi/12,math.pi/15),\n",
        "        padding_mode='zeros',\n",
        "        prob=0.5,\n",
        "        as_tensor_output=False),\n",
        "    mn.transforms.ToTensord(keys=[\"img\",\"label\"]),\n",
        "    \n",
        "])\n",
        "val_transforms=mn.transforms.Compose([\n",
        "    mn.transforms.LoadImageD(keys=mri_to_use),\n",
        "    mn.transforms.AddChannelD(keys=mri_to_use),\n",
        "    mn.transforms.SpacingD(keys=mri_to_use,pixdim=(1,1,1), meta_key_postfix='meta_dict'),\n",
        "    mn.transforms.Resized(keys=mri_to_use, spatial_size=(240,240,155)),\n",
        "\n",
        "    ###QUESTION TIME! Last time :) ###  \n",
        "    mn.transforms.NormalizeIntensityD(keys=mri_to_use,channel_wise=##Your code here##),\n",
        "    #channel_wise (bool) – if True, calculate on each channel separately, otherwise, calculate on the entire image directly. default to False.\n",
        "    #''' https://docs.monai.io/en/stable/transforms.html#normalizeintensityd '''\n",
        "  \n",
        "    mn.transforms.ConcatItemsd(keys=mri_to_use, name=\"img\"),\n",
        "    mn.transforms.ToTensord(keys=[\"img\",\"label\"]),\n",
        "    \n",
        "])"
      ],
      "id": "572ce3c3",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "040b095e-e71c-4fa6-8f1c-1eb26410bc91"
      },
      "source": [
        "### Create Dataloader/Dataset/Sampler"
      ],
      "id": "040b095e-e71c-4fa6-8f1c-1eb26410bc91"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ff21f92d"
      },
      "source": [
        "'''Creating Dataset'''\n",
        "train_ds = mn.data.Dataset(data=train_list, transform=train_transforms)\n",
        "val_ds = mn.data.Dataset(data=val_list, transform=val_transforms)"
      ],
      "id": "ff21f92d",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1208268f",
        "outputId": "ff963a5c-9d04-405a-d1a6-d31b57f873b1",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "for data in train_ds[:2]:\n",
        "    print(data['img'].shape)"
      ],
      "id": "1208268f",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/functional.py:568: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at  ../aten/src/ATen/native/TensorShape.cpp:2228.)\n",
            "  return _VF.meshgrid(tensors, **kwargs)  # type: ignore[attr-defined]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "torch.Size([2, 240, 240, 155])\n",
            "torch.Size([2, 240, 240, 155])\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "sample_img=train_ds[0]['img']\n",
        "fig,axes=plt.subplots(1,2,figsize=(10,10));\n",
        "axes[0].imshow(sample_img[0,:,:,100],cmap='gray');\n",
        "axes[0].title.set_text('T1C')\n",
        "axes[1].imshow(sample_img[1,:,:,100],cmap='gray');\n",
        "axes[0].title.set_text('T2WI')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 318
        },
        "id": "Bc6KHigv3psV",
        "outputId": "fcdfba4f-17a4-4a1c-c7b5-cffd9b8337a4"
      },
      "id": "Bc6KHigv3psV",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x720 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "'''Sampler'''\n",
        "class_weights=[1.1,1]\n",
        "sample_weights=[0]*len(train_ds)\n",
        "for idx,data in enumerate(train_ds):\n",
        "    class_weight=class_weights[data['label']]\n",
        "    sample_weights[idx]=class_weight\n",
        "sampler= WeightedRandomSampler(sample_weights,num_samples=len(sample_weights),replacement=True)"
      ],
      "metadata": {
        "id": "Ub2KXTh-R6vI"
      },
      "id": "Ub2KXTh-R6vI",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c36a2ae6"
      },
      "source": [
        "\"\"\"How make batch in dataloader\"\"\"\n",
        "def collater(data):\n",
        "    batch_imgs = [s['img'] for s in data]\n",
        "    batch_labels = [s['label'] for s in data]\n",
        "    imgs = torch.stack(batch_imgs, axis=0)\n",
        "    targets = torch.stack(batch_labels, axis=0)\n",
        "    targets=targets.squeeze(-1)\n",
        "    return {'imgs': imgs, 'targets': targets}"
      ],
      "id": "c36a2ae6",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "68522725"
      },
      "source": [
        "\"\"\"Creating train and validation dataloader\"\"\"\n",
        "train_loader = DataLoader(train_ds, batch_size=bs, num_workers=2, pin_memory=torch.cuda.is_available(),,sampler=sampler,collate_fn=collater,prefetch_factor=1)\n",
        "val_loader = DataLoader(val_ds, batch_size=bs, num_workers=2, pin_memory=torch.cuda.is_available(),collate_fn=collater,prefetch_factor=1)"
      ],
      "id": "68522725",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "c4b10926-9dd7-469f-a777-a37febf9077a"
      },
      "source": [
        "### Create model"
      ],
      "id": "c4b10926-9dd7-469f-a777-a37febf9077a"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dee9b8f1"
      },
      "source": [
        "model = mn.networks.nets.DenseNet121(spatial_dims=3, in_channels=2, out_channels=2).to(device)\n",
        "loss_function = torch.nn.CrossEntropyLoss()\n",
        "optimizer = torch.optim.Adam(model.parameters(), lr)"
      ],
      "id": "dee9b8f1",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "943bc88b-1e8f-4bb8-9b4e-3454aab4f8e2"
      },
      "source": [
        "### Training and evaluation on Validation set\n",
        "\n"
      ],
      "id": "943bc88b-1e8f-4bb8-9b4e-3454aab4f8e2"
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "e98ac062"
      },
      "source": [
        "val_interval = 1\n",
        "best_metric = -1\n",
        "best_metric_epoch = -1\n",
        "wandb.init(project='SIIM_early_fusion', entity='Define your entity here')#Define your wandb entity \n",
        "config = wandb.config\n",
        "config.learning_rate = lr\n",
        "config.batch_size=bs\n",
        "config.mode='3D'\n",
        "config.backbone='DenseNet121'\n",
        "config.optimizer='Adam'\n",
        "config.sampler='True'\n",
        "config.resample='True'\n",
        "config.registration='True'\n",
        "config.normalization='Per patient'\n",
        "config.epochs=epochs\n",
        "config.augmentation='Affine'\n",
        "wandb.run.name=f'BraTS_3D_Densenet121_{lr}_bs{bs}_2Channel(1c_seg)_normalization_pp'\n",
        "wandb.watch(model)\n",
        "\n",
        "for i,epoch in enumerate(tqdm(range(epochs))):\n",
        "    print(\"-\" * 10)\n",
        "    print(f\"epoch {epoch + 1}/{epochs}\")\n",
        "    model.train()\n",
        "    epoch_loss = 0\n",
        "    val_epoch_loss=0\n",
        "    step = 0\n",
        "    val_step=0\n",
        "    for batch_data in tqdm(train_loader):\n",
        "        step += 1\n",
        "        inputs, labels = batch_data[\"imgs\"].to(device), batch_data[\"targets\"].to(device)\n",
        "        optimizer.zero_grad()\n",
        "        outputs = model(inputs)\n",
        "        loss = loss_function(outputs, labels)\n",
        "        loss.backward()\n",
        "        optimizer.step()\n",
        "        epoch_loss += loss.item()\n",
        "        epoch_len = (len(train_ds) // train_loader.batch_size)+1\n",
        "        print(f\"{step}/{epoch_len}, train_loss: {loss.item():.4f}\")\n",
        "        wandb.log({\"loss\": loss})\n",
        "    epoch_loss /= step\n",
        "    wandb.log({'epoch_loss':epoch_loss})\n",
        "    print(f\"epoch {epoch + 1} average loss: {epoch_loss:.4f}\")\n",
        "\n",
        "    \n",
        "    model.eval()\n",
        "    with torch.no_grad():\n",
        "        y_pred = torch.tensor([], dtype=torch.float32, device=device)\n",
        "        y = torch.tensor([], dtype=torch.long, device=device)\n",
        "        for val_data in val_loader:\n",
        "            val_step +=1\n",
        "            val_images, val_labels = val_data[\"imgs\"].to(device), val_data[\"targets\"].to(device)\n",
        "            val_output=model(val_images)\n",
        "            val_loss= loss_function(val_output,val_labels)\n",
        "            val_epoch_loss += val_loss.item()\n",
        "            val_epoch_len = len(val_ds) // val_loader.batch_size\n",
        "            total_val_step=(len(val_ds)//bs)+1\n",
        "            print(f\"{val_step}/{total_val_step}, val_loss: {val_loss.item():.4f}\")\n",
        "            wandb.log({'val_loss':val_loss})\n",
        "            y_pred = torch.cat([y_pred, model(val_images)], dim=0)\n",
        "            y = torch.cat([y, val_labels], dim=0)\n",
        "            \n",
        "\n",
        "        val_epoch_loss /= val_step\n",
        "        print(f\"epoch {epoch + 1} average val_loss: {val_epoch_loss:.4f}\")\n",
        "        wandb.log({'val_epoch_loss':val_epoch_loss})\n",
        "\n",
        "        acc_value_val = torch.eq(y_pred.argmax(dim=1), y)\n",
        "        acc_metric_val = acc_value_val.sum().item() / len(acc_value_val)\n",
        "        if acc_metric_val > best_metric:\n",
        "            best_metric = acc_metric_val\n",
        "            best_metric_epoch = epoch + 1\n",
        "            ##torch.save(model.state_dict(), \"best_metric_model_classification3d_multimodal_early.pth\") Specify Directory\n",
        "            print(\"saved new best metric model\")\n",
        "        print(f\"current epoch: {epoch + 1} current accuracy: {acc_metric_val:.4f} best accuracy: { best_metric:.4f}\")\n",
        "        wandb.log({\"loss\": loss,\n",
        "                    'Accuracy_val':acc_metric_val,})\n",
        "        y_pred = torch.tensor([], dtype=torch.float32, device=device)\n",
        "        y = torch.tensor([], dtype=torch.long, device=device)\n",
        "        \n",
        "        step = 0\n",
        "        train_step=0\n",
        "        for train_data in train_loader:\n",
        "            train_step +=1\n",
        "            train_images, train_labels = train_data[\"imgs\"].to(device), train_data[\"targets\"].to(device)\n",
        "            train_output=model(train_images)\n",
        "            train_epoch_len = len(train_ds) // train_loader.batch_size\n",
        "            total_train_step=(len(train_ds)//bs)+1\n",
        "            y_pred = torch.cat([y_pred, model(train_images)], dim=0)\n",
        "            y = torch.cat([y, train_labels], dim=0)\n",
        "            \n",
        "\n",
        "        acc_value_train = torch.eq(y_pred.argmax(dim=1), y)\n",
        "        acc_metric_train = acc_value_train.sum().item() / len(acc_value_train)\n",
        "        wandb.log({'Accuracy_train':acc_metric_train})\n",
        "\n",
        "print(f\"train completed, best_metric: {best_metric:.4f} at epoch: {best_metric_epoch}\")"
      ],
      "id": "e98ac062",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "##Intermediate Fusion: Adding Age!\n",
        "###In this section we will add the age of the patient after extracting features from MRI, and then we will train another neural net (while simultaneously training the first neural net).\n",
        "![image.png](data:image/png;base64,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)"
      ],
      "metadata": {
        "id": "Q9O0BYiYF2lB"
      },
      "id": "Q9O0BYiYF2lB"
    },
    {
      "cell_type": "markdown",
      "source": [
        "Create List of dictionaries...."
      ],
      "metadata": {
        "id": "vabNjDkmJoov"
      },
      "id": "vabNjDkmJoov"
    },
    {
      "cell_type": "code",
      "source": [
        "train_list=[]\n",
        "val_list=[]\n",
        "for i,idx in enumerate(path_list_0):\n",
        "    data_dict={}\n",
        "    id_num=int(idx.split('/')[-1].split('_')[1])\n",
        "    df_selected=df.loc[lambda df: df['BraTS21ID'] == id_num]\n",
        "    idx_str=idx.split('/')[-1].split('_')[1]\n",
        "    try:\n",
        "      label=list(df_selected['MGMT_value'])[0]\n",
        "      data_dict['img_t1']=f'{idx}/BraTS2021_{idx_str}_t1.nii.gz'\n",
        "      data_dict['img_t2']=f'{idx}/BraTS2021_{idx_str}_t2.nii.gz'\n",
        "      data_dict['img_t1c']=f'{idx}/BraTS2021_{idx_str}_t1ce.nii.gz'\n",
        "      data_dict['img_flair']=f'{idx}/BraTS2021_{idx_str}_flair.nii.gz'\n",
        "      data_dict['seg']=f'{idx}/BraTS2021_{idx_str}_seg.nii.gz'\n",
        "      data_dict['age']=torch.randint(low=40, high=90,size=(1,))# Since BraTS2021 dataset doesn't contain age of the patients, here we are generating random age for the patients!\n",
        "      data_dict['label']=label\n",
        "    except:\n",
        "      pass\n",
        "    if id_num in list(df_train['BraTS21ID']):\n",
        "        train_list.append(data_dict)\n",
        "    elif id_num in list(df_val['BraTS21ID']):\n",
        "        val_list.append(data_dict)\n",
        "print(f'Number of train samples:{len(train_list)}')\n",
        "print(f'Number of validation samples:{len(val_list)}')\n",
        "print(f'Each dictionary would be like:\\n{train_list[0]}')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5G1ba1PU9LA8",
        "outputId": "a798e131-076b-499d-a215-1dd8809964d4"
      },
      "id": "5G1ba1PU9LA8",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of train samples:24\n",
            "Number of validation samples:6\n",
            "Each dictionary would be like:\n",
            "{'img_t1': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00160/BraTS2021_00160_t1.nii.gz', 'img_t2': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00160/BraTS2021_00160_t2.nii.gz', 'img_t1c': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00160/BraTS2021_00160_t1ce.nii.gz', 'img_flair': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00160/BraTS2021_00160_flair.nii.gz', 'seg': '/content/drive/MyDrive/RSNA_ASNR_MICCAI_BraTS2021_TrainingData/BraTS2021_00160/BraTS2021_00160_seg.nii.gz', 'age': tensor([51]), 'label': 1}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "'''Creating Dataset'''\n",
        "train_ds = mn.data.Dataset(data=train_list, transform=train_transforms)\n",
        "val_ds = mn.data.Dataset(data=val_list, transform=val_transforms)"
      ],
      "metadata": {
        "id": "v_5VM2n7BwPv"
      },
      "id": "v_5VM2n7BwPv",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "'''Sampler'''\n",
        "class_weights=[1.1,1]\n",
        "sample_weights=[0]*len(train_ds)\n",
        "for idx,data in enumerate(train_ds):\n",
        "    class_weight=class_weights[data['label']]\n",
        "    sample_weights[idx]=class_weight\n",
        "sampler= WeightedRandomSampler(sample_weights,num_samples=len(sample_weights),replacement=True)"
      ],
      "metadata": {
        "id": "1xwHCoApR1Se"
      },
      "id": "1xwHCoApR1Se",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "\"\"\"How make batch in dataloader\"\"\"\n",
        "def collater(data):\n",
        "    batch_imgs = [s['img'] for s in data]\n",
        "    batch_labels = [s['label'] for s in data]\n",
        "    batch_ages=[s['age'] for s in data]\n",
        "    imgs = torch.stack(batch_imgs, axis=0)\n",
        "    targets = torch.stack(batch_labels, axis=0)\n",
        "    ages=torch.stack(batch_ages, axis=0)\n",
        "    targets=targets.squeeze(-1)\n",
        "    ages=ages.squeeze(-1)\n",
        "    return {'imgs': imgs, 'targets': targets, 'ages':ages}"
      ],
      "metadata": {
        "id": "paf9Faa-B7UP"
      },
      "id": "paf9Faa-B7UP",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "\"\"\"Creating train and validation dataloader\"\"\"\n",
        "train_loader = DataLoader(train_ds, batch_size=bs, num_workers=2, pin_memory=torch.cuda.is_available(),sampler=sampler,collate_fn=collater,prefetch_factor=1)\n",
        "val_loader = DataLoader(val_ds, batch_size=bs, num_workers=2, pin_memory=torch.cuda.is_available(),collate_fn=collater,prefetch_factor=1)"
      ],
      "metadata": {
        "id": "SdzVVzFMCATz"
      },
      "id": "SdzVVzFMCATz",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Consider re adding the figure explaininig the intermediate fusion"
      ],
      "metadata": {
        "id": "eLps2S6KNivN"
      },
      "id": "eLps2S6KNivN"
    },
    {
      "cell_type": "code",
      "source": [
        "extracted_features=10 #Feel free to change the number of extracted feature\n",
        "model = mn.networks.nets.DenseNet121(spatial_dims=3, in_channels=2, out_channels=extracted_features).to(device)\n",
        "intrmd_model=torch.nn.Sequential(\n",
        "     torch.nn.Linear(extracted_features+1, 5),\n",
        "     torch.nn.ReLU(),\n",
        "     torch.nn.Linear(5,2))#Feel free to change the number of layes, neurons , and the activation function\n",
        "intrmd_model.to(device)\n",
        "loss_function = torch.nn.CrossEntropyLoss()\n",
        "params = list(model.parameters()) + list(intrmd_model.parameters())# Creating a list containing the parameters of the first and second neural networks\n",
        "optimizer = torch.optim.Adam(params, lr)# Simultaneously optimizing both networks."
      ],
      "metadata": {
        "id": "9qXhl_xMEKkw"
      },
      "id": "9qXhl_xMEKkw",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#@title If you want to check the shape of the extracted brain MRI featurees and concatenated vector run this cell and reveal the code.\n",
        "for batch_data in tqdm(train_loader):\n",
        "        inputs, labels,ages= batch_data[\"imgs\"].to(device), batch_data[\"targets\"].to(device), batch_data['ages'].to(device)\n",
        "        ages=ages.reshape(2,1)\n",
        "        optimizer.zero_grad()\n",
        "        outputs = model(inputs)\n",
        "        print(f'Shape of the extracted features in a batch:{outputs.shape}')\n",
        "        print('label:',labels)\n",
        "        outputs = torch.cat((outputs,ages),dim=1)\n",
        "        print(f'Shape of the concatenated vector in a batch:{outputs.shape}')\n",
        "        outputs = intrmd_model(outputs)\n",
        "        print(f'Shape of the final output in a batch:{outputs.shape}')\n",
        "        break"
      ],
      "metadata": {
        "id": "F6M4-xWmHEcx",
        "cellView": "form"
      },
      "id": "F6M4-xWmHEcx",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "##Intermediate fusion training loop"
      ],
      "metadata": {
        "id": "lsTzM5MOxmPR"
      },
      "id": "lsTzM5MOxmPR"
    },
    {
      "cell_type": "code",
      "source": [
        "val_interval = 1\n",
        "best_metric = -1\n",
        "best_metric_epoch = -1\n",
        "wandb.init(project='SIIM_intermediate_fusion', entity='Define your entity here')#Define your wandb entity \n",
        "config = wandb.config\n",
        "config.learning_rate = lr\n",
        "config.batch_size=bs\n",
        "config.mode='3D'\n",
        "config.backbone='DenseNet121'\n",
        "config.optimizer='Adam'\n",
        "config.sampler='True'\n",
        "config.resample='True'\n",
        "config.registration='True'\n",
        "config.normalization='Per patient'\n",
        "config.epochs=epochs\n",
        "config.augmentation='Affine'\n",
        "wandb.run.name=f'BraTS_3D_Densenet121_{lr}_bs{bs}_2Channel(1c_seg)_normalization_pp'\n",
        "wandb.watch(model)\n",
        "\n",
        "for i,epoch in enumerate(tqdm(range(epochs))):\n",
        "    print(\"-\" * 10)\n",
        "    print(f\"epoch {epoch + 1}/{epochs}\")\n",
        "    model.train()\n",
        "    epoch_loss = 0\n",
        "    val_epoch_loss=0\n",
        "    step = 0\n",
        "    val_step=0\n",
        "    for batch_data in tqdm(train_loader):\n",
        "        step += 1\n",
        "        inputs, labels,ages= batch_data[\"imgs\"].to(device), batch_data[\"targets\"].to(device), batch_data['ages'].to(device)\n",
        "        ages=ages.reshape(bs,1)# Makes the dimensions compatible for concatenation.\n",
        "        optimizer.zero_grad()\n",
        "        outputs = model(inputs)\n",
        "        outputs = torch.cat((outputs,ages),dim=1)# Concatenation of the extracted features with age.\n",
        "        outputs = intrmd_model(outputs)# Feeding the concatenated vector to the second neural network.\n",
        "        loss = loss_function(outputs, labels)\n",
        "        loss.backward()\n",
        "        optimizer.step()\n",
        "        epoch_loss += loss.item()\n",
        "        epoch_len = (len(train_ds) // train_loader.batch_size)+1\n",
        "        print(f\"{step}/{epoch_len}, train_loss: {loss.item():.4f}\")\n",
        "        wandb.log({\"loss\": loss})\n",
        "    epoch_loss /= step\n",
        "    wandb.log({'epoch_loss':epoch_loss})\n",
        "    print(f\"epoch {epoch + 1} average loss: {epoch_loss:.4f}\")\n",
        "\n",
        "    \n",
        "    model.eval()\n",
        "    with torch.no_grad():\n",
        "        y_pred = torch.tensor([], dtype=torch.float32, device=device)\n",
        "        y = torch.tensor([], dtype=torch.long, device=device)\n",
        "        for val_data in val_loader:\n",
        "            val_step +=1\n",
        "            val_images, val_labels,val_ages = val_data[\"imgs\"].to(device), val_data[\"targets\"].to(device),val_data['ages'].to(device)\n",
        "            val_ages=val_ages.reshape(bs,1)# Makes the dimensions compatible for concatenation.\n",
        "            val_output=model(val_images)\n",
        "            val_output=torch.cat((val_output,val_ages),dim=1)# Concatenation of the extracted features with age.\n",
        "            val_output=intrmd_model(val_output)# Feeding the concatenated vector to the second neural network.\n",
        "            val_loss= loss_function(val_output,val_labels)\n",
        "            val_epoch_loss += val_loss.item()\n",
        "            val_epoch_len = len(val_ds) // val_loader.batch_size\n",
        "            total_val_step=(len(val_ds)//bs)+1\n",
        "            print(f\"{val_step}/{total_val_step}, val_loss: {val_loss.item():.4f}\")\n",
        "            wandb.log({'val_loss':val_loss})\n",
        "            y_pred = torch.cat([y_pred, model(val_images)], dim=0)\n",
        "            y = torch.cat([y, val_labels], dim=0)\n",
        "            \n",
        "\n",
        "        val_epoch_loss /= val_step\n",
        "        print(f\"epoch {epoch + 1} average val_loss: {val_epoch_loss:.4f}\")\n",
        "        wandb.log({'val_epoch_loss':val_epoch_loss})\n",
        "\n",
        "        acc_value_val = torch.eq(y_pred.argmax(dim=1), y)\n",
        "        acc_metric_val = acc_value_val.sum().item() / len(acc_value_val)\n",
        "        if acc_metric_val > best_metric:\n",
        "            best_metric = acc_metric_val\n",
        "            best_metric_epoch = epoch + 1\n",
        "            ##torch.save(model.state_dict(), \"best_metric_model_classification3d_multimodal_.pth\")# Specify Directory\n",
        "            print(\"saved new best metric model\")\n",
        "        print(f\"current epoch: {epoch + 1} current accuracy: {acc_metric_val:.4f} best accuracy: { best_metric:.4f}\")\n",
        "        wandb.log({\"loss\": loss,\n",
        "                    'Accuracy_val':acc_metric_val,})\n",
        "        y_pred = torch.tensor([], dtype=torch.float32, device=device)\n",
        "        y = torch.tensor([], dtype=torch.long, device=device)\n",
        "        \n",
        "        step = 0\n",
        "        train_step=0\n",
        "        for train_data in train_loader:\n",
        "            train_step +=1\n",
        "            train_images, train_labels, train_ages = train_data[\"imgs\"].to(device), train_data[\"targets\"].to(device), train_data['ages'].to(device)\n",
        "            train_ages=train_ages.reshape(bs,1)# Makes the dimensions compatible for concatenation.\n",
        "            train_output=model(train_images)\n",
        "            train_output=torch.cat((train_output,train_ages),dim=1)# Concatenation of the extracted features with age.\n",
        "            train_output=intrmd_model(train_output)# Feeding the concatenated vector to the second neural network.\n",
        "            train_epoch_len = len(train_ds) // train_loader.batch_size\n",
        "            total_train_step=(len(train_ds)//bs)+1\n",
        "            y_pred = torch.cat([y_pred, model(train_images)], dim=0)\n",
        "            y = torch.cat([y, train_labels], dim=0)\n",
        "            \n",
        "\n",
        "        acc_value_train = torch.eq(y_pred.argmax(dim=1), y)\n",
        "        acc_metric_train = acc_value_train.sum().item() / len(acc_value_train)\n",
        "        wandb.log({'Accuracy_train':acc_metric_train})\n",
        "\n",
        "print(f\"train completed, best_metric: {best_metric:.4f} at epoch: {best_metric_epoch}\")"
      ],
      "metadata": {
        "id": "HcsQ568xFa-0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 135,
          "referenced_widgets": [
            "fe261f7e333c49789718b8da1f40a928",
            "83266fabcf184519b6c6171dbb04e4a2",
            "d04f41a2f2564080b93fd0f23d4d8842",
            "925254b1e9ef4d96876d923d20e1ec87",
            "4e2481b28d9142da9dfd237255fa35a4",
            "ed12f53ef2344a45a99df0a6da4a6e84",
            "034b312ee7ec4d309fd8a8b1f289a942",
            "73b2e9137f61433fa22a859b39462fbc",
            "b8d4249c0d544b30bfe7cb655d703d19",
            "e5e04d54574d4fb8a595e0e4fafbb482",
            "266c41902ec24b5385172180bd11c3b5",
            "4bdbb02bbb2d43f08c9ca0f57a453031",
            "f6c41bb9bcf843eba85dfbbdf725f232",
            "79e204ff103d450991ac815e6b4c11ba",
            "24dce259922b49ba8eaa137728c69b78",
            "20d0b9894414426493025ef8307e0bb5",
            "8b3bbd669a3042d08685b3eeb8bbbe98",
            "9d7dbb7b13034e97b7229a0e31761dee",
            "3c5823b55b834d3398e04d84906656d2",
            "d8495f189f454b449e5bfb3125eccb8b",
            "d93e5b276bd341769d15a687d79ee4d0",
            "020470f3eec84edea07c31450fa7a964"
          ]
        },
        "outputId": "b55c09e6-1494-4358-cff5-17b136c5d183"
      },
      "id": "HcsQ568xFa-0",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "  0%|          | 0/3 [00:00<?, ?it/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "fe261f7e333c49789718b8da1f40a928"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------\n",
            "epoch 1/3\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "  0%|          | 0/12 [00:00<?, ?it/s]"
            ],
            "application/vnd.jupyter.widget-view+json": {
              "version_major": 2,
              "version_minor": 0,
              "model_id": "4bdbb02bbb2d43f08c9ca0f57a453031"
            }
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "1/13, train_loss: 8.6068\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "##Late Fusion: Model ensemble!\n",
        "### In this section we want to do result ensemble of three imaginary different models.\n",
        "In this imaginary scenario we trained the first model using T1W, the second model using T2W, and the third model using T1C. "
      ],
      "metadata": {
        "id": "0WRHFoNT4BYG"
      },
      "id": "0WRHFoNT4BYG"
    },
    {
      "cell_type": "code",
      "source": [
        "Pred_1=torch.tensor([1,1,1,1,1,0])# A model trained with T1WI \n",
        "print(f\"T1WI's model accuracy:{torch.eq(Pred_1,Labels).sum()/len(Labels)}\")\n",
        "Pred_2=torch.tensor([0,0,0,0,1,0])# A model trained with T2WI \n",
        "print(f\"T2WI's model accuracy:{torch.eq(Pred_2,Labels).sum()/len(Labels)}\")\n",
        "Pred_3=torch.tensor([1,0,1,0,0,1])# A model trained with T1C \n",
        "print(f\"T1C's model accuracy:{torch.eq(Pred_3,Labels).sum()/len(Labels)}\")\n",
        "Labels=torch.tensor([1,0,1,0,1,0])# Ground truth"
      ],
      "metadata": {
        "id": "UNdf6V3F3-8p"
      },
      "id": "UNdf6V3F3-8p",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "'''Late Fusion'''\n",
        "Pred_aggregated=Pred_1+Pred_2+Pred_3\n",
        "Pred_aggregated=torch.where(Pred_aggregated>1.5,1,0)# Performing majority voting\n",
        "print(f'Model ensemble accuracy:{torch.eq(Pred_aggregated,Labels).sum()/len(Labels)}')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "3nMHz9Fz5p-t",
        "outputId": "e04d2e24-0344-4baf-eea1-90ab54477453"
      },
      "id": "3nMHz9Fz5p-t",
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model ensemble accuracy:1.0\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "##Some disscused questions:\n",
        "###1- What is the best approach for normalizing the age?\n",
        "###2- Pros and Cons of early, intermediate and late fusion?\n",
        "###3- In what circumstances should early, intermediate, or late fusion be used?\n",
        "###4- Why don't we always do model ensembles and late fusions?\n",
        "##Please feel free to contact us with any questions you may have or if you wish to discuss more:\n",
        "Faghani.shahriar@mayo.edu\n",
        "\n",
        "Conte.gianmarco@mayo.edu\n",
        "\n",
        "Moassefi.mana@mayo.edu"
      ],
      "metadata": {
        "id": "woj-YXooT0xm"
      },
      "id": "woj-YXooT0xm"
    },
    {
      "cell_type": "markdown",
      "source": [
        "###References:\n",
        "1. Bellot, D., Boyer, A. and Charpillet, F. (2002) ‘A new definition of qualified gain in a data fusion process: application to telemedicine’, in Proceedings of the Fifth International Conference on Information Fusion. FUSION 2002. (IEEE Cat.No.02EX5997), pp. 865–872 vol.2.\n",
        "2. Gadzicki, K., Khamsehashari, R. and Zetzsche, C. (2020) ‘Early vs Late Fusion in Multimodal Convolutional Neural Networks’, in 2020 IEEE 23rd International Conference on Information Fusion (FUSION), pp. 1–6.\n",
        "3. Karpathy, A. et al. (2014) ‘Large-scale video classification with convolutional neural networks’, in Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, pp. 1725–1732.\n",
        "4. Huang, S.-C. et al. (2020) ‘Fusion of medical imaging and electronic health records using deep learning: a systematic review and implementation guidelines’, npj Digital Medicine. doi:10.1038/s41746-020-00341-z.\n",
        "5. Haileleol Tibebu (2020), Introduction To Data Fusion, URL: https://medium.com/haileleol-tibebu/data-fusion-78e68e65b2d1\n",
        "6. Yoo, Y. et al. (2019) ‘Deep learning of brain lesion patterns and user-defined clinical and MRI features for predicting conversion to multiple sclerosis from clinically isolated syndrome’, Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 7(3), pp. 250–259.iomed. Eng. Imaging Vis. 7, 250–259, (2019).\n",
        "7. Boonn, William W., and Curtis P. Langlotz. 2009. “Radiologist Use of and Perceived Need for Patient Data Access.” Journal of Digital Imaging 22 (4): 357–62.\n",
        "8. Haskins, Grant, Uwe Kruger, and Pingkun Yan. 2020. “Deep Learning in Medical Image Registration: A Survey.” Machine Vision and Applications 31 (1): 8.\n",
        "9. Huang, Shih-Cheng, Anuj Pareek, Saeed Seyyedi, Imon Banerjee, and Matthew P. Lungren. 2020. “Fusion of Medical Imaging and Electronic Health Records Using Deep Learning: A Systematic Review and Implementation Guidelines.” NPJ Digital Medicine 3 (October): 136.\n",
        "10. Liu, Shi-Feng, Jian Lu, Hong Wang, Yan Han, De-Feng Wang, Li-Li Yang, Zi-Xiang Li, and Xiao-Kun Hu. 2019. “Computed Tomography-Magnetic Resonance Imaging Fusion-Guided Iodine-125 Seed Implantation for Single Malignant Brain Tumor: Feasibility and Safety.” Journal of Cancer Research and Therapeutics 15 (4): 818–24.\n"
      ],
      "metadata": {
        "id": "oELTViNW2EIg"
      },
      "id": "oELTViNW2EIg"
    }
  ]
}