Skip to content

jonathan-githubofficial/Computational-Pathology-with-CNNs

Repository files navigation

Computational Pathology with CNNs Open In Colab

This repository contains the research and development of machine learning models, particularly Convolutional Neural Networks (CNNs), applied to computational pathology and computer vision tasks.

Project Overview

We aim to leverage CNNs to automate tissue classification in computational pathology and explore applications in computer vision, such as animal face classification. Our primary goal is to achieve high accuracy and address inherent challenges in the datasets.

Datasets

  1. Colorectal Cancer Classification: A reduced dataset containing 6K image patches split into three distinct classes.
  2. Prostate Cancer Classification: Comprises 6k image patches from a larger dataset of 120k images, classifying tissue types.
  3. Animal Faces Classification: Focuses on 6k images out of 16k, classifying animal types such as Cats, Dogs, and wildlife animals.

Getting Started

Prerequisites

  • Python 3.x
  • PyTorch
  • Other dependencies listed in requirements.txt

Installation

  1. Clone the repo:
git clone https://github.com/jonathan-githubofficial/COMP-432
  1. Install the required packages:
pip install -r requirements.txt

Usage

  1. run the _init.py the check the cuda condition to see if you are capable to use GPU acceleration
  2. run the model with tweaked hyperparameters, once the model is trained, it will generate a file. like *modelname*_Seed=*seed*.pth in the Model/Saved folder, this is the checkpoint of the model.
  3. run the ReportDriver.py with desired seed, which you can find it in Model/Saved/hyperparameter_log, then you can observe the output and the t-SNE report

Challenges & Solutions

  • Data Variability: Addressed through data preprocessing and augmentation techniques.
  • Imbalanced Datasets: Utilized techniques like SMOTE and undersampling to balance the classes.
  • Transfer Learning: Conducted experiments to fine-tune models across different datasets.
  • Computational Demands: Recommended usage of GPU acceleration for training deep CNNs.

Certainly! Here's the updated "Team Members" table with the provided information:

Team Members

Name ID GitHub
Jonathan Haddad 40111053 jonathangithubofficial
Jainil Jaha 40067468 jjaha99
Yixin Liu 40115632 Jiejue233
Kevin Hong 40176625 krocden
Yichen Huang 40167688 prprtracy

License

MIT

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages