SMIT CT Lung GTV segmentation model

models/msk_smit_lung_gtv/config/default.yml

@@ -0,0 +1,34 @@
+general:
+  data_base_dir: /app/data
+  version: 1.0.0
+  description: Default configuration for SMIT model (dicom to dicom)
+
+execute:
+- DicomImporter
+- NiftiConverter
+- SMITRunner
+- DsegConverter
+- DataOrganizer
+
+modules:
+  DicomImporter:
+    source_dir: input_data
+    import_dir: sorted_data
+    sort_data: true
+    meta: 
+      mod: '%Modality'
+
+  SMITRunner:
+    a_min: -500
+    a_max: 500
+    # Can add other config paremeters here
+
+  DsegConverter:
+    model_name: SMIT
+    body_part_examined: CHEST
+    source_segs: nifti:mod=seg
+    skip_empty_slices: true
+
+  DataOrganizer:
+   targets:
+   - dicomseg:mod=seg-->[i:sid]/smit.seg.dcm%    

models/msk_smit_lung_gtv/dockerfiles/Dockerfile

@@ -0,0 +1,24 @@
+FROM mhubai/base:latest
+
+# Update authors label
+LABEL authors="[email protected],[email protected],[email protected],[email protected]"
+
+RUN apt update
+
+ARG MHUB_MODELS_REPO
+#ENV MHUB_MODELS_REPO=https://github.com/locastre/models.git
+RUN buildutils/import_mhub_model.sh msk_smit_lung_gtv ${MHUB_MODELS_REPO}
+
+#ENV WORK_DIR=/app/models/msk_smit_lung_gtv/src
+ENV WORK_DIR=/app
+
+WORKDIR ${WORK_DIR}/msk_smit_lung_gtv/src
+ENV WEIGHTS_URL=https://mskcc.box.com/shared/static/sf7jic4m2dk67413cipbbq6hddvhpj61.gz
+ENV CONDA_URL=https://mskcc.box.com/shared/static/d580gfjzzmt26v8klwp8pivb6wafomag.gz
+RUN wget ${WEIGHTS_URL} -O weights.tar.gz && tar xvf weights.tar.gz && rm weights.tar.gz
+RUN mkdir conda-pack && chmod -R 777 conda-pack
+RUN cd conda-pack && wget ${CONDA_URL} -O conda.tar.gz && tar xvf conda.tar.gz && rm conda.tar.gz
+
+
+ENTRYPOINT ["mhub.run"]
+CMD ["--config", "/app/models/msk_smit_lung_gtv/config/default.yml"]

models/msk_smit_lung_gtv/meta.json

@@ -0,0 +1,102 @@
+{
+    "id": "", 
+    "name": "msk_smit_lung_gtv", 
+    "title": "CT Lung GTV SMIT Segmentation",
+    "summary": {
+    "description": "GTV segmentation from CT scan",
+    "inputs": [
+    {
+        "label": "Input Image",                     
+        "description": "The CT scan of a patient.", 
+        "format": "NIFTI", 
+        "modality": "CT",  
+        "bodypartexamined": "Chest",
+        "slicethickness": "5mm",    
+        "contrast": true,     
+        "noncontrast": true   
+    }
+    ],
+    "outputs": [
+        {
+            "label": "Segmentation of the lung GTV",  
+            "description": "Segmentation of the lung GTV from NIfTI CT images.",
+            "type": "Segmentation",                       
+            "classes": [  
+                "GTV"
+            ]
+        }
+    ],
+    "model": {
+        "architecture": "Swin Transformer based segmentation, self-supervised pretrained with 10k CT data", 
+        "training": "supervised",   
+        "cmpapproach": "3D" 
+    },
+    "data": {
+        "training": {
+            "vol_samples": 377  
+        },      
+        "evaluation": {
+            "vol_samples": 139
+        },
+        "public": true,   
+        "external": false 
+    }
+    },
+    "details": {
+        "name": "SMIT",   
+        "version": "1.0.0", 
+        "devteam": "",  
+        "authors": ["Jue Jiang, Harini Veeraraghavan"],      
+        "type": "it is a 3D Swin transformer based segmentation net",   
+        "date": {
+            "code": "11.03.2025",   
+            "weights": "11.03.2025",
+            "pub": "15.July.2024"    
+        },
+        "cite": "Jiang, Jue, and Harini Veeraraghavan. Self-supervised pretraining in the wild imparts image acquisition robustness to medical image transformers: an application to lung cancer segmentation. Proceedings of machine learning research 250 (2024): 708.",
+		"license": {
+            "code": "GNU General Public License",         
+            "weights": "GNU General Public License"
+        },
+        "publications": [
+			{
+			"title": "Self-supervised pretraining in the wild imparts image acquisition robustness to medical image transformers: an application to lung cancer segmentation",  
+            "url": "https://openreview.net/pdf?id=G9Te2IevNm"
+			},   
+			{
+			"title":"Self-supervised 3D anatomy segmentation using self-distilled masked image transformer (SMIT)",
+			"url":"https://link.springer.com/chapter/10.1007/978-3-031-16440-8_53"
+        	}
+	],
+        "github": "https://github.com/The-Veeraraghavan-Lab/CTRobust_Transformers.git"
+    },     
+    "info": {   
+    	"use": {
+			"title": "Intended use",
+			"text": "This model is intended to be used on CT images (with or without contrast)",
+			"references": [],
+			"tables": []
+
+		},
+    	"evaluation": {
+			"title": "Evaluation data",
+			"text": "To assess the model's segmentation performance in the NSCLC Radiogenomics dataset, we considered that the original input data is a full 3D volume. The model segmented not only the labeled tumor but also tumors that were not manually annotated. Therefore, we evaluated the model based on the manually labeled tumors. After applying the segmentation model, we extracted a 128*128*128 cubic region containing the manual segmentation to assess the model’s performance.",
+			"references": [],
+			"tables": ["validation_data_id and DSC value, Validation data is 139 data in the NSCLC Radiogenomics data:https://www.cancerimagingarchive.net/collection/nsclc-radiogenomics/, AMC-001:0.023977216,AMC-005:0.84385232,AMC-006:0.844950109,AMC-011:0.885911774,AMC-013:0.786724403,AMC-014:0.628335342,AMC-016:0.708633094,AMC-019:0.791600435,AMC-020:0.882119609,AMC-021:0.834135707,AMC-022:0.688767807,AMC-026:0.801595536,R01-001:0.738330143,R01-002:0.826459454,R01-003:0.724166437,R01-004:0.643794147,R01-005:0.8740986,R01-006:0.816578249,R01-007:0.736460458,R01-008:0.570397112,R01-010:0.901700554,R01-011:0.836905321,R01-012:0.26011073,R01-013:0.760693274,R01-014:0.605606001,R01-015:0.921568729,R01-016:0.748842593,R01-018:0.899090049,R01-019:0.777296896,R01-020:0.858735841,R01-021:0.674536904,R01-022:0.773468955,R01-023:0.851143174,R01-024:0.63791364,R01-025:0.667036976,R01-026:0.867828559,R01-027:0.849266954,R01-028:0.914362163,R01-029:0.796479193,R01-030:0.742501087,R01-031:0.771934798,R01-032:0.546395241,R01-033:0.668465959,R01-034:0.491623711,R01-035:0.861957664,R01-036:0.834929738,R01-039:0.640360767,R01-040:0.843040538,R01-041:0.255910987,R01-042:0.827863856,R01-043:0.358487119,R01-045:0.556983182,R01-046:0.798674399,R01-047:0.875100294,R01-048:0.86953796,R01-049:0.831395349,R01-050:0.736791014,R01-051:0.863763708,R01-052:0.853056081,R01-054:0.890185037,R01-055:0.721171698,R01-056:0.646278311,R01-057:0.819531018,R01-060:0.755168662,R01-061:0.831325301,R01-062:0.621616202,R01-063:0.887817849,R01-064:0.503693754,R01-065:0.900957261,R01-066:0.863084304,R01-067:0.793478908,R01-068:0.706467662,R01-069:0.652887756,R01-070:0.156561781,R01-071:0.794301598,R01-072:0.71873941,R01-073:0.656626506,R01-074:0.686797136,R01-075:0.769153952,R01-076:0.658746901,R01-077:0.515673556,R01-078:0.805609871,R01-079:0.768960982,R01-080:0.465984655,R01-082:0.764202063,R01-083:0.420652174,R01-084:0.679731288,R01-085:0.768992248,R01-086:0.493431042,R01-087:0.488001239,R01-088:0.593974567,R01-089:0.933253651,R01-090:0.891955114,R01-091:0.726296959,R01-092:0.557369092,R01-093:0.827921054,R01-094:0.809129332,R01-095:0.713630679,R01-096:0.728150443,R01-097:0.445709849,R01-099:0.786909219,R01-101:0.826549971,R01-103:0.818544249,R01-105:0.800283429,R01-107:0.77209806,R01-109:0.526077667,R01-110:0.497560976,R01-111:0.511410894,R01-112:0.907062065,R01-113:0.44661508,R01-114:0.902224058,R01-115:0.78721174,R01-116:0.561519405,R01-117:0.570513745,R01-118:0.594700407,R01-119:0.61825917,R01-121:0.839111393,R01-122:0.519057377,R01-124:0.594308036,R01-125:0.734829593,R01-126:0.426915017,R01-127:0.191945712,R01-128:0.781319407,R01-129:0.538877476,R01-131:0.544844598,R01-132:0.557804821,R01-133:0.491422557,R01-134:0.431908166,R01-135:0.554446119,R01-136:0.407775136,R01-137:0.248216534,R01-138:0.835014493,R01-139:0.680349407,R01-140:0.858731552,R01-141:0.081384615,R01-142:0.703421009,R01-144:0.657289694,R01-145:0.787378659,R01-146:0.850732088"],
+			"limitations": "The model might produce minor false positives but this could be easilily removed by post-processing such as constrain the tumor segmentation only in lung slices"
+		},
+		"training": {
+			"title": "Training data",
+			"text": "Training data was from 377 data in the TCIA NSCLC-Radiomics data, references: Aerts, H. J. W. L., Wee, L., Rios Velazquez, E., Leijenaar, R. T. H., Parmar, C., Grossmann, P., Carvalho, S., Bussink, J., Monshouwer, R., Haibe-Kains, B., Rietveld, D., Hoebers, F., Rietbergen, M. M., Leemans, C. R., Dekker, A., Quackenbush, J., Gillies, R. J., Lambin, P. (2014). Data From NSCLC-Radiomics (version 4) [Data set]. The Cancer Imaging Archive."
+
+		},
+    	"analyses": {
+			"title": "Quantitative Analyses",
+			"text": "DSC was used to compute the accuracy of the model"
+		},
+		"limitations": {
+			"title": "Limitations",
+			"text": "The model might produce minor false positives but this could be easilily removed by post-processing such as constrain the tumor segmentation only in lung slices"
+		}
+    }
+}

models/msk_smit_lung_gtv/src/README.md

@@ -0,0 +1,4 @@
+## References
+[1] Jiang, Jue, and Harini Veeraraghavan. "Self-supervised pretraining in the wild imparts image acquisition robustness to medical image transformers: an application to lung cancer segmentation." In Medical Imaging with Deep Learning. 2024.
+
+[2] Jiang, Jue, Neelam Tyagi, Kathryn Tringale, Christopher Crane, and Harini Veeraraghavan. "Self-supervised 3D anatomy segmentation using self-distilled masked image transformer (SMIT)." In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 556-566. Cham: Springer Nature Switzerland, 2022.

models/msk_smit_lung_gtv/src/bash_run_SMIT_Segmentation.sh

@@ -0,0 +1,49 @@
+#!/bin/bash
+#
+#
+# Input arguments: 
+# $1 data_dir
+# $2 save_folder
+# $3 load_weight_name
+# $4 input_nifti
+
+source ./conda-pack/bin/activate
+
+#Use SMIT
+use_smit=1 #Use SMIT not SMIT+
+
+#Data folder and there need a 'data.json' file in the folder 
+data_dir="$1"
+
+#Segmentation output folder 
+save_folder="$2"
+
+#Some configrations for the model, no need to change
+#Trained weight 
+load_weight_name="$3"
+
+input_nifti="$4"
+
+a_min=-500
+a_max=500
+space_x=1.5
+space_y=1.5
+space_z=2.0
+out_channels=2
+
+python utils/gen_data_json.py $input_nifti
+
+python run_segmentation.py \
+    --roi_x 128 \
+    --roi_y 128 \
+    --roi_z 128 \
+    --space_x $space_x \
+    --space_y $space_y \
+    --space_z $space_z \
+    --data_dir $data_dir \
+    --out_channels $out_channels \
+    --load_weight_name $load_weight_name \
+    --save_folder $save_folder \
+	--a_min=$a_min \
+	--a_max=$a_max \
+    --use_smit $use_smit