Wei Xia, et al. "Multicenter Evaluation of a Weakly Supervised Deep Learning Model for Lymph Node Diagnosis in Rectal Cancer on MRI." Radiology: Artificial Intelligence (2024): e230152.
Weakly supervISed model DevelOpment fraMework (WISDOM) to construct lymph node diagnosis model with preoperative MRI data coupled with postoperative patient-level pathological information.
- 0_pretraining_CIFIR10.py: use the CIFIR10 images to pretrain the intensity diagnostic network.
- 1_MI_training.py: build the intensity diagnostic model MI using the T2W-MRI image with weak supervision component.
- 2_MIS_training.py: build the integrated diagnostic model MIS using the short, long diameters, diamter ratio, and the img predictions of MI.
- 3_MISA_training.py: build the integrated diagnostic model MISA using the short, long diameters, diamter ratio, ADC value, and the img predictions of MI.
- 4_MISA_test_output_preds.py: generate the predictions of MISA for statistic analysis.
- 5_diagnostic_network_extract.py: extract the diagnostic networks.
- 6_Heatmap_generation.py: generate the heatmap by Grad-CAM to highlight the region related to metastasis.
- 7_pytorch_implementation: new implementation using pytorch. The size, ADC value, and image features were fused in one stage.