Paper: https://arxiv.org/pdf/2311.13752.pdf
Data:
- Input Data: We use Medical Segmentation Decathlon corresponding to four organs: Colon, Liver, Lung and Pancreas.
- Training/test splits: https://github.com/abachaa/3D-MIR/tree/main/Data
Code:
1) Organ Segmentation: We use Total Segmentator to segment and index individual organs.
- (a) Organ Segmentation: run_totalsegmentator.py
2) Image Processing and Quantification
-
(a) Liver Quantification: msd_colon_2D_3D_metrics_extraction.ipynb
-
(b) Pancreas Quantification: msd_pancreas_2D_3D_metrics_extraction.ipynb
-
(c) Colon Quantification: msd_colon_2D_3D_metrics_extraction.ipynb
-
(d) Lung Quantification: msd_lung_2D_3D_metrics_extraction.ipynb
3) Embeddings Generation (BiomedCLIP)
4) Retrieval Methods & Evaluation:
-
(a) Slice-based Retrieval: Method-1-Slice-based-Retrieval.ipynb (described in Section 4.1)
-
(b) Volume-based Retrieval: Method-2-Volume-based-Retrieval.ipynb (described in Section 4.2)
-
(c) Multi-modal Retrieval: Method-3-Multimodal-Retrieval.ipynb (described in Section 4.3) and GPT-4-based captions