Weakly Supervised Semantic Segmentation on retina OCT images

Anomaly-Guided (AGM)

🎉 05/2024 Paper is accepted in Medical Image Analysis (MedAI):

Anomaly-guided weakly supervised lesion segmentation on retinal OCT images

Fully-supervised

semantic segmentation with pixel-level annotation on RESC dataset, for comparison

Structural and Textual Guided (Under Review)

🎉 11/2024 Paper is on arXiv preprint:

A Multimodal Approach Combining Structural and Cross-domain Textual Guidance for Weakly Supervised OCT Segmentation

Download data

1. The datasets, RESC and Duke, are used by all projects in this repo.

2. The datasets can be downloaded from this Google Drive link

3. The RESC and Duke are .tar.gz files which need to be extracted and put at the repo root path: tar -xzf datasets.tar.gz. The structure of /your_dir/datasets/ should be organized as follows:

---2015_BOE_Chiu/
       --segment_annotation
            --gan_healthy
            --images
            --labels
            --mask
       --train.csv
       --valid.csv

---oct_kaggle/
       --normal_mask
       --origin
            --test/DME
            --train/DME
       --test
            --0.normal
            --1.abnormal
       --train
            --0.normal
            --gan_healthy

---RESC/
       --mask
            --train
            --valid
       --train
            --gan_healthy
            --label_images
            --original_images
       --valid
            --gan_healthy
            --label_images
            --original_images
       --resc_cls_labels.npy