Symptom Disentanglement in Chest X-ray Images for Fine-Grained Progression Learning

The official repo for [MICCAI'24] "Symptom Disentanglement in Chest X-ray Images for Fine-Grained Progression Learning"

Introduction

We propose a multi-stage framework to model the complex progression from symptom perspective. Specifically, we introduce two consecutive modules namely Symptom Disentangler (SD) and Symptom Progression Learner (SPL) to learn from static diagnosis to dynamic disease development.

Methods

Installation

git clone https://github.com/zhuye98/SDPL.git 
cd SDPL

Install torch and torchvision required packages:

• torch == 2.0.0+cu118
• python == 3.8
• einops == 0.3.2

Data Pre-processing

Step 0: Download the datasets used in this paper:

• Chest Imagenome
• MIMIC-CXR-JPG

Step 1: Follow CheXRelFormer for the data pre-processing, and structured as follows:

data
 ├─A
 ├─B
 ├─label
 ├─list
 └─split
     ├─train.csv
     ├─val.csv
     └─test.csv

Step 2: Extract the symptom labels and regenerate the progression label:

python data_processing/get_symptom_label.py
python data_processing/get_progression_label.py

Three files are generated, including:

• symptom_label_chexpert.csv. This file records the ['subject_id', 'study_id', 'dicom_id', 'path', 'Lung Opacity', 'Pleural Effusion', 'Atelectasis', 'Cardiomegaly', 'Edema', 'Pneumothorax', 'Consolidation', 'Pneumonia', 'split'] for each case;
• symptom_label_chexpert.json. The dictionary of symptom labels {dicom_id: symptom_list}, i.e. "dicom_id": [NaN, 0.0, NaN, NaN, 0.0, 1.0, NaN, 0.0];
• progression_label.json. The dictionary of symptom labels {dicom_id_cur_dicom_id_pri: progression_label_list}, i.e. "dicom_id_cur_dicom_id_pri": [NaN, NaN, NaN, NaN, NaN, NaN, NaN, 2.0], which indicates 'no change' of symptom 'Pneumonia'.

Step 3: Regenerate the split:

python data_processing/regenerate_split.py

Finally, the data is structured as follows:

data
 ├─A
 ├─B
 ├─label
 ├─list 
 ├─my_list 
 ├─symptom_label_chexpert.csv
 ├─symptom_label_chexpert.json
 ├─progression_label.json
 ├─split
     ├─train.csv
     ├─val.csv
     └─test.csv

Training and Evaluation

cd code
sh scripts/run_CheXRelFormer.sh
sh scripts/run_SDPL.sh

sh scripts/eval_CheXRelFormer.sh
sh scripts/eval_SDPL.sh

Acknowledgements

Our code is origin from CheXRelFormer. We are grateful to these authors for their valuable contributions.