Awesome

Longitudinal-Chest-X-Ray

This is the implementation of Utilizing Longitudinal Chest X-Rays and Reports to Pre-Fill Radiology Reports.

The preprocess data of MIMIC-CXR is from https://github.com/cuhksz-nlp/R2Gen.

The structure of annotation.json is as follows:

{ "train": [ { "id": "XXXXXX", "study_id": XXXXXX, "subject_id": XXXXX, "report": "XXXXXXXXXX", "image_path": ["p10/p10000032/s50414267/XXXXXXXXX.jpg"], "split": "train" } ] }

Wr arrange data based on the "StudyDate" from https://physionet.org/content/mimic-cxr-jpg/2.0.0/ Metadata files.

I have also extracted the image paths from the annotation.json file and added them as train.json, test.json, and val.json in this GitHub repository

Run bash run_mimic_cxr.sh to train a model on the data.