Improving model fairness in image-based computer-aided diagnosis

Datasets

The first dataset is provided by Medical Imaging and Data Resource Center (MIDRC) and is available through this website (https://data.midrc.org/). The AREDS dataset is publicly available on NCBI dbGAP (https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000001.v3.p1). The OHTS dataset is available upon request due to patient protection (https://ohts.wustl.edu/). The MIMIC-CXR dataset is publicly available on PhysioNet (https://www.physionet.org/content/mimic-cxr-jpg/).

Getting started

Prerequisites

• python >=3.6
• pytorch = 1.11.0
• torchvision = 0.12.0
• sklearn = 0.23.2
• pandas = 1.4.1
• opencv = 4.5.0
• skimage = 0.17.2
• tqdm = 4.48.2
• json = 0.9.6
• pickle = 2.2.1

Quickstart

I used the experiment on the MIMIC-CXR dataset on the intersectional groups as an example.

python train_mimic_intersection.py

Reference

Lin M, Li T, Yang Y, Holste G, Ding Y, Van Tassel SH, Kovacs K, Shih G, Wang Z, Lu Z, Wang F, Peng Y. Improving model fairness in image-based computer-aided diagnosis. Nat Commun. 2023 Oct 6;14(1):6261. doi: 10.1038/s41467-023-41974-4. PMID: 37803009; PMCID: PMC10558498.

Acknowledgment

This work was supported by the National Library of Medicine under Award No. 4R00LM013001, NSF CAREER Award No. 2145640, and Amazon Research Award.