Official Repository for the MICCAI 2022 paper titled DRGen: Domain Generalization in Diabetic Retinopathy Classification
Domain Generalization is a challenging problem in deep learning especially in medical image analysis because of the huge diversity between different datasets. Existing papers in the literature tend to optimize performance on single target domains, without regards to model generalizability on other domains or distributions. High discrepancy in the number of images and major domain shifts, can therefore cause single-source trained models to under-perform during testing. In this paper, we address the problem of domain generalization in Diabetic Retinopathy (DR) classification. The baseline for comparison is set as joint training on different datasets, followed by testing on each dataset individually. We therefore introduce a method that encourages seeking a flatter minima during training while imposing a regularization. This reduces gradient variance from different domains and therefore yields satisfactory results on out-of-domain DR classification. We show that adopting DR-appropriate augmentations enhances model performance and in-domain generalizability. By performing our evaluation on 4 open-source DR datasets, we show that the proposed domain generalization method outperforms separate and joint training strategies as well as well-established methods.
The application code we use is based on backbone codes from both SWAD(domainbed)[1] and Fishr [2].
First install requirements using (make sure you are in the main drgen folder first):
pip install -r requirements.txt
Configuration available in config.yaml and further default arguments can be changed from train_all.py:
python train_all.py
- Cha, J., Cho, H., Lee, K., Park, S., Lee, Y., Park, S.: Domain generalization needs stochastic weight averaging for robustness on domain shifts. CoRR arXiv:2102.08604 (2021)
- Ramé, A., Dancette, C., Cord, M.: Fishr: invariant gradient variances for out-of-distribution generalization. CoRR arXiv:2109.02934 (2021)