SWSSL - Sliding window-based self-supervised learning for anomaly detection in high-resolution images (IEEE Trans. on Medical Imaging 2023)

By Haoyu Dong, Yifan Zhang, Hanxue Gu, Nicholas Konz, Yixin Zhang, and Maciej Mazurowski.

Paper Link here

This is the official repository for our image anomaly detection model SWSSL in IEEE Trans. on Medical Imaging 2023. In this paper, we extend anomaly detection to high-resolution images by proposing to train the network and perform inference at the patch level, through the sliding window algorithm. We further study the augmentation function in the context of medical imaging when learning augmentation-invariant features. In particular, we observe that the resizing operation, a key augmentation in general computer vision literature, is detrimental to detection accuracy, and the inverting operation can be beneficial. We also propose a new module that encourages the network to learn from adjacent patches to boost detection performance.

Model training and evaluation

Step 0: Data Acquisition:

You can obtain the Chest XRay dataset from here, and the DBT dataset from here. We are also happy to provide pre-processed data upon request.

Step 1: Model Running:

You can run "bash run_chest.sh" to obtain the performance reported in the paper. Running on the DBT dataset is very similar, but you need to change some hyperparameters as described in the paper.

Citation

Please cite our paper if you use our code or reference our work (published version citation forthcoming):

@article{dong2023swssl,
  title={SWSSL: Sliding window-based self-supervised learning for anomaly detection in high-resolution images},
  author={Dong, Haoyu and Zhang, Yifan and Gu, Hanxue and Konz, Nicholas and Zhang, Yixin and Mazurowski, Maciej A},
  journal={IEEE Transactions on Medical Imaging},
  year={2023},
  publisher={IEEE}
}