PatchCL-AE: Anomaly detection for medical images using patch-wise contrastive learning-based auto-encoder

Shuai Lu, Weihang Zhang, Jia Guo, Hanruo Liu, Huiqi Li, Ningli Wang

Abstract: Highlights•A patch-wise contrastive learning-based auto-encoder is proposed to optimize local semantic reconstruction for medical anomaly detection.•A novel local semantic discrepancy-based anomaly score is proposed to achieve improved localization to assist clinicians in lesion detection.•Experiments validate that the proposed method outperforms state-of-the-art anomaly detection results across three modalities, i.e., brain MRI, retinal OCT, and chest X-ray images.