Anomaly Detection in Medical Imaging With Deep Perceptual Autoencoders

Anomaly detection is the problem of recognizing abnormal inputs based on the seen examples of normal data. Despite recent advances of deep learning in recognizing image anomalies, these methods still prove incapable of handling complex images, such as those encountered in the medical domain. Barely visible abnormalities in chest X-rays or metastases in lymph nodes on the scans of the pathology slides resemble normal images and are very difficult to detect. To address this problem, we introduce a new powerful method of image anomaly detection. It relies on the classical autoencoder approach with a re-designed training pipeline to handle high-resolution, complex images, and a robust way of computing an image abnormality score. We revisit the very problem statement of fully unsupervised anomaly detection, where no abnormal examples are provided during the model setup. We propose to relax this unrealistic assumption by using a very small number of anomalies of confined variability merely to initiate the search of hyperparameters of the model. We evaluate our solution on two medical datasets containing radiology and digital pathology images, where the state-of-the-art anomaly detection models, originally devised for natural image benchmarks, fail to perform sufficiently well. The proposed approach suggests a new baseline for anomaly detection in medical image analysis tasks.