Abstract
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 medical images, such as barely visible abnormalities in chest X-rays and metastases in lymph nodes. 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 at all 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 natural image datasets with a known benchmark, as well as on two medical datasets containing radiology and digital pathology images. The proposed approach suggests a new strong baseline for image anomaly detection and outperforms state-of-the-art approaches in complex medical image analysis tasks.
Highlights
A NOMALY detection is a crucial task in the deployment of machine learning models, where knowing the “normal” data samples should help spot the “abnormal” ones [5], [6]
2) We introduce a new anomaly detection approach that utilizes the autoencoder with the perceptual loss
We propose the use of autoencoders that are much simpler to set up and train, while combining it with perceptual loss was shown to be more powerful than Generative Adversarial Networks (GANs) utilizing the perceptual loss
Summary
A NOMALY detection is a crucial task in the deployment of machine learning models, where knowing the “normal” data samples should help spot the “abnormal” ones [5], [6]. We present how to achieve a smooth growth of perceptual information in the loss function, and show that this greatly improves the quality of anomaly detection in the high-resolution medical data. A. CONTRIBUTIONS 1) We compare the three strongest SOTA anomaly detection methods (the hyperparameters of which we fine-tuned to their optima) in two challenging medical tasks: Chest X-rays and H&E-stained histological images. We further extend the proposed method with progressive growing training (in particular, we introduce how to gradually grow the perceptual information in the loss function), allowing us to adapt the anomaly detection to the high-resolution medical data. We believe that such a simple solution will standardize tuning of hyperparameters of different models, eliminate the ground for misunderstanding, and improve reproducibility
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have