In-batch Relational Features Enhance Precision in An Unsupervised Medical Anomaly Detection Task

TL;DR

This paper introduces a population-aware embedding method that uses in-batch relational features to significantly improve unsupervised medical image anomaly detection accuracy, reducing false positives.

Contribution

The novel approach augments CNN autoencoder representations with batch-wise relational features via hypergraph estimation and graph convolution, enhancing anomaly detection in medical images.

Findings

Achieved 0.90 AUC-ROC, a 5.7% absolute gain over baseline.

Improved average precision by 16%, reaching 0.78 AP.

Performance scales positively with batch size, enabling tunable healthy variation integration.

Abstract

Confounding pathology with normal anatomical variation remains a significant challenge in unsupervised medical-image anomaly detection, resulting in numerous false positives. To enhance integration of healthy variation, we augment the latent representation of a CNN autoencoder with contextual similarities within a normal cohort through batch-wise hypergraph estimation and a shared-weights graph convolution layer, producing a population-aware embedding. On a heterogeneous brain-tumor dataset of 2D MRI scans, the method improves separability between healthy and pathological samples, achieving an AUC-ROC of 0.90 (95% CI 0.84-0.95, 5.7% absolute gain), and a 16% absolute improvement in average precision (0.78 AP, 95% CI 0.66-0.89), thereby lowering false-positive rates. Moreover, both anomaly detection and downstream tumor versus no-tumor classification performance improve with the size of the mini-batch context captured in the augmented representation, suggesting a tunable lever for integrating healthy variation.