Panda: Unsupervised Pelvic Anomaly Detection for Real-Time MR Imaging

2026-07-27

AI Summary

This work addresses the challenge of real-time anomaly detection in female pelvic MRI, where physiological motion, tissue deformation, and instrument-related artifacts complicate reliable identification. To circumvent the need for annotated data, the authors propose an unsupervised anomaly detection method built upon the Dinomaly framework. By freezing a pretrained DINOv2 Vision Transformer (ViT) encoder and coupling it with a noisy MLP bottleneck and a linear-attention decoder, this approach uniquely integrates a frozen ViT with a lightweight decoding architecture—effectively mitigating identity mapping while balancing accuracy and computational efficiency. Token-level cosine distances yield interpretable spatial anomaly maps, achieving 88.06% pixel-level AUROC and 95.45% frame-level specificity on a uterine fibroid subset, with an inference speed of 40.5 slices per second, thereby meeting clinical requirements for real-time deployment.

Abstract

Female pelvic diseases remain an under researched area characterized by often delayed diagnosis. While pelvic MRI offers superior soft-tissue contrast for diagnosis and image-guided procedures, real-time anomaly detection remains challenging due to physiological motion, tissue deformation, and instrument artifacts. Existing supervised approaches are impractical, as adverse events are rare, heterogeneous, and difficult to annotate. We present a Dinomaly-based unsupervised anomaly detection framework adapted for pelvic MRI that learns normative representations from healthy cases and flags deviations without requiring labels. Our approach leverages a frozen DINOv3 Vision Transformer encoder combined with a noisy MLP bottleneck and Linear Attention decoder to prevent identity mapping while maintaining computational efficiency. Anomalies are localized via per-token cosine distance between encoder and decoder representations, yielding spatial anomaly maps that provide immediate feedback at the scanner to support radiologist decision-making and adaptive protocol adjustment. Evaluated on a curated subset of the Uterine Myoma Dataset, the framework achieves a pixel-level AUROC of 88.06% and high specificity (95.45%) at frame level at 40.5 slices/s, meeting real-time clinical deployment requirements. The spatial anomaly maps and frame-level scores provide immediate, localized feedback at the scanner to support radiologist decision-making and adaptive protocol adjustment during active procedures.

Problem

• pelvic MRI
• anomaly detection
• real-time imaging
• unsupervised learning
• medical imaging

Innovation

• unsupervised anomaly detection
• real-time MRI
• Vision Transformer
• spatial anomaly mapping
• DINOv3