Anomaly-Driven Approach for Enhanced Prostate Cancer Segmentation

2025-04-30

AI Summary

Prostate cancer MRI segmentation faces challenges including severe class imbalance, variable lesion scales, and scarcity of expert annotations. Method: We propose Anomaly-Driven U-Net (adU-Net), the first framework to incorporate dual-parametric MRI (T2-weighted + ADC) anomaly maps—generated via Fixed-Point GAN—as prior-guidance signals directly into the U-Net encoder; notably, the ADC-derived anomaly map enhances sensitivity to clinically significant prostate cancer (csPCa). This design requires no additional manual annotations. Contribution/Results: adU-Net effectively mitigates small-lesion under-detection and limited generalizability. On an external test cohort, it achieves a combined AUROC–Average Precision score of 0.618, significantly outperforming nnU-Net (0.605), thereby improving csPCa localization accuracy and robustness. The method establishes a novel, interpretable, and transferable paradigm for weakly supervised medical image segmentation.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Segmentation
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising

Abstract

Magnetic Resonance Imaging (MRI) plays an important role in identifying clinically significant prostate cancer (csPCa), yet automated methods face challenges such as data imbalance, variable tumor sizes, and a lack of annotated data. This study introduces Anomaly-Driven U-Net (adU-Net), which incorporates anomaly maps derived from biparametric MRI sequences into a deep learning-based segmentation framework to improve csPCa identification. We conduct a comparative analysis of anomaly detection methods and evaluate the integration of anomaly maps into the segmentation pipeline. Anomaly maps, generated using Fixed-Point GAN reconstruction, highlight deviations from normal prostate tissue, guiding the segmentation model to potential cancerous regions. We compare the performance by using the average score, computed as the mean of the AUROC and Average Precision (AP). On the external test set, adU-Net achieves the best average score of 0.618, outperforming the baseline nnU-Net model (0.605). The results demonstrate that incorporating anomaly detection into segmentation improves generalization and performance, particularly with ADC-based anomaly maps, offering a promising direction for automated csPCa identification.

Problem

• Improving prostate cancer segmentation in MRI using anomaly-driven methods
• Addressing data imbalance and lack of annotations in csPCa detection
• Enhancing tumor identification via anomaly maps in deep learning models

Innovation

• Anomaly-Driven U-Net for prostate cancer segmentation
• Fixed-Point GAN generates anomaly maps
• Integration improves AUROC and AP scores