Cinepro: Robust Training of Foundation Models for Cancer Detection in Prostate Ultrasound Cineloops

2025-01-21

AI Summary

Localizing prostate cancer regions in ultrasound cineloops remains challenging under weak supervision, as biopsy-derived labels provide only coarse-grained tumor proportion estimates (introducing substantial label noise) and lack anatomical priors and temporal context modeling. Method: We propose the first weakly supervised localization framework specifically designed for ultrasound cineloops, featuring: (i) a proportion-aware robust loss function that explicitly incorporates pathological tumor proportions into the supervision signal to mitigate label noise; (ii) multi-frame temporally consistent strong data augmentation to enhance stability of cancer-related features; and (iii) a multi-center weakly supervised training paradigm integrating foundation models with global anatomical context. Results: Evaluated on a multi-center dataset, our method achieves an AUROC of 77.1% and a balanced accuracy of 83.8%, significantly outperforming existing approaches.

Technology Category

• Machine Learning: Semi-Supervised Learning
• Computer Vision: Multi-modal Vision
• Search and Optimization: Local Search

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: On-Device user modeling, personalization, and recommendation

Abstract

Prostate cancer (PCa) detection using deep learning (DL) models has shown potential for enhancing real-time guidance during biopsies. However, prostate ultrasound images lack pixel-level cancer annotations, introducing label noise. Current approaches often focus on limited regions of interest (ROIs), disregarding anatomical context necessary for accurate diagnosis. Foundation models can overcome this limitation by analyzing entire images to capture global spatial relationships; however, they still encounter challenges stemming from the weak labels associated with coarse pathology annotations in ultrasound data. We introduce Cinepro, a novel framework that strengthens foundation models' ability to localize PCa in ultrasound cineloops. Cinepro adapts robust training by integrating the proportion of cancer tissue reported by pathology in a biopsy core into its loss function to address label noise, providing a more nuanced supervision. Additionally, it leverages temporal data across multiple frames to apply robust augmentations, enhancing the model's ability to learn stable cancer-related features. Cinepro demonstrates superior performance on a multi-center prostate ultrasound dataset, achieving an AUROC of 77.1% and a balanced accuracy of 83.8%, surpassing current benchmarks. These findings underscore Cinepro's promise in advancing foundation models for weakly labeled ultrasound data.

Problem

• Prostate Cancer Detection
• Imprecise Annotations
• Structural Information Incorporation

Innovation

• Cinepro method
• label noise reduction
• temporal information enhancement