Breast Cancer Detection from Multi-View Screening Mammograms with Visual Prompt Tuning

TL;DR

Breast Cancer Detection from Multi-View Screening Mammograms with Visual Prompt Tuning presents MVPT-NET, a two-stage framework that combines high-resolution single-view pretraining with multi-view visual prompt tuning to efficiently integrate MLO and CC mammograms. The backbone Swin Transformer is kept frozen during multi-view learning, while a small set of learnable prompts enables cross-view fusion and knowledge distillation between single-view and multi-view predictions. Evaluated on a large OPTIMAM-derived dataset from four NHS screening sites, MVPT-NET achieves state-of-the-art AUROC: 0.842 for Benign vs Malignant and 0.852 for Benign vs DCIS vs Invasive, outperforming several multi-view baselines. The results demonstrate the practicality of prompt-based fine-tuning to preserve high-resolution detail without heavy downsampling, with implications for scalable breast cancer screening across institutions.

Abstract

Accurate detection of breast cancer from high-resolution mammograms is crucial for early diagnosis and effective treatment planning. Previous studies have shown the potential of using single-view mammograms for breast cancer detection. However, incorporating multi-view data can provide more comprehensive insights. Multi-view classification, especially in medical imaging, presents unique challenges, particularly when dealing with large-scale, high-resolution data. In this work, we propose a novel Multi-view Visual Prompt Tuning Network (MVPT-NET) for analyzing multiple screening mammograms. We first pretrain a robust single-view classification model on high-resolution mammograms and then innovatively adapt multi-view feature learning into a task-specific prompt tuning process. This technique selectively tunes a minimal set of trainable parameters (7\%) while retaining the robustness of the pre-trained single-view model, enabling efficient integration of multi-view data without the need for aggressive downsampling. Our approach offers an efficient alternative to traditional feature fusion methods, providing a more robust, scalable, and efficient solution for high-resolution mammogram analysis. Experimental results on a large multi-institution dataset demonstrate that our method outperforms conventional approaches while maintaining detection efficiency, achieving an AUROC of 0.852 for distinguishing between Benign, DCIS, and Invasive classes. This work highlights the potential of MVPT-NET for medical imaging tasks and provides a scalable solution for integrating multi-view data in breast cancer detection.

Figures (1)

• Figure 1: Network architecture of MVPT-NET.