Reconsidering Explicit Longitudinal Mammography Alignment for Enhanced Breast Cancer Risk Prediction

2025-06-24

AI Summary

This study investigates the optimal paradigm for explicit image alignment in longitudinal mammography-based breast cancer risk prediction, focusing on two critical design choices: alignment space (input vs. representation space) and optimization strategy (joint vs. independent). We propose performing image-level explicit alignment in the input space via deformable registration and optimizing the alignment and risk prediction tasks independently. Experiments demonstrate that joint optimization in the representation space degrades alignment quality, whereas our decoupled approach significantly improves both deformation field accuracy and risk prediction performance. Comparative evaluations against implicit alignment methods—including Transformer-based architectures—confirm the superiority of explicit input-space alignment. Our method achieves higher risk prediction accuracy in long-term screening scenarios. The implementation is publicly available.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Optimization
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Regular mammography screening is essential for early breast cancer detection. Deep learning-based risk prediction methods have sparked interest to adjust screening intervals for high-risk groups. While early methods focused only on current mammograms, recent approaches leverage the temporal aspect of screenings to track breast tissue changes over time, requiring spatial alignment across different time points. Two main strategies for this have emerged: explicit feature alignment through deformable registration and implicit learned alignment using techniques like transformers, with the former providing more control. However, the optimal approach for explicit alignment in mammography remains underexplored. In this study, we provide insights into where explicit alignment should occur (input space vs. representation space) and if alignment and risk prediction should be jointly optimized. We demonstrate that jointly learning explicit alignment in representation space while optimizing risk estimation performance, as done in the current state-of-the-art approach, results in a trade-off between alignment quality and predictive performance and show that image-level alignment is superior to representation-level alignment, leading to better deformation field quality and enhanced risk prediction accuracy. The code is available at https://github.com/sot176/Longitudinal_Mammogram_Alignment.git.

Problem

• Optimizing explicit mammography alignment for better risk prediction
• Comparing input vs. representation space alignment effectiveness
• Balancing alignment quality and predictive performance trade-offs

Innovation

• Explicit alignment in representation space
• Joint optimization of alignment and risk
• Image-level alignment improves prediction accuracy