We propose LMV-Net, a longitudinal multi-view model that combines explicit longitudinal alignment with analysis of CC and MLO mammograms.

By Solveig Thrun, Doctoral Research Fellow at SFI Visual Intelligence

Breast cancer is the most common cancer among women worldwide, and mammography screening is essential for early detection. However, current screening programs recommend the same screening interval for all women, despite differences in individual breast cancer risk.

More accurate risk prediction models could enable personalized screening by identifying women who would benefit from more frequent examinations while reducing unnecessary screening for those at lower risk.

Recent advances in deep learning have shown that mammography images contain valuable information for predicting a woman's risk of developing breast cancer up to five years in advance. Early models analyzed only the most recent mammogram, whereas newer approaches incorporate prior screening examinations to capture subtle tissue changes over time, more closely reflecting the workflow of radiologists. Mammograms are acquired in two complementary views, the craniocaudal (CC) and mediolateral oblique (MLO) views, which together provide a more complete representation of the breast.

Most longitudinal models rely on implicit alignment relying on the network to learn temporal correspondence without explicitly enforcing spatial consistency between examinations. However, breast tissue undergoes substantial non-rigid deformation due to differences in positioning and compression across screening visits, making direct feature comparisons challenging.

Recent work has demonstrated that explicitly aligning mammograms across time improves the detection of subtle anatomical changes. Existing explicit alignment methods, however, process each mammographic view independently without jointly modeling the complementary information available across CC and MLO views.

To address these limitations, we propose LMV-Net, a longitudinal multi-view model that combines explicit longitudinal alignment with joint analysis of CC and MLO mammograms. The model aligns prior and current examinations for each view using image-based deformation fields before integrating longitudinal information through a dual-stream attention mechanism. This enables LMV-Net to capture subtle temporal changes while effectively combining complementary information from both mammographic views.

We evaluate LMV-Net on the public EMBED and CSAW-CC breast cancer screening datasets and compare it with several state-of-the-art risk prediction models. Our results demonstrate consistent improvements in predicting breast cancer risk up to five years in advance, across the overall population as well as different breast density groups and cancer subtypes. These findings highlight the potential of longitudinal multi-view analysis to improve breast cancer risk prediction and support the development of personalized screening strategies.

Accurate breast cancer risk prediction from screening mammography is critical for enabling personalized screening intervals and early detection. Recent deep learning methods have shown the value of longitudinal data and explicit temporal alignment. However, existing approaches either perform explicit alignment using a single mammographic view or model multiple views without explicit longitudinal alignment, limiting their ability to exploit the complementary spatial–temporal information used in clinical practice. To address this gap, we propose LMV-Net, a longitudinal multi-view breast cancer risk prediction model that jointly analyzes anatomically complementary CC and MLO views within an explicitly aligned longitudinal framework. We evaluate our approach on the public EMBED and CSAW-CC datasets, comparing it to state-of-the-art breast cancer risk prediction methods. Our model consistently outperforms existing approaches in overall risk prediction performance and across different breast density and cancer subgroups. Importantly, these improvements highlight the potential of longitudinal multi-view modeling to enhance risk stratification, paving the way for future work on personalized screening, earlier identification of high-risk patients, and more efficient screening resource allocation. The code is available at https://github.com/sot176/LMV-Net.