Event date
Event location
Organisé par :
Meet the winner of the 3rd edition of our PhD Student Award!
This study presents LongiMam, an end-to-end deep learning model designed to improve breast cancer risk prediction by leveraging longitudinal mammography. Unlike most existing approaches that rely on single or limited prior exams, LongiMam integrates the current mammogram with up to four prior screenings using a combined convolutional–recurrent architecture.
The model was trained and evaluated on a large, population-based screening dataset characterized by substantial class imbalance and heterogeneous follow-up. Across multiple input configurations, incorporating prior exams with the current one suggested improved predictive performance. Priors alone were less effective, underscoring the importance of recent imaging. Subgroup analyses showed good performance in women with dense breasts, in those aged 55 years or older, and in women exhibiting changes in mammographic density over time.
These findings suggest that longitudinal modeling enhances breast cancer prediction and support the use of repeated mammograms to refine risk stratification in screening programs.
LongiMam is available as open-source software.
Keywords: Breast cancer, screening, mammography, deep learning neural networks, longitudinal data
