Reconsidering Spatial Alignment for Longitudinal Breast Cancer Risk Prediction

Solveig Thrun, Stine Hansen, Zijun Sun, Nele Blum, Suaiba Amina Salahuddin, XIN WANG, Kristoffer Knutsen Wickstrøm, Elisabeth Wetzer, Robert Jenssen, Maik Stille, Michael Kampffmeyer

Keywords: Breast cancer, Risk prediction, Longitudinal mammograms, Registration

Abstract: Regular mammography screening is vital for early breast cancer detection, and deep learning enables more personalized strategies. However, misalignment across time points can obscure subtle tissue changes and reduce prediction accuracy. This study evaluates image-based, feature-level, and implicit alignment methods on two large mammography datasets, showing that our proposed image-based registration model achieves the highest accuracy and anatomically plausible deformations, highlighting the importance of precise alignment in longitudinal risk prediction.

Serve As Reviewer: ~Solveig_Thrun1

Submission Number: 13