Predicting long term breast cancer risk using longitudinal mammographic screening history

Risk assessment of breast cancer (BC) seeks to enhance individualized screening and prevention strategies. Recent deep learning (DL) risk models based on mammography have shown superiority in short-term risk prediction compared to traditional risk factor-based models. However, these models primarily rely on single-time exams, which emphasize the detection of existing lesions and may ignore the temporal changes in breast tissues. In this study, we present the Multi-Time Point Breast Cancer Risk Model (MTP-BCR), a novel temporospatial DL risk model that integrates traditional BC risk factors and longitudinal mammography data to identify subtle changes in breast tissue indicative of future malignancy. Utilizing a large inhouse dataset comprising risk factors and 171,168 mammograms involving 9,133 women, we evaluate the performance of the MTP-BCR model in long-term risk prediction. Our model demonstrates a significant improvement in 10-year risk prediction with an area under the receiver operating characteristics (AUC) of 0.80, outperforming the traditional BCSC 10-year risk model, our pure image model (without risk factors), and is also superior to other SOTA methods at 5-year AUC in various screening cohorts. Furthermore, MTP-BCR provides unilateral breast-level predictions with AUCs up to 0.81 and 0.77 for 5-year and 10-year risk assessments, respectively. External validation in the public CSAW-CC dataset demonstrates the consistent advantage of our multi-time point-based model compared to the single-time point-based method. For the prediction of breast cancer recurrence, our MTP-BCR obtains a 5-year AUC of 0.71 which also surpasses other methods. The heatmaps derived from our model may help clinicians better understand the progression from normal tissue to cancerous growth, enhancing interpretability in breast cancer risk assessment.