An Integrated Tool for Data-driven Breast Cancer Screening and Risk Assessment
Early detection of breast cancer is essential for efficient subsequent treatments. Current screening guidelines are designed for the "average woman", and hence perform poorly for many subgroups of women, failing to detect many malignant tumors while at the same time yielding many false positives. This work develops and applies an integrated tool (Hippolyta) for data-driven breast cancer risk stratification and screening. Hippolyta relies on a Hidden Markov Model to understand the patient's longitudinal breast cancer state trajectory and on Partially Observable Markov Decision Processes to learn a personalized screening policy for each individual patient. Hippolyta greatly outperforms all previous algorithms; for example, the rate of false positives (which often result in unnecessary invasive and costly procedures) is reduced by almost 40% (while maintaining the same rate of correct detection.
Prof. Mihaela van der Schaar
Dr. Camelia Davtyan
Dr. Arash Naeim