FR

PRE-ACT - Prediction of Radiotherapy side Effects using explainable AI

Ingénierie et Architecture

Radiotherapy is a widely used cancer treatment, however some patients suffer side effects. In breast cancer, side effects can include breast atrophy, arm lymphedema, and heart damage. Some factors that increase risk for side effects are known, but current approaches do not use all available complex data.

Prediction for risk of side effects from several different types of data: clinical; images; type of cancer therapy; and genomics.

Radiotherapy is a widely used cancer treatment, however some patients suffer side effects. In breast cancer, side effects can include breast atrophy, arm lymphedema, and heart damage. Some factors that increase risk for side effects are known, but current approaches do not use all available complex imaging and genomics data. The time is now ripe to leverage the huge potential of AI towards prediction of side effects. This project will use rich datasets from three patient cohorts to design and implement an AI tool that predicts the risk of side effects, including arm lymphedema in breast cancer patients and provides an easily understood explanation to support shared decision-making between the patient and physician.

The PRE-ACT consortium combines the expertise in computing (MDW, AUEB-RC), AI (HES-SO, CENTAI), radiation oncology (MAASTRO, UNICANCER), medical physics (THERA), genetics (ULEIC), psychology (CNR) and health economics (UM) that is necessary to tackle this problem.

The project will integrate data from the three cohorts and build AI predictive models with built-in explainability for each of the key side effects of breast cancer radiotherapy. These AI models will be incorporated into an existing commercial radiotherapy software platform to create a world-leading product. The extended platform will be validated in a clinical trial to support treatment decisions regarding the irradiation of lymph nodes. The trial will adopt an innovative design in which the patients and medical team in the test arm will receive the risk prediction, but those in the control arm will not. A communication package built with a co-design methodology will ensure that AI outcomes are tailored to stakeholders effectively. The trial will evaluate whether using the AI platform changed the arm lymphedema rate and impacted treatment decisions and quality-of-life. Generalizability of the AI models for other types of cancer will be sought through transfer learning techniques.