Next frontier in radiation therapy: AI-based real-time adaptive planning


Cancer is the chimera of modern society in terms of health-related diseases in both developed and developing countries. It is expected that about 40% of men and women will be diagnosed with cancer at some point in their lives. In developed countries, nearly half of cancer patients are treated with radiotherapy during their illness. Imaging is the main tool for planning and therapeutic guidance in radiotherapy. The standard process involves the acquisition of high-resolution anatomical images on which a manual annotation is made to delimit the structures of interest (target region and regions to be saved). However, it is a long and tedious task, and not very reproducible. But despite tremendous advances in equipment, radiation oncologists agree that the current treatment planning system has several shortcomings, such as poor tumor response prediction and suboptimal dosage management plans for external radiotherapy. Radiotherapy also requires positioning the patient in the same reference system as that used for treatment planning, through an image-based transformation of the patient’s anatomy. In such a context – and in a general way – the deformation of organs, due to the loss of weight of the patient or the elastic nature of the organs, is not taken into account in the dose administration. The overall objective is to propose a patientspecific treatment planning system based on recent developments in machine learing, adapting to the changing patient throughout radiotherapy with optimal dose planning using interventional MRI. We have designed a multi-disciplinary/multi-center collaboration to train and validate predictive models on over 20,000 retrospective cases treated over a 10-yr period at CHUM and MUHC, with a particular focus on head/neck cancers for demonstration purposes. These innovations will help radiation oncologists and medical physicists detect not only tumors during treatment, but also optimally avoid critical structures surrounding the tumor. By better avoiding healthy tissue, the treatment could be concentrated on a reduced number of treatments, thus improving convenience and access to treatment.