Toward a Deep Learning Prediction of the Biodistirbution of Radioactivity in Selective Internal Radiotherapy

Selective Internal Radio Therapy, is a treatment for hepatocellular carcinoma, consisting in the administration of radioactive microsphere inside the arterial blood-flow feeding the tumors [1]. It is an effective treatment that requires a precise planning to maximize the radioactive activity delivered to the tumor and minimize the toxicity to healthy structures. It is in this clinical context that we are developing a deep learning assisted method to infer a patient specific activity distribution from pre-interventional, diagnostic images, to assist the planning and patient selection for the SIRT. Two networks are presented in this study, one using a diagnostic scanner, guided by post-operative information, and a second, using only data derived from pre-operative imagery.