Explainability.

Oncology is a unique field of medicine, as there are rarely any cut and dried decisions. That means an extra heavy reliability on predictions and probabilities, which patients just loooove. Oncologists spend years memorizing and synthesizing the stats associated with each treatment for each stage of each disease, all while artfully incorporating our wisdom and gestalt related to individual patient performance status, values and resources. Who can do it better? Machines, of course. This review touches on the real-world utility of machine-learning regarding individualized predictions of outcomes following radiation treatment by not only assessing its accuracy but also its interpretability by physicians and patients. TBL: Digitized predictions of outcomes are a welcome future for oncology, but it has to be easily understood by physicians and easily explained to patients. | Luo, Br J Radiol 2019