Adaptive radiotherapy: image registration that takes the clinical record into account increases the proportion of acceptable contours from 45 per cent to 82.5 per cent

Based on an analysis of 16,361 CT scans from 3,473 patients, CAP-DIR utilises contours, dose data, clinical notes and patient history to track anatomical changes throughout the course of treatment.

During radiation therapy, the patient's anatomy evolves: the tumor regresses, the organs move. Adapting the treatment plan involves matching precisely targets, organs at risk and dose from one scanner to another. However, current recalibration methods treat each pair of images in isolation, without knowing the context of radiation therapy or the patient's previous evolution.

An informed and longitudinal bundling

CAP-DIR, developed by a team of many radiotherapists, integrates the original contours, dose card, clinical text and characteristics of the previous recalings of the same patient. Assessed from 16,361 CT examinations of 3,473 patients from four institutional cohorts and five public games, it brings the standardized cross-correlation in risk organ regions from 84.76% to 96.92% compared to the best learning competitor, and remains at the top of seven external cohorts.

In a blind reading of 20 cases followed over time, the proportion of clinically acceptable contours increased from 45% to 82.5% for target volumes and from 22.5% to 62.5% for at-risk organs, with an estimated reduction in correction time. The overall dose deferral error also decreased by 34.5%.