LEARN: Learning Environment for Artificial Intelligence in Radiotherapy New Technology

Overview

This study will develop a whole-of-body markerless tracking method for measuring the motion of the tumour and surrounding organs during radiation therapy to enable real-time image guidance. Routinely acquired patient data will be used to improve the training, testing and accuracy of a whole-of-body markerless tracking method. When the markerless tracking method is sufficiently advanced, according to the PI of each of the data collection sites, the markerless tracking method will be run in parallel to, but not intervening with, patient treatments during data acquisition.

This observational study will access routinely acquired radiation therapy treatment data from 300 patients including brain, breast, head and neck, kidney, liver, pancreas, prostate, spine and cardiac anatomic sites. At least 30 patients will be recruited from each anatomic site to enable sufficient data for the markerless tracking method training, testing and validation. The clinical data will be used to develop, train, test and validate a markerless target tracking method. After the treatment, the ground truth and the variability in the ground truth will be computed. The patient images, the markerless tracking results, the ground truth and the variability will be uploaded to an in-house developed clinical trial learning system. Uploading additional data to the learning system automatically triggers the model building of the deep learning system. In this manner, the learning system gets both more accurate and more robust with each patient accrued. As the patient data accrues, the primary hypothesis of targeting accuracy can be tested. The developed markerless tracking software will be applied by study personnel to the treatment imaging data for each anatomic site using five-fold cross-validation where 80% of the data is used for training and the remaining unseen 20% of the data is used for testing. Target positions produced by the markerless tracking will be compared with a 'ground truth'.

Conditions