Validation of Multi-modality Image Registration for Glioblastoma Treatments on the MR-Linac

Motivation: Improve consistency, efficiency and accuracy of verifying image registrations, which are a potential major failure mode in MR-Linac treatments. Goal(s): Develop a quantitative method for automatically verifying image co-registration. Approach: Brain MR and CT image registrations were assessed using five different similarity metrics. Validation of registration involved comparing the vendor-provided registration algorithm with an independent algorithm to identify the most useful image similarity metrics. Results: A process for automated registration analysis was developed. SSIM and MSE were the most promising metrics for measuring registration accuracy. Impact: A new tool was developed for verifying the accuracy of registration of MR images, which could be used to automatically detect inaccurate registrations during the treatment workflow on the MR-Linac.