Comparison of subject-to-template registration schemes using CT and MR radiotherapy images with brain lesions

IntroductionVoxel-based analyses have been used more widely in radiotherapy in recent years. The purpose of this study is to compare eight different methods of registering images on to a template space for such analyses. A novel way, using both CT and MR data, is proposed.MethodsCT and MR brain images from 85 participants in the CoDe-B-Rad study (NCT06466720) were registered on an age-specific template. The registration schemes used included sCT-Template (linear and non-linear); MR-Template (linear, non-linear, masked, and enantiomorphic); and Dual-Template (linear and non-linear). The registrations were compared qualitatively and quantitatively against the template MR images with scores ranging from 1 (lowest) to 5 (highest) and quantitatively (via Jaccard, ASD, and HD95).ResultsQualitatively, the best registration scheme was the Dual-Template-non-linear registration, with 60 participants scoring above 4. The second best was the MR-Template-enantiomorphic, with 53 participants scoring above 4. Quantitatively, the Dual-Template-non-linear method outperformed on the mean (±SD) for the ASD and HD95, with 0.918 (±0.1774) and 2.965 (±0.5392) respectively. For ASD the difference compared to other methods was significant (p = 0.03). The MR-Template-masked outperformed on the Jaccard mean (±SD), median (IQR), and ASD, achieving values of 0.567 (±0.0557), 0.555 (0.055), and 0.808 (0.162) respectively. The Dual-Template-non-linear and MR-Template-masked and non-linear had the same result for the median HD95: 2.639. The performance of all linear schemes was inadequate both quantitatively and qualitatively. All non-linear registration schemes had issues with distortions of tissue and landmarks, however, for the Dual-Template scheme these were minimal.ConclusionThe Dual-Template-non-linear registration scheme is a new way of registering lesioned brain images for use with voxel-wise techniques in radiotherapy, which utilises both CT and MR image data. The scheme provides fidelity of the underlying soft tissue as well as the surrounding skull, minimising anatomical and dosimetric distortions.