Author: Danna Gurari, Moyed Miften, Sarah Milgrom, Atharva Rajesh Peshkar, Willem Schreuder, David H. Thomas 👨🔬
Affiliation: University of Colorado Boulder, University of Colorado School of Medicine, University of Colorado Anschutz, Thomas Jefferson University 🌍
Purpose: To evaluate the accuracy of a novel avatar-based patient positioning technique.
Methods: We developed a modified surface-guided radiation therapy (SGRT) technique, 'Avatar-Guided Radiation Therapy' (AgRT), which uses patient-specific 'avatars' to predict anatomy from surface scans. Unlike SGRT, AgRT leverages realistic, patient-specific full-body surface models combined with skeletal models predicted from surface scans. AgRT incorporates prior knowledge from the patient's CT imaging to establish a correlation map between a patient's surface imaging and internal anatomy. To evaluate positioning accuracy, we compared SGRT and AgRT performance using a retrospective longitudinal dataset of 30 patients, each with multiple CBCT scans. Daily CT scans were used as surrogates for CBCT scans, with extracted body surfaces from the CT scans used to mimic SGRT. The methods used only the first (day=0) CT scan for generating organ meshes, and organ meshes from later treatment days are used as ground truth. SGRT alignment was based on surface information alone. In contrast, for AgRT, the day=0 body surface was used to predict subsequent skeletal avatars throughout the treatment. Then, alignment was based on both the surface and skeletal information. Rigid transformations from both approaches were applied to day=0 organ meshes to predict positions on subsequent treatment days. Root-mean-squared error (RMSE) was calculated between predicted and ground truth organ positions for both approaches.
Results: AgRT showed improvement in aligning internal anatomy compared to SGRT for all organs, particularly the liver (mean RMSEAgRT = 21.13 mm vs mean RMSESGRT = 37.79 mm). Additionally, AgRT exhibited less variation over time compared to SGRT.
Conclusion: Realistic full-body modeling in SGRT can address issues caused by insufficient surface anatomic variation, which can lead to poor correlation and large errors in current techniques. Inferring skeletal anatomy enables alignment to a patient’s X-ray imaging, improving the correspondence between surface imaging and internal anatomy.