Intraoperative Navigation with Preoperative Models: Image-to-Patient Registration for Robot-Assisted Minimally Invasive Esophagectomy
Summary
Background: RAMIE is a complex procedure for treating esophageal cancer, where maintaining anatomical orientation can be challenging due to intricate anatomy and highly zoomed-in visualization. 3D-3D image-to-patient registration, the fusion of an anatomical 3D model with the stereoscopic intraoperative video feed, might improve surgical navigation. This study investigates the feasibility of 3D-3D image-to-patient registration for RAMIE by developing a prototype registration method.
Method: A dataset of preoperative CT scans and intraoperative stereoscopic video footage of 10 RAMIE patients was created. 3D reconstructions of both were generated as point clouds via CT scan segmentation and stereo matching of the stereoscopic video frames. RANSAC, manual correspondence selection (MCS) and ICP, were tested as registration methods. Performance was measured via success rate, surface registration error (SRE), and processing time. A panel of surgeons assessed qualitative results and answered survey questions regarding image-to-patient registration.
Results: The combination of MCS and ICP performed best, achieving a success ratio of 0.86 and a SRE of 10.98 mm. Although slower (average 25.42 seconds), it provided the most accurate results. RANSAC, while faster (0.07 seconds), had the lowest success ratio (0.18) and higher error rates. In a survey, surgeons rated registration results as moderately accurate (3-4 out of 5).
Conclusion:This study demonstrates the feasibility of image-to-patient registration for RAMIE. However, challenges remain, including time-intensive registration and the disregard for the deformation of anatomical structures. Survey results indicate a clear clinical need for improved image-to-patient registration. Future work will focus on automating registration, addressing deformable anatomy, and improving suitability in clinical settings based on stakeholder input.