Preoperative CT and Intraoperative CBCT Image Registration and Evaluation in Robotic Cochlear Implant Surgery

Abstract

Robotic cochlear implantation is an effective way to treat deafness and severe losses in hearing, which can reduce errors caused by human factors. It requires the pre-operative CT and intra-operative CBCT image registration to map the preoperatively computed drilling trajectory into the intra-operative space, and has extremely high requirements for registration speed and accuracy. At present, the research on the registration method is mature, while the evaluation method is not effective. The current evaluation metrics are mostly limited to the similarity, lacking of geometric information. Whereas in clinical surgery, we are more concerned with the target registration error (TRE). In this work, we complete the CT-CBCT registration by the commonly used intensity-based method and do the process with the open source tool Elastix. We do experiment on 2 cadaver head datasets with 8 screws implanted and 14 human head datasets. We calculate the centroid distance of the screws in CBCT image and registered CT image. Meanwhile, we use SIFT to extract key points in images and calculate the average Euclidean distance between corresponding points. Results show that the registration time is less than one minute. The average centroid distances of the screws in two cadaver heads are 0.19 mm and 0.12 mm, and the average Euclidean distances of the key points in two cadaver heads are 0.196 mm and 0.239 mm. TRE of all 16 datasets are within one voxel. The TRE calculated by SIFT key points is very close to the result obtained from implanted screws. We can use SIFT feature extraction method to evaluate the registration accuracy instead of implanting screws into the patient’s head during pre-operation period, which will greatly simplify surgical procedure and avoid unnecessary injury.