Deep-Learning-Enhanced Atlas-Based Preoperative and Intraoperative Registration for Cochlear Implant Surgery Navigation
No Thumbnail Available
Date
2025-05-22
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This dissertation provides the groundwork for intraoperative registration in cochlear implant surgery though the developed Vision6D pose annotation tool and a series of deep-learning-based methods. The primary contributions can be summarized as follows. First, self-supervised ossicles registration and segmentation, as detailed in Chapter 2. Second, the development of the Vision6D software and its comprehensive user study using two public 6D pose estimation datasets, as introduced in Chapter 3. Third, 2D monocular microscope views to 3D CT registration using the incus of the ossicles as a landmark, which is described in Chapter 4. Fourth, mastoidectomy shape prediction to extract the postmastoidectomy mesh directly from preoperative CT scans, as shown in Chapters 5, 6, and 7. Fifth, postmastoidectomy surface multi-view synthesis from a single microscope image is proposed in Chapter 8. Sixth, surgical scene completion for the synthetic postmastoidectomy surface multi-views through single-step denoising diffusion GAN, as illustrated in Chapter 9. Finally, Chapter 10 utilizes the prior contributions from Chapters 5 to 8 to perform the monocular patient-to-image intraoperative registration for cochlear implant surgery that leverages the synthetic surgical views. These combined components provide numerous opportunities for future intraoperative navigation systems and surgical applications.
Description
Keywords
Cochlear Implant, Deep Learning, Intraoperative Registration, Atlas-based Registration and Segmentation.