Yi Sun obtained his PhD in Biomedical Sciences, Master of Medical imaging and Bachelor in Electronic Engineering. Since 2007, he worked in the field of computer assistant surgery planning, with focus on oral and maxillofacial surgery. Currently he is responsible for the 3D surgical simulation team in the department of oral and maxillofacial surgery (UZ Leuven). In the past years, he and the team members developed several computer assisted surgical applications in dental implant placement, cranio-maxillofacial reconstruction and patient specific implant design.    

He has published more than 60 articles in peer-reviewed journals and has contributed three book chapters. His research interests are 1) Computer assisted reconstruction of large bone defects in cranio-maxilofacial region; 2) Deep learning model based on biological data to design patient-specific implants.

Topic title: Automatic Designing Framework of Orbital Patient-Specific Implants via 2.5-Dimensional Autoencoder Based on Depth Feature Maps

Abstract:

Orbital fractures pose clinical challenges because they can impair vision and impact essential facial structures. Patient-Specific Implants (PSI), designed using computed tomography (CT) scanning techniques and additive manufacturing, offer a promising tool to these challenges. However, current manually or semi-automatically designing of PSIs in clinical settings have serious limitations, including poor convenience, complicated operations and low efficiency.

METHODS: Therefore, an asymmetric 2.5-dimensional autoencoder framework based on depth feature maps was proposed to overcome them. Firstly, based on the orbital geometric anatomy landmarks, a novel depth feature generation algorithm is designed to extract spatial distribution information of each orbit. Next, outputs are introduced into our 2.5-Dimensional autoencoder frameworks, to guide the network towards automatic repairment of depth maps with defects. Additionally, we propose a method that combines the inception architecture with dilated convolution to further enhance the network's performance.

RESULTS: In our study, all 80 clinical cases were employed for training, while another set of 5 cases were used exclusively for testing purposes. Our evaluation results revealed that our proposed architecture achieved an average repairing accuracy of 0.42±0.22 mm.

CONCLUSIONS: The outcomes demonstrate a notable precision in the reconstruction of orbital implants using our developed framework. This framework is distinctively engineered to accommodate a wide array of defect profiles, irrespective of their distribution, layout, or typology, thereby ensuring its broad applicability in the domain of orbital implant reconstruction. Our framework manifests considerable promise for future integration into the surgical design process of Patient-Specific Implants (PSIs), heralding a significant evolution in orbital reconstructive methodologies.