AI-based 3D-printing strategies for patient specific implants in maxillofacial surgery

Abstract

An orbital floor fracture can be the result of a blunt trauma and may lead to enophthalmus, diplopia or even vision loss [1]. In the case of a severe injury, reconstructive surgery with titanium implants can be indicated [2]. Because of the limited intraoperative view and the complexity of the anatomical region, there is special need for a patient-specific solution to ensure precise reconstruction [3]. The field of application of patient-specific implants (PSI) has expanded constantly due to their superior ability to restore the complex anatomical conditions of the facial skull compared to standardized implants and therefore leading to a more predictable treatment outcome [4]. A major obstacle for the use of PSIs is the currently time consuming and expensive manufacturing process, which is characterized by many individual manual steps. In this project, the processes were fully automated and integrated into an end-to-end automatic digital value chain. First, an artificial intelligence (AI) based automatic segmentation of the bony skull out of CT-Data is performed using Dense U-Net [5]. Subsequently, the defect is reconstructed using a statistical shape model (SSM), which was trained by using a public Dataset [6]. The outcome of the SSM is used to create a virtual PSI design, which is once again designed by an artificial intelligence. Finally, the implant is produced via additive manufacturing (AM), followed by post-processing to obtain the required surface qualities. In summary, the exclusive input consists of medical image data that is automatically processed into an output of the physical implant. This study focuses on the created SSM. The input consists of a 3D model of the skull with an orbital floor defect and the output of a 3D model with the reconstructed orbit. The aim of this study is to determine the precision of the "Statistical Shape Model", created in the course of the project „DigiMed“ (REACT-EU), for reconstruction of the orbit using orbital volume measurement. Currently the evaluation is still in progress, therefore the final results cannot be given yet.

Author’s statement
Conflict of interest: All Authors state no conflict of interest. Acknowledgement: We thank the EU for the funding of this study (51164122).

References
[1] Winegar, B.A. and J.E. Gutierrez, Imaging of Orbital Trauma and Emergent Non-traumatic Conditions.Neuroimaging Clin N Am, 2015. 25(3): p. 439-56.
[2] Yi, W.S., et al., Reconstruction of complex orbital fracture with titanium implants. Int J Ophthalmol, 2012. 5(4): p. 488-92.
[3] Chepurnyi, Y., et al., Clinical efficacy of peek patient-specific implants in orbital reconstruction. J Oral Biol Craniofac Res, 2020. 10(2): p. 49-53.
[4] Timoshchuk, M.A., et al., Do Patient-Specific Implants Decrease Complications and Increase Orbital Volume Reconstruction Accuracy in Primary  Orbital Fracture Reconstruction? J Oral Maxillofac Surg, 2022. 80(4): p. 669-675.
[5] Kola?ík, M., et al., Optimized High Resolution 3D Dense-U-Net Network for Brain and Spine Segmentation.Applied Sciences, 2019. 9(3): p. 404.
[6] Kodym, O., et al., SkullBreak / SkullFix - Dataset for automatic cranial implant design and a benchmark for volumetric shape learning tasks. Data Brief, 2021. 35: p. 106902.