3D-segmentation in modeling of bone defect implants

Abstract

In recent years, methods for efficient and rapid prosthetics of bone defects and injuries have advanced significantly, largely due to the use of computer modeling and 3D printing of implants. Unfortunately, the demand for such implants continues to grow, especially in Ukraine, where the number of patients with bone defects has reached record levels due to ongoing military actions. Currently, the most common method for creating cranial bone implants involves modeling and designing a mesh titanium cranioplasty plate using specialized software based on 3D scans of the patient. This process still requires the involvement of professional engineers to create multiple prototypes and the final implant model. To simplify, accelerate, and automate this process, it is necessary to develop a data preparation system for 3D printing that includes intelligent analysis of the patient’s 3D scans. Such a system must necessarily include a 3D image segmentation procedure – dividing the image into related pixel groups by assigning labels and grouping them accordingly. This step is especially important when processing human anatomy, as different bones have fundamentally distinct structural and functional characteristics that must be taken into account during implant modeling. In the case of 3D images, the process deals not with pixels but with voxels — three-dimensional spatial units. For modeling bone substitutes, existing datasets or 3D atlases of the human skeleton are often used as input data. When working with data from real patients, medical confidentiality must be strictly observed. Therefore, if existing datasets are not used, the data must be collected with the patients’ written consent, and all metadata must be removed before inclusion in the dataset. This paper examines methods for treating bone defects and the main techniques of bone plastic surgery, comparing their advantages and disadvantages. The process of alloplastic transplantation is described, along with the current applications of 3D modeling in this field. An analysis of segmentation methods is presented, demonstrating that the use of neural networks appears to be a promising approach, as it can simplify the modeling of bone defect implants based on insights from previous studies. The paper also highlights the potential challenges that may arise when applying this methodology.