The facial contour plays a crucial role in determining the smoothness and overall harmony of the face. Positioned at the center of the face, the zygomatic complex holds a prominent role in shaping the bony contour of the midface. Its form and prominence significantly impact the appearance of not only the midface but also the entire face. In regions such as Asia, facial features often exhibit a flatter profile, and pronounced cheekbones can create a discordant aesthetic with the preference for softer and more harmonious facial contours. Consequently, there has been a growing interest and demand for cheekbone reduction surgery in Asian countries (Hwang and Lee, 2019; Park, 2020).

The “L-shaped” zygomatic osteotomy stands out as a highly effective method for diminishing the prominence of the zygomatic body, aligning with the aesthetic preferences of patients. This surgical procedure currently stands as the mainstream approach for achieving this objective. For a visual representation, please refer to the osteotomy diagram provided for illustration (Fig. 1).

The zygomatic arch osteotomy surgery frequently employs either an intraoral or preauricular incision approach, aiming to minimize scarring (Chen et al., 2011; Mu, 2010). Nevertheless, these approaches are associated with small openings and unfavorable angles, which restrict the surgeon's field of view and pose challenges in precisely identifying the osteotomy line and crucial anatomical structures. Consequently, the postoperative complication rate for zygomatic arch osteotomy tends to be elevated, encompassing concerns such as vascular and nerve damage, bilateral asymmetry following osteotomy, postoperative infections, nonunion of bones, and complications in achieving proper bite alignment (Myung et al., 2017).

The rapid evolution of digital technology in computers has provided robust technical support for addressing the aforementioned issues. This encompasses 3D reconstruction technology, computer-aided design, computer-aided manufacturing, computer-aided surgical simulation, and computer-aided navigation systems.

By harnessing digital computer technology, the completion of preoperative osteotomy design, simulation evaluation, and the production of osteotomy guides based on preoperative design results can significantly guide intraoperative procedures. This approach helps in reducing surgical errors arising from blind spots, minimizing postoperative complications, and facilitating the development of personalized osteotomy plans to meet individual patient needs. This, in turn, markedly diminishes communication barriers between medical staff and patients preoperatively (Tao et al., 2021; Shi et al., 2022; Singh et al., 2022).

Nevertheless, the design of osteotomy necessitates the consideration of postoperative symmetry, safety, effectiveness, aesthetics, and other indicators before determining the osteotomy plane. Taking the osteotomy design method of the author's research group as an example, doctors are required to switch between Mimics and 3-Matic software, transfer data, and perform a sequence of operations. These include neural tube labeling, determining feature points for the osteotomy surface, making adjustments, virtual osteotomy, and cutting and trimming.

The current situation necessitates the utilization of various software, involving numerous steps and an extended, time-consuming process, thereby significantly depleting the time and energy resources of clinical doctors. Furthermore, the protracted learning curve associated with bone cutting design by physicians adds to the complexity. Additionally, the design outcomes are susceptible to the aesthetic preferences of the bone cutting operator. All these factors collectively impede the widespread adoption and implementation of digital preoperative simulation technology within the medical field. However, with the advancing landscape of machine learning and artificial intelligence technologies, novel and viable solutions have emerged to address these challenges, paving the way for the effective application of digital medical simulation technology in the medical domain.

Artificial Intelligence (AI), in a broad sense, refers to all technologies facilitating machines to replicate human thinking and capabilities for problem-solving. Over recent years, several AI deep learning frameworks, including GoogLeNet, VGGNet, and ResNet, have surfaced. Within the realm of medical research, these deep learning frameworks have the potential to substitute for doctors in executing mechanical tasks. Computers, being more attuned to grayscale and color disparities in images than the human eye, yield heightened accuracy in diagnoses. Consequently, artificial intelligence finds extensive application in image-assisted diagnosis.

At present, AI technology predominantly centers around the analysis of two-dimensional medical images, while research pertaining to three-dimensional image reconstruction remains significantly constrained. Additionally, the underlying logic of the aforementioned artificial intelligence relies on the decision-making principles of big data, essentially operating on a binary logic of either “yes” or “no” (Mitsala et al., 2021; Nabizadeh et al., 2022). Consequently, this binary logic is employed in the development of diagnostic software for diseases. AI algorithms grounded in three-dimensional data primarily leverage the PointNet open-source deep learning framework, a system that has been extensively trained at Stanford University (Qi et al., 2017).

The AI-based segmentation system utilizing 3D point cloud data models has been initially employed in military recognition and traffic management (He et al., 2024). However, the AI recognition and segmentation system, which relies on point cloud models, demand a substantial amount of computing power and pose relatively high requirements on computer configurations. Standard computers commonly found in hospitals may undergo a reduction in processing speed owing to the extensive data they need to handle, thereby falling short of meeting practical requirements. Although we have devised an AI mandibular osteotomy software, the routine computers within the department prove incapable of running the AI osteotomy software due to an excessive load issue (Qiu et al., 2022). Therefore, the creation of AI recognition and segmentation software, grounded in three-dimensional image data for surgical planning, presents a novel challenge by concurrently minimizing the computational workload. Conventional 3D data segmentation algorithms, including “RANSAC (random sample consensus)" and “NEPN (nonequivalent point network)," prove intricate to implement in typical medical scenarios (Zheng et al., 2019; Chen et al., 1999). The iterative implementation of a Point-Cloud segmentation algorithm introduces a fresh technical challenge, marking the commencement of the practical integration of artificial intelligence within the realm of three-dimensional medical data.

Taking into account the aforementioned considerations, our research team aims to investigate a novel AI cranial osteotomy software employing deep learning technology. The objective of this software is to offer an artificial intelligence solution for the preoperative design of cranial osteotomy, with the capability to supplant manual precise bone cutting design. This innovation aims to significantly decrease preoperative design time while enhancing the symmetry and safety of craniofacial contour surgery. Furthermore, the software has relatively low computational requirements, enabling it to operate seamlessly on standard clinic computers and deliver instantaneous design results.