Artificial intelligence-assisted three-dimensional reconstruction in thoracic surgery: a narrative review

Background and Objective: The increasing emphasis on more precise surgical procedures, such as segmentectomy, demands greater expertise in surgical and anatomical knowledge. Accurate localization of lesions and identification of smaller pulmonary structures, such as segmental arteries and veins, remains a significant challenge in thoracic surgery. Artificial intelligence-assisted three-dimensional reconstruction (AI-3DR) presents a potential solution to overcome these challenges. The purpose of this review is to examine and discuss the current progress and status of AI-3DR technology in thoracic surgery. Methods: A comprehensive literature search of the PubMed, Cochrane Library, and Embase databases was conducted using keywords related to AI-assisted 3D reconstruction. The final search was completed on September 30, 2024. Key Content and Findings: AI-3DR has shown promising results in improving both preoperative planning and intraoperative precision. Studies have demonstrated that AI-3DR provides superior accuracy in localizing pulmonary lesions and classifying pulmonary structures, such as segmental bronchi, arteries, and veins, compared to traditional two-dimensional (2D) computed tomography images. In several cases, the use of AI-3DR led to a change in surgical approach, such as converting planned lobectomies to segmentectomies, thereby preserving more lung tissue. Moreover, real-time deformable augmented reality imaging supported by AI-3DR has proven valuable in enhancing intraoperative decision-making. However, despite these advances, limitations remain, including the small datasets used to train AI systems, which can limit generalizability, and occasional misclassification of small pulmonary structures. Additionally, the AI-3DR technology is still in experimental phases and have not yet been widely adopted for clinical use. Conclusions: AI-3DR technology holds great promise for improving the precision and outcomes of thoracic surgeries. Although further refinement is needed to enhance its generalizability and robustness, the technology is well-positioned to become a valuable tool in clinical practice as it continues to evolve.