An In vitro study on the application of artificial intelligence in trephination pathway planning for endodontic...

This study aims to develop an artificial intelligence (AI) method using Vision-Language Models (VLMs) to determine the apicoectomy site and plan the trephination parameters-including pathway and depth-for digital guide-assisted endodontic microsurgery. Furthermore, it evaluates the feasibility of applying this method in the design of surgical guide templates, thereby providing a reference for the future development of automated and intelligent guide systems.

AI technology based on VLMs was employed to assist in the design of sleeveless guided systems by determining the precise site for apicoectomy and planning the trephination pathway and depth. Twelve teeth per group (total 24) were included. A single experienced clinician performed the simulated surgeries, blinded to the guide type. This study evaluated an AI-assisted system for endodontic microsurgery through a two-part experiment. First, virtual planning accuracy and efficiency were compared between AI and manual methods. Second, guided surgeries were performed on models using sleeveless guides derived from both plans. Data were analyzed using independent-samples t tests to compare the manual-designed and AI-designed surgical guides.

The AI-based method demonstrated superior accuracy and efficiency in virtual planning. The mean absolute deviation from the ideal resection length (3 mm) was significantly smaller in the AI group (0.031 ± 0.030 mm) than in the manual group (0.341 ± 0.234 mm) (p < 0.001). Likewise, the mean absolute deviation from the ideal resection angle (90°) was 4.06 ± 2.27° in the AI group, compared with 9.95 ± 4.24° in the manual group (p < 0.001). The average planning time was also significantly reduced in the AI group (98.48 ± 33.91 s) relative to the manual group (157.8 ± 22.45 s, p < 0.001). These accuracy benefits were confirmed in the in vitro model: resections performed with AI‑designed guides yielded mean absolute deviations of 0.193 ± 0.090 mm for length and 5.94 ± 2.93° for angle, whereas manual guides resulted in 0.473 ± 0.178 mm and 11.01 ± 3.45°, respectively (both p < 0.001).

Compared with manual design, the incorporation of AI-assisted techniques in the design of endodontic microsurgical guides significantly enhances both accuracy and efficiency.

This study addresses key challenges in surgical guide design for endodontic microsurgery. By integrating AI, it reduces technical barriers and shortens the preoperative workflow, enabling clinicians without extensive digital training to fabricate high-quality guides efficiently. This promotes broader adoption of targeted endodontic microsurgery and supports the preservation of natural teeth.