A visualized AI-assisted framework for pre-surgical evaluation of implant pathways on periapical radiographs

Implant pathway assessment is a crucial part of pre-operative planning before performing implant surgery at the site of a missing tooth. Dentists frequently use the periapical radiograph (PA) as a tool for preliminary evaluation of the implant pathway. However, the manual interpretation process is time-consuming and inherently subjective, making it susceptible to variations in PA-based implant pathway judgment due to the operator's experience. This study aims to establish an automated and objective framework that integrates an algorithm and deep learning to analyze the implant pathway preliminarily.

We analyzed 1,200 PAs containing missing teeth and reserved 120 PAs for testing. We first used segmentation model to identify the teeth adjacent to the missing tooth site. This segmentation served as the foundation for the GeoPath algorithm for visualized implant pathway. The results were compared with actual preliminary clinical assessment outcomes to evaluate feasibility.

The segmentation model achieved a precision of 96.73 %, a sensitivity of 97.12 %, and a specificity of 99.27 %, and obtained a Cohen's κ score of 0.93, which confirmed an agreement between the automated auxiliary detection results and the dentist's ground truth annotations. Furthermore, the angular error was only 2.79°, which confirmed a lower average root mean square error (RMSE) of 0.10433. This effectively reduced the time required for dentists' clinical PA assessment by 87.19 %.

The automated implant pathway assessment framework for PAs can serve as an auxiliary tool for senior dentists during initial clinical diagnosis and as a method for training new dentists.