Evaluation of yolov8 object detection algorithm for implant identification

• Published in: International dental journal Vol. 74; pp. S49 - S50
• Main Authors: Uğur, Tarık Ali, Taş, Ayşe, Yıldırımyan, Nelli, Delilbaşı, Çağrı
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01-10-2024
• ISSN: 0020-6539
• DOI: 10.1016/j.identj.2024.07.721

The objective of this study is to evaluate the effectiveness of a modern object detection algorithm in precisely identifying dental implants.
The dataset consisted of 3673 implant fixture images in a total of 763 panoramic images from two different dental school hospitals. The entire dataset is divided into three segments: 80% designated for training, 10% for validation, and the remaining 10% for testing. YOLOv8, one of the latest object detection algorithms, was used to identify the implants. The training of the algorithm was stopped when no better results were obtained and the trained weights were used to evaluate the performance of the algorithm using test dataset. Intersection over Union (IoU) was set to 0.5, and precision, recall, mean average precision (mAP), and F1 score were utilized as performance metrics.
The overall precision, recall, mAP and F1 scores were 0.777, 0.842, 0.834, and 0.78, respectively. The metrics obtained from implants with sparse representation in the database and comparable macroscopic characteristics yield low values. Conversely, implants showcasing distinct macroscopic features yield high results, even with a limited sample size.
The findings of this study underscore the proficiency of YOLOv8, an object detection methodology, in accurately identifying dental implants within panoramic radiographs, despite the utilization of a limited dataset encompassing various implant types. Furthermore, it emphasizes the critical role of dataset representation in achieving robust performance.