(PR84) Periapical Lesion Segmentation with YOLOV5 Architecture on Dental Periapical Images

Objective: The objective of this study was to use a YOLOV5 architecture based on an AI-supported convolutional neural network (CNN) model to segment the periapical lesion on the dental periapical radiographs. Materials and

Methods: In the study, 375 periapical lesions obtained from 284 dental periapical images were used to segmentation of the lesions. All labeled images were separated into training (80%), validation (10%), and testing (10%). The AI model was trained using the YOLOv5x architecture with 500 epochs and a learning rate of 0.01. A confusion matrix was used in the statistical evaluation of the results.

Results: The accuracy of the periapical lesion segmentation model was 0.8648, the precision was 0.8888, the sensitivity was 0.9696, and the F1 score value was 0.9275. mAP value at 0.5 Inter-over Union (IoU) value was found as 0.7192.

Conclusion: The performance of YOLOV5 architecture in the segmentation of periapical lesions on dental periapical radiographs has the potential for clinical use to assist dentists. Future studies should be made on an expanded number of images.