Detection and Differentiation of OPMD from Photographic Images Using Transfer Learning

*Corresponding author: (e-mail: [e-mail]).

Abstract

Background: This study leverages artificial intelligence (AI) and transfer learning from existing YOLO models to enhance the detection of oral potentially malignant disorders (OPMDs) like leukoplakia and ulcers. It highlights the role of AI in improving diagnostic accuracy in oral health care.

Materials and Methods: In this study, 696 images featuring 761 leukoplakia and 386 ulcer lesions were analyzed using a transfer learning-enhanced YOLOv8 model for OPMD detection. The model’s performance was evaluated based on precision, recall, F1-scores, and a confusion matrix to assess its effectiveness in identifying these conditions.

Results: The YOLO model displayed notable precision in the detection of leukoplakia and ulcers on the test set, achieving 97.44% for leukoplakia and a perfect 100% for ulcers. Despite this high precision, the model encountered some limitations in the recall, with rates of 64.41% for leukoplakia and 58.62% for ulcers. These figures indicate a tendency of the model to miss certain cases. The F1-scores, standing at 77.55% for leukoplakia and 73.91% for ulcers, suggest a competent balance between precision and recall, though they also point to potential areas for improvement in overall detection efficacy.

Conclusion: The model demonstrates high precision in detecting key oral conditions, with notable accuracy in identifying true positives. Its performance, particularly in precision, positions it as a valuable tool in early oral disease detection despite some missed detections indicated by the recall rates.

Publication History

Article published online:
08 July 2024