AUTOMATED DETECTION OF ORAL CANCER IN RADIOGRAPHIC IMAGES USING YOLOV8

Abstract

Oral Squamous Cell Carcinoma (OSCC) early diagnosis is crucial in improving patient survival. The traditional diagnostic methods such as visual inspection and histopathology are time-consuming, subjective and rely heavily on the expertise of the clinician and may result in delayed diagnosis and treatment, thereby decreasing the success of intervention. One potential solution is the use of artificial intelligence (AI), especially deep learning, which can detect lesions quickly, non-invasively, and accurately. But there has been little research on deep learning–based oral cancer detection in African countries. The aim of this research is to create a deep learning model for automatic detection of oral lesions using YOLOv8 algorithm from radiographic (X-ray) images. Images of 613 annotated images across six classes (normal, cancer-inner, cancer-outer, herpes-inner, herpes-outer, and ulcer) were obtained from Roboflow and then split into a training, validation, and test set of 70%, 15%, and 15%, respectively, with images resized to 640×640 pixels. The accuracy, precision, recall, F1-score, mAP@50, and mAP@50–95 were used to assess the model's performance. The model achieved a mAP score of 0.84 and 0.56 in mAP@50 and mAP@50-95 respectively, and the overall precision, recall and f1 score were 0.88, 0.75 and 0.81 respectively, best performance in cancer classes. Results show that YOLOv8 can be used to complete the early oral cancer screening in resource-limited environments.