Examination of new clinical dental caries in school children using real intra oral photos with artificial intelligence model YOLO-V8x

Background Dental caries is one of the most common chronic diseases in school-aged children, with a prevalence above 80% in Indonesia. Traditional diagnostic practices are time-consuming and dependent on the number of healthcare professionals available, and as such, have led to the creation of AI-based alternatives, such as the HI Bogi application. This research used the YOLO-v8 model to assist in the detection of dental caries, which is faster, more efficient, and highly accurate, and can be developed to enhance existing dental health programs in schools in Indonesia Materials and Methods A dataset of 3,221 JPG images labeled using the ICDAS D0 to D1 method was prepared and processed using Roboflow labeling software, and the images were resized to 640×640 pixels to standardize the input for model training. The dataset was divided into training (2,266 images), validation (635 images), and testing (320 images) subsets. The YOLOv8x algorithm was used for deep learning and the performance of the model was evaluated using confusion matrix analysis to calculate True Positive (TP), False Positive (FP), and False Negative (FN) values. Statistical Mann–Whitney tests were conducted to compare the classification accuracy between the AI model and dentists across the ICDAS categories, whereas a diagnostic speed test assessed the efficiency of the AI model relative to dentists. Results The YOLO-v8 model showed encouraging results, with a recall of 41.1%, accuracy of 72.6%, and mAP of 45.8%. P-values of 0.301 for D1, 0.690 for D2, 0.621 for D3, 0.693 for D4, 0.634 for D5, and 0.302 for D6 were obtained from comparative testing with dentists. These results showed no significant variations in the sensitivity, specificity, PPV, or NPV among the ICDAS categories (p > 0.05). Furthermore, AI performed much faster than the dentists during the examinations (p = 0.000). AI may improve the efficiency and efficacy of early caries diagnosis as demonstrated by these findings. Conclusion The YOLO-v8 model integrated into the application showed promising results in early caries detection, comparable to those of dentists across all ICDAS criteria. AI significantly outperformed dentists in terms of examination speed and completed tasks four times faster. Future research should explore transformer-based models to improve accuracy and expand the datasets to enhance the ability of the model to identify diverse caries categories, including rare lesions.