Automatic Detection and Classification of Dental Anomalies and Tooth Types Using Transformer-Based Yolo With GA Optimization

Dental pathology detection and tooth type categorization are crucial yet often error-prone and time-consuming tasks in dental diagnostics. This paper presents a novel approach using Transformer-Based You Only Look Once (YOLO) for automating dental pathology detection and tooth type categorization from panoramic X-ray images. Manual dental examinations are error-prone and time consuming. Therefore automated systems are required. Preprocessing techniques such as image resizing, rotation, Gaussian smoothing, and histogram scaling are used to enhance image quality and analysis. A significant enhancement is the integration of a multi-head self-attention block into the Cross Stage Partial Darknet (CSPDarknet) backbone within the YOLOv4 architecture. This addition enabled the model to capture comprehensive contextual information and learn distinct feature representations from the dental images. Inspired by Transformer models, the multi-head self-attention mechanism improves the capacity of the model to intricate spatial relationships in images, enhancing detection and classification performance. In addition, the system employs Genetic Algorithm (GA) optimization for hyperparameter tuning, which fine-tunes model parameters to maximize performance metrics and enhance accuracy and robustness in dental diagnostics. The system achieved an average precision, recall rate of 99.5% and F1 score of 99.41% and a detection and classification accuracy of 99.31%. These results highlight the effectiveness of advanced preprocessing, deep learning enhancements, and GA optimization for precise dental pathology detection and tooth type categorization using panoramic radiography.