Efficacy of Various Deep Learning Models for Automated Diagnosis in Oral and Maxillofacial Lesions

Abstract

Recent advancements in deep learning (DL) have significantly enhanced automated diagnostic capabilities in oral and maxillofacial radiology. Convolutional neural networks (CNNs) and their variants, such as YOLO, U-Net, Mask R-CNN, and hybrid CNN–Transformer architectures, have demonstrated superior accuracy in detecting, segmenting, and classifying lesions in panoramic, periapical, and CBCT images. These models improve clinical workflows by enabling rapid interpretation, reducing observer variability, and ensuring consistent precision in identifying caries, cysts, and neoplastic lesions. YOLO models facilitate real-time object detection, U-Net variants deliver detailed segmentation, and Mask R-CNN allows instance-level delineation. Emerging CNN–Transformer hybrids combine contextual and spatial reasoning, leading to robust diagnostic performance. Overall, DL-based image analysis provides a reliable adjunct to clinical decision-making, advancing precision-driven dental radiology.