Hybrid Vision Transformer for Detection of Dentigerous Cysts in Dental Radiography Images

Dentigerous cysts, a prevalent form of odontogenic lesions, present a significant challenge in dental diagnostics due to their often asymptomatic nature and potential for complications if left untreated. Early and accurate detection of these cysts is crucial for optimal patient care and treatment planning. This study introduces a novel approach to automate the detection of dentigerous cysts in dental radiographic images using advanced deep learning techniques. We present a Hybrid Adaptive Vision Transformer (HA-ViT) architecture that uniquely combines convolutional neural networks (CNNs) with Vision Transformers (ViTs) to process both full panoramic and focused dental radiographs. This dual-path approach allows for comprehensive analysis of dental structures at multiple scales, addressing the varied imaging scenarios encountered in clinical practice. Our model was trained and evaluated on a large dataset of dental radiographs, labeled by a dentist. The HA-ViT architecture demonstrated an excellent performance, achieving an accuracy of 94.44%, sensitivity of 90.64%, and specificity of 96.74% in dentigerous cyst detection. The model's robustness is further evidenced by an Area Under the Receiver Operating Characteristic curve (AUC-ROC) of 0.9829. These results underscore the potential of our approach to significantly enhance the efficiency and accuracy of dentigerous cyst detection in clinical settings.