Artificial Intelligence-Assisted Detection of Odontogenic Lesions on Panoramic Radiographs: Development and Validation of a Deep Learning Model for Nigerian Clinical Settings

Panoramic radiography is the most widely available dental imaging modality in Nigerian tertiary and secondary health facilities, and its systematic interpretation for early detection of odontogenic cysts and tumors is constrained by clinician workload, training variability, and limited access to specialist oral radiology review. Artificial intelligence (AI) tools, specifically deep learning-based convolutional neural networks (CNNs), have demonstrated promise in automating detection of radiographic abnormalities, but no validated AI diagnostic tool exists for odontogenic lesion detection in Nigerian or broader African clinical contexts. This PhD study develops and validates a deep learning model for automated detection of odontogenic cysts and tumors on panoramic radiographs acquired from Nigerian patients. A training dataset of 2,400 panoramic radiographs with confirmed odontogenic lesions was curated from five teaching hospitals spanning four geopolitical zones, with annotations performed by two consultant oral and maxillofacial surgeons and one oral radiologist. A ResNet-50 architecture was fine-tuned and evaluated using 5-fold cross-validation. External validation was performed on 300 radiographs from two independent institutions. The model achieved a sensitivity of 86.2% and specificity of 89.4% for odontogenic lesion detection, outperforming junior resident detection accuracy. Performance was comparable for radicular cysts, dentigerous cysts, and ameloblastomas but lower for early-stage odontogenic keratocysts. The study establishes proof-of-concept for AI-assisted oral surgical diagnosis in Nigerian radiology and recommends a prospective clinical implementation trial at resource-limited institutions. Keywords: artificial intelligence, deep learning, panoramic radiograph, odontogenic lesions, Nigeria.