Quantitative Assessment of Third Molar Extraction Difficulty and Nerve Injury Risk Using Artificial Intelligence and...

Tooth pattern and surrounding tissue can affect the complexity of third molar (3M) extraction surgery. This study aimed to develop a new, more precise 3M extraction difficulty assessment scoring system by applying image computational techniques and artificial intelligence (AI)-based technologies to panoramic radiography images.

We used dental panoramic X-ray data of patients aged 16 to 86 years. The mandibular canal (MC), inner area of alveolar bone (IAAB), and 3M were detected and segmented using an AI-based deep learning model. We developed algorithms to evaluate the difficulty of 3M extraction and the risk of inferior alveolar nerve damage. The inclination of the 3M (1-4 points), impaction depth of the 3M (1-4 points), and proximity between the mandibular third molar (M3M) and MC (0-4) were classified and scored. The scores obtained from each algorithm were summed to classify the difficulty index as very easy (2-4), easy (5-6), slightly difficult (7-9), and very difficult (10-12).

The average precisions of the maxillary 3M and M3M detection models were 0.93 and 0.97, respectively. Moreover, the average accuracies of the 3M inclination classification algorithm, 3M impaction depth calculation algorithm, and the proximity of M3M and MC algorithms were 0.8544, 0.9515, and 0.8991, respectively.

The developed extraction difficulty scoring model can aid dentists in assessing the risk of 3M extraction and establish an appropriate surgical plan. Future studies should include more diverse patient data and improve the performance of the AI models to realize clinical implementation.