Prediction of extraction difficulty for impacted maxillary third molars with deep learning approach.

Area of Science:

• Dentistry
• Oral and Maxillofacial Surgery
• Artificial Intelligence in Medicine

Background:

• Surgical difficulty assessment for impacted maxillary third molars is crucial for patient outcomes.
• Current methods rely on subjective interpretation of radiographic features.
• Predictive tools can enhance pre-operative planning and reduce surgical risks.

Purpose of the Study:

• To evaluate the efficacy of a deep learning (DL) model in predicting surgical difficulty.
• To utilize panoramic radiographs for pre-operative assessment of impacted maxillary third molars.
• To develop an AI-driven decision support system for oral surgeons.

Main Methods:

• A dataset of 708 panoramic radiographs was analyzed.
• A deep learning model (YoloV5x architecture) was employed for segmentation and classification.
• Surgical difficulty was scored based on depth, angulation, and anatomical relations (sinus, ramus).
• Data was divided into training (80%), validation (10%), and testing (10%) sets.

Main Results:

• The DL model achieved high performance in predicting impacted maxillary third molar segmentation.
• Sensitivity, precision, and F1 scores reached 0.9705, 0.9428, and 0.9565, respectively.
• Specific anatomical relation models showed varying degrees of success, with the S-model scoring 0.8974, 0.6194, and 0.7329.

Conclusions:

• The developed DL model demonstrates effectiveness in predicting surgical difficulty for impacted maxillary third molars.
• This AI approach can serve as a valuable decision support tool for clinicians.
• Utilizing panoramic radiographs with DL can improve peri-operative decision-making in oral surgery.