Detection and Segmentation of Radiolucent Lesions in the Lower Jaw on Panoramic Radiographs Using Deep Neural...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oral and Maxillofacial Radiology

Background:

• Radiolucent lesions in the mandible can be challenging to detect and segment accurately.
• Early and precise diagnosis is crucial for effective treatment planning and patient outcomes.

Purpose of the Study:

• To develop and evaluate a deep learning model, specifically You Only Look Once (YOLO) v8, for autonomous detection and segmentation of radiolucent mandibular lesions.
• To assess the model's performance using various metrics and augmentation techniques.

Main Methods:

• Analysis of 226 panoramic radiographs featuring various radiolucent lesions (e.g., cysts, ameloblastomas).
• Data augmentation techniques including translation, scaling, rotation, flipping, and mosaic effects were applied.
• A deep neural network with five-fold cross-validation was employed for detection and segmentation tasks.
• Performance was evaluated using Intersection over Union (IoU), precision, recall, and mean average precision (mAP).

Main Results:

• The augmented YOLOv8 model significantly improved detection performance, achieving 95.2% precision, 94.4% recall, and 97.5% mAP@50.
• Segmentation task results were also enhanced with augmentation, reaching 100% precision, 94.5% recall, and 96.6% mAP@50.
• Augmentation notably improved mAP@50-95 scores for both detection (68.7%) and segmentation (72.2%).

Conclusions:

• The developed YOLOv8 deep learning model demonstrates a remarkable capability for the automatic detection and segmentation of radiolucent mandibular lesions.
• The model's performance, particularly with data augmentation, indicates its potential as a valuable tool in dental diagnostics.
• This AI-driven approach has the potential to enhance patient care through more efficient and accurate lesion identification.