Automated Classification Radiograph of Periodontal Bone Loss Using Deep Learning.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Deep Learning for Periodontal Disease Detection

Background:

• Periodontitis is a prevalent inflammatory condition leading to tooth loss if untreated.
• Early and accurate classification of periodontal bone loss from dental radiographs is vital for effective patient management.
• Orthopantomograms (OPGs) are commonly used dental radiographs for assessing periodontal health.

Purpose of the Study:

• To evaluate and compare the performance of three deep learning architectures: InceptionV3, InceptionV4, and ResNet-50.
• To assess their efficacy in classifying OPGs based on grades of periodontal bone loss.
• To identify the most effective deep learning model for automated OPG analysis in periodontitis detection.

Main Methods:

• A comparative experimental design utilizing convolutional neural network (CNN) architectures.
• Training and evaluation of InceptionV3, InceptionV4, and ResNet-50 on OPG images.
• Implementation of image data augmentation and hyperparameter tuning (epochs, learning rates, optimizers) to optimize model performance.

Main Results:

• ResNet-50 achieved the highest accuracy (96.8%) by the 16th epoch.
• ResNet-50 outperformed InceptionV3 and InceptionV4 in precision, recall, and F1-score.
• The study utilized MATLAB and a GeForce RTX 4060 GPU for model training and evaluation.

Conclusions:

• ResNet-50 offers superior accuracy and reliability for classifying periodontal bone loss from OPGs compared to InceptionV3 and InceptionV4.
• Deep learning-based OPG classification systems show significant potential to aid dental professionals in early periodontal disease detection.
• Further improvements can be achieved through dataset expansion and advanced hyperparameter tuning.