Advanced Deep Learning Models for Classifying Dental Diseases from Panoramic Radiographs

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Deep Learning for Healthcare

Background:

• Dental diseases pose significant oral health challenges, necessitating early diagnosis.
• Panoramic radiographs offer detailed dental structure visualization, suitable for automated diagnostic systems.
• Existing datasets often suffer from class imbalance and inconsistencies, hindering accurate automated diagnosis.

Purpose of the Study:

• To investigate the efficacy of advanced deep learning models for multiclass classification of dental diseases at a sub-diagnosis level.
• To address data inconsistencies and class imbalance in panoramic radiograph datasets.
• To evaluate the performance of various convolutional neural network architectures for dental disease classification.

Main Methods:

• Utilized a dataset of 10,580 high-quality panoramic radiographs, consolidated into 35 classes.
• Applied preprocessing techniques including class consolidation, mislabeled entry correction, redundancy removal, and augmentation to mitigate class imbalance.
• Assessed five convolutional neural network (CNN) architectures: InceptionV3, EfficientNetV2, DenseNet121, ResNet50, and VGG16.

Main Results:

• InceptionV3 achieved the highest performance with 97.51% accuracy and 96.61% mean average precision (mAP).
• EfficientNetV2 and DenseNet121 also demonstrated strong classification performance with accuracies of 97.04% and 96.70%, respectively.
• ResNet50 and VGG16 provided competitive accuracy rates, highlighting the potential of multiple CNN architectures.

Conclusions:

• Deep learning models, particularly InceptionV3, are highly effective for automated dental disease classification using panoramic radiographs.
• The study provides a foundation for developing efficient and accurate automated diagnostic systems in dentistry.
• Future research should focus on dataset expansion, ensemble learning, and explainable AI for enhanced clinical utility.