Benign and Malignant Oral Lesion Image Classification Using Fine-Tuned Transfer Learning Techniques.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oral Pathology

Background:

• Oral lesions are common indicators of oral diseases, necessitating prompt detection for effective treatment.
• Automated detection of oral lesions using deep learning shows significant potential.

Purpose of the Study:

• To utilize deep learning-based image classification algorithms for the identification of oral lesions.
• To evaluate the performance of VGG19, DeIT, and MobileNet models in classifying oral lesions.

Main Methods:

• A dataset of oral images, categorized into benign and malignant lesions, was used for model training and evaluation.
• Three deep learning models (VGG19, DeIT, MobileNet) were employed for image classification.

Main Results:

• VGG19 and MobileNet models achieved a 100% accuracy rate in identifying oral lesions.
• The DeIT model demonstrated a high accuracy rate of 98.73%.

Conclusions:

• Deep learning algorithms are highly effective for the automated detection of oral lesions.
• The VGG19 and MobileNet models are particularly well-suited for oral lesion classification tasks, offering high accuracy and reliability.