Artificial Intelligence in Oral Diagnosis: Detecting Coated Tongue with Convolutional Neural Networks

Abstract

Background/Objectives: Coated tongue is a common oral condition with notable clinical relevance, often overlooked due to its asymptomatic nature. Its presence may reflect poor oral hygiene and can serve as an early indicator of underlying systemic diseases. This study aimed to develop a robust diagnostic model utilizing convolutional neural networks and machine learning classifiers to improve the detection of coated tongue lesions. Methods: A total of 200 tongue images (100 coated and 100 healthy) were analyzed. Images were acquired using a DSLR camera (Nikon D5500 with Sigma Macro 105 mm lens, Nikon, Tokyo, Japan) under standardized daylight conditions. Following preprocessing, feature vectors were extracted using CNN architectures (VGG16, VGG19, ResNet, MobileNet, and NasNet) and classified using Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Multi-Layer Perceptron (MLP) classifiers. Performance metrics included sensitivity, specificity, accuracy, and F1 score. Results: The SVM + VGG19 hybrid model achieved the best performance among all tested configurations, with a sensitivity of 82.6%, specificity of 88.23%, accuracy of 85%, and an F1 score of 86.36%. Conclusions: The SVM + VGG19 model demonstrated high accuracy and reliability in diagnosing coated tongue lesions, highlighting its potential as an effective clinical decision support tool. Future research with larger datasets may further enhance model robustness and applicability in diverse populations.