Tongue image quality assessment based on a deep convolutional neural network.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Traditional Chinese Medicine (TCM)

Background:

• Tongue diagnosis is crucial for modernizing TCM.
• High-quality tongue images are essential for building standard TCM diagnostic datasets.
• Automated quality control is vital for efficient and accurate tongue image analysis in TCM.

Purpose of the Study:

• To develop an automatic, efficient, and accurate model for tongue image quality assessment (IQA).
• To compare the performance of various machine learning models for TCM tongue IQA.
• To identify the optimal model for screening qualified tongue images for a standard TCM database.

Main Methods:

• Evaluated multiple machine learning models, including Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Adaptive Boosting Algorithm (Adaboost), Naïve Bayes, Decision Tree (DT), Residual Neural Network (ResNet), Visual Geometry Group (VGG) Convolutional Neural Network (CNN), and Densely Connected Convolutional Networks (DenseNet).
• Compared model performance using accuracy, precision, recall, and F1-Score metrics.
• Focused on differentiating between good-quality and poor-quality tongue images.

Main Results:

• Deep learning models, particularly CNNs, demonstrated high accuracy (over 96%).
• ResNet-152 and DenseNet-169 achieved accuracies exceeding 98%.
• ResNet-152 achieved 99.04% accuracy, 99.05% precision, 99.04% recall, and 99.05% F1-score, outperforming other evaluated models.

Conclusions:

• Deep CNNs are feasible and effective for evaluating tongue image quality.
• ResNet-152 is selected as the optimal quality-screening model for tongue IQA.
• This research supports the development of intelligent tongue diagnosis technology in TCM.