Tooth-Marked Tongue Recognition Using Multiple Instance Learning and CNN Features.

Area of Science:

• Integrative Medicine
• Medical Diagnostics
• Artificial Intelligence in Healthcare

Background:

• Tooth-marked tongue diagnosis is crucial in Traditional Chinese Medicine (TCM).
• Existing automated recognition methods struggle with the diverse appearances and variations of tooth marks.
• Current approaches often rely on concavity, leading to inconsistent performance.

Purpose of the Study:

• To develop a robust automated method for tooth-marked tongue recognition.
• To overcome the limitations of existing methods that rely solely on concavity information.
• To improve the diagnostic accuracy of tooth-marked tongue analysis in TCM.

Main Methods:

• A three-stage approach was proposed for tooth-marked tongue classification.
• Stage 1: Utilized concavity information to identify suspected regions.
• Stage 2 & 3: Employed a convolutional neural network (CNN) for feature extraction and a multiple-instance classifier for final decision-making.

Main Results:

• The proposed three-stage method demonstrated effectiveness in tooth-marked tongue recognition.
• The approach addresses challenges posed by variations in tooth mark appearance and shape.
• Experimental results validated the superior performance of the developed method.

Conclusions:

• The novel three-stage approach offers a more reliable solution for automated tooth-marked tongue recognition.
• This method enhances the potential for integrating AI in TCM diagnostics.
• Further research can build upon this framework for improved tongue-based health assessments.