ROC-Boosting: A Feature Selection Method for Health Identification Using Tongue Image.

Area of Science:

• Biomedical imaging
• Computational pathology
• Traditional Chinese Medicine (TCM)

Background:

• Tongue diagnosis is a key component of Traditional Chinese Medicine (TCM) for health assessment.
• Objective selection of diagnostic features from tongue images remains a challenge.
• Automated analysis of tongue images can potentially enhance diagnostic accuracy.

Purpose of the Study:

• To identify significant Haar-like features from tongue images for health identification.
• To develop a novel boosting method for feature selection in the ROC space.
• To provide scientific validation for TCM tongue diagnosis.

Main Methods:

• Collected tongue images and health data from 1,322 cases, classified into health, ill (normal tongue), and ill (abnormal tongue) groups.
• Extracted Haar-like features from tongue images.
• Applied a new boosting method in the ROC space for feature selection.

Main Results:

• A total of 27 Haar-like features were initially obtained.
• Seven significant features were selected from the health and ill (normal tongue) groups.
• Twenty-five significant features were selected from the health and ill (abnormal tongue) groups.

Conclusions:

• The selected significant features were primarily located in the root, top, and side regions of the tongue.
• These findings align with traditional tongue partitions used in TCM.
• The study provides scientific evidence supporting the use of TCM tongue diagnosis for health identification.