Principal Component Vector Rotation of the Tongue
Color Spectrum to Predict Mibyou (Disease-Oriented State)

Satoshi Yamamoto  Norimichi Tsumura  Toshiya Nakaguchi 
Takao Namiki  Yuji Kasahara  Keiko Ogawa-Ochiai 
Katsutoshi Terasawa  Yoichi Miyake

Chiba University, ,Japan

International Journal of Computer Assisted Radiology and Surgery 2011 Mar; 6(2): 209-215.

Abstract
Purpose Kampo medicine (Japanese traditionalherbal medicine) contains concepts useful forpreventive medicine. For example, Mibyou (diseaseorientedstate) aims to prevent illness by early recognition.Kampo diagnosis is based on subjective examinations, such as tongue inspection, by trained specialistphysicians. An objective metric of the tongue colorspectrum was developed as a surrogate for subjectivevisual inspection
Methods Tongue images were acquired with a hyperspectralimaging system, and the uncoated tongueregion was segmented automatically. The spectral informationof the uncoated tongue area was analyzed byprincipal component analysis (PCA). The componentvector most representative of each clinical symptom was found by rotating the vector on a plane spanned by two arbitrary principal component vectors.Results The system was tested in human volunteers.
Results The system was tested in human volunteers.Forty-four hyperspectral images were acquired from 30healthy male subjects for initial testing.The Oketsu(blood stagnation) score was determined by an experiencedclinician in Kampo medicine from 27 of 30 subjects. The correlation between respective principal components and Oketsu score was 0.67 at maximum, and increased to 0.73 by linear combination, while it was0.75 by vector rotation. Significant correlations for many disorders were demonstrated, and vector rotation showed better correlation than linear combination.Conclusions A PCA-based algorithm was developed to objectively evaluate patients using color images of the tongue surface. Testing showed that this method was a feasible surrogate for expert visual tongue analysis.
This tool should help non-trained people identify Mibyou health status for individuals. The algorithm is free of empirical criteria, and it may be it applicable tomany hyperspectral image types.

[PDF] (to appeare)


Back