PURPOSE: Kampo medicine (Japanese traditional herbal medicine) can identify "Mibyou" (disease-oriented state) based on visual tongue inspection by a trained physician. Surface colors of uncoated tongue provide essential features. A hyperspectral imaging system which performs regional image analysis of the tongue color spectrum was developed to automatically inspect the uncoated tongue. METHODS: Hyperspectral tongue surface images were acquired by a camera equipped with an integrating sphere. The color spectrum from 4 tongue areas--uncoated tongue, coated tongue, lip, and perioral areas--were quantified. The average spectrum of each area was determined and 4 images calculated by focusing on the differences between respective spectra; highlight, shadow, tongue coat, and lip-weighted images were calculated. The uncoated tongue area was extracted by subtraction. RESULTS: Reproducibility was evaluated by applying the camera and extraction algorithm to 44 images from human subjects. Eighty-two percentage were judged as acceptable by a Kampo medicine physician expert. CONCLUSIONS: Using a hyperspectral camera and extraction algorithm, the tongue color of the uncoated part was automatically extracted. This technique is suitable for tongue color analysis and may help non-trained users to identify "Mibyou".
PURPOSE: Kampo medicine (Japanese traditional herbal medicine) can identify "Mibyou" (disease-oriented state) based on visual tongue inspection by a trained physician. Surface colors of uncoated tongue provide essential features. A hyperspectral imaging system which performs regional image analysis of the tongue color spectrum was developed to automatically inspect the uncoated tongue. METHODS: Hyperspectral tongue surface images were acquired by a camera equipped with an integrating sphere. The color spectrum from 4 tongue areas--uncoated tongue, coated tongue, lip, and perioral areas--were quantified. The average spectrum of each area was determined and 4 images calculated by focusing on the differences between respective spectra; highlight, shadow, tongue coat, and lip-weighted images were calculated. The uncoated tongue area was extracted by subtraction. RESULTS: Reproducibility was evaluated by applying the camera and extraction algorithm to 44 images from human subjects. Eighty-two percentage were judged as acceptable by a Kampo medicine physician expert. CONCLUSIONS: Using a hyperspectral camera and extraction algorithm, the tongue color of the uncoated part was automatically extracted. This technique is suitable for tongue color analysis and may help non-trained users to identify "Mibyou".