A novel approach based on computerized image analysis for traditional Chinese medical diagnosis of the tongue

https://doi.org/10.1016/S0169-2607(99)00031-0Get rights and content

Abstract

This research is aimed at building a computerized tongue examination system (CTES) based on computerized image analysis for the purpose of quantizing the tongue properties in traditional Chinese medical diagnosis. The chromatic algorithm is developed to identify the colors of the tongue and the thickness of its coating. The textural algorithm is used to detect the grimy coating. CTES is shown to be significantly consistent within itself with P>0.05 using the Hotelling multivariate statistical test. The overall rate of correctness for CTES to identify the colors of tongue, verifying the thickness of its coating and detecting of any grimy coating exceeds 86%. Therefore, the CTES is helpful to provide the physicians a systematic and objective diagnostic standard for the tongue diagnosis in the clinical practice and research.

Introduction

The principles of traditional Chinese medical diagnosis are based on the information obtained from four diagnostic processes, which are inspection, listening and smelling, inquiry and palpation [1]. The inspection process is to examine the patient by observing his or her shape, expression, tongue [2], etc. The listening and smelling process is to collect information for diagnosis by smelling the odor and listening to the voice of the patient. The inquiry process is to diagnose through conversation with the patient. The palpation process is to examine the pulses of the patient’s wrist at some important locations [3]. After completing these diagnostic processes, the health conditions of the patient are summarized, the possible causes of diseases can be concluded and the treatment is then implemented.

Among these diagnostic processes mentioned above, the examination of tongue is one of the most important approaches for getting significant evidences in diagnosing the patient’s health conditions. In western civilization and in Chinese culture, people seeking health care expect their tongue to be routinely examined [4]. However, the clinical competence of tongue diagnosis was determined by the experience and knowledge of the physicians who adopted the tongue diagnosis. Also, environmental factors such as the difference of light sources and brightness had great influence on the physicians in making good diagnostic results through the tongue. Moreover, most of the precious experiences in traditional tongue diagnosis could not be retained scientifically and quantitatively. Therefore, it is necessary to build an objective diagnostic standard for tongue diagnosis. The impetus of this research is to design a computerized tongue examination system (CTES) based on some image processing techniques for the purposes of providing a systematic and quantitative diagnostic standard and lowering the diagnostic discrepancy of tongue diagnosis among physicians. In addition, the experiences of tongue diagnosis can be retained by this knowledge-based expert system and the tongue images can also be retained by computer for future reference.

The color image analysis is currently being used in several medical applications. For instance, Mattsson et al. [5] applied the computer image analysis in oral pathology to observe and analyze the lichenoid reactions. Takeichi and Sato [6] used the computer-assisted image analyses to perform on the color of the tongues of 95 medical students to enhance the accuracy and objectivity of tongue inspections for determining blood stasis. They used the slide scanner to digitize the color slides of tongues and the digital information was transmitted to a personal computer for sequent feature extraction and analysis. Caprioli et al. [7] used computerized analysis of stereoscopic video images in ophthalmology to examine the reproducibility of optic measurements. In tissue analysis, Sun and Horng [8] presented an automatic computer system based on color morphometry for objectively and quantitatively assessing liver fibrosis in each liver section. Experimental results demonstrated that their system was highly reliable for the severity of liver fibrosis. The clinical competence should be built on repetitive practice of skills, together with intellect, wisdom and above all, curiosity [9]. Any abnormalities should arouse this curiosity but small changes need to be observed more carefully. Establishing a computerized analysis would enable more objective scientific analyses of physiological parameters than that of visual observation.

In traditional Chinese medicine, tongue diagnosis is usually based on the capacity of the eye for detailed discrimination. Therefore, it depends on the subjective analysis of the examiners. The first exploratory experiment for the quantitative analysis of color identification of the tongue was carried out in 1985 by the Anhui Traditional Chinese Medical College and the University of Science and Technology of China [10]. In their work, tongue images were captured from some tongue pictures by a CCD (charge-coupled device) camera and an RGB (redness, greenness, blueness) model was used to represent the colors in the images. They compared tongue images from diverse categories by means of various statistics of RGB values, such as mean, minimum, maximum and standard deviation. Their results indicated that tongue images from different categories had different statistics of RGB values. Thus, they had shown the potential of quantitative analysis of tongue diagnosis based on image processing techniques. In 1987, the Peking Traditional Medical Research Center carried out an experiment which was aimed at quantitative analysis for tongue diagnosis based on chromatic properties of tongues [11]. They chose 150 patients who were suffering from different categories of disease such as liver ailment, kidney and digestive illness. Tongue images of these patients were pictured on slides. Each slide was further transformed to three images by projecting them through the red, green and blue light sources. The projected images were then captured by a CCD camera and quantized to digital images by computer. By comparing the RGB values, one could decide the 3-dimensional regions corresponding to these diseases in the RGB feature space. Therefore, the colors of the pixels in tongue images for testing were properly recognized and assigned to different categories by means of deciding which region each set of RGB values of tongue image fell into. Note that the identified colors were represented in specific codes (either characters or numbers). The results of color identification of chosen pixels were claimed to be reasonably consistent with the diagnostic results obtained by physicians. Moreover, the identified colors were also used for the verification of thickness of coating. By comparing the RGB values of the pixels, the detection of grimy coating could be done. The results were acceptable only for certain specific categories of tongue images. One can understand easily that the chromatic properties of tongue images could not reveal too much information of the coating. It requires more features based on other properties to obtain better quantitative analysis. Based on these two preliminary exploratory experiments, the potential of color identification of the tongue by means of quantitatively chromatic approaches were demonstrated. In 1990, the China government released a color comparison system for tongue diagnosis [10]. This system was composed of standardized color-boards which were developed by physicians and scientific researchers. By comparing the colors appearing in the tongue of the patient, the physicians could know easily the exact colors of the patient’s tongue. However, this approach is somehow tedious for tongue diagnosis and presents difficulty in recording certain tongue properties quantitatively for future reference.

Recently, C.C. Chiu et al. [12] proposed a structural texture recognition algorithm which adopted the RGB model for mapping the tongue colors to some known categories and used certain features such as spatial gray-tone dependency matrices (SGTDM) [13] and Fourier power spectrum (FPS) [14], [15] to verify or identify certain properties of coating on the tongue. The results obtained in their experiments were encouraging for building a standardized CTES. This research, which is an extension of Chiu’s research, is focused on developing algorithms for quantizing the chromatic and textural properties of tongue. The following section will present the general outline of the newly developed CTES.

Before introducing the general outline of CTES, it is necessary to understand the major factors relating to traditional tongue diagnosis. The conceptual processes of tongue diagnosis are shown in Fig. 1. Generally, tongue diagnosis can be obtained by means of observing the substance and coating of the tongue [1], [2]. The substance of the tongue can be characterized by its color, luster, shape and movement. The coating of tongue can be distinguished by its color and certain textural properties as listed in Fig. 1. Note that the shaded blocks in the figure are major foci for CTES.

In this paper, a standardized hardware implementation of a computerized tongue diagnosis system is presented. Its goal is to reduce the environmental effects in order to improve the diagnostic results. Also, the structural recognition approaches based on chromatic and spatial textural properties of tongue are developed for collecting certain information regarding the substance and coating of tongue, such as colors of the substance and coating and all the properties of the coating except for moisture and dryness. Note that the properties of moisture and dryness on the tongue can be easily identified by human beings but still present a challenge for the machine.

The hardware implementation is composed of a color CCD, a standardized light source with day-light quality, a head supporting mechanism for sampling tongue images and a computer. The choice of the color CCD and light source is based on the high fidelity of color requirements for tongue images. The self-designed head supporting mechanism is to stabilize the tongue image of a patient and therefore improve the image quality by fine tuning of the mechanism under the controlled standard light source and focus of lens.

The software implementation of the CTES is composed of several modules, which are a structural recognition approach module, an image database management module, a tongue diagnosis rulebase and the reasoning module. The structural recognition approaches shown in Fig. 2 can be divided into two parts which are based on two important properties of tongue images, i.e. chromatic and spatial textural properties. Algorithms for determining the chromatic-related properties of the tongue are based on the HSL (hue, saturation, luminance) color model since the way it represents the colors is very suitable for describing the colors found in the tongue especially when the substance is covered by a coating. Based on the HSL model, it can be easier to define a feature for thickness measurement of the coating. From the textural point of view, the grimy coating can be described by certain features of SGTDM by using the monochrome image. More details of these algorithms will be presented in the following sections.

The paper is organized as follows: in Section 2, the algorithms based on chromatic properties for color classification of the tongue and thickness verification of its coating are discussed. In Section 3, the algorithm based on spatial textural properties for the detection of grimy coating is presented. In Section 4, the implementation of this system including hardware implementation and software development is described. Also, certain experimental results are illustrated and discussed. Finally, the conclusions are summarized in Section 5.

Section snippets

The color relaxation method

Color plays an important role in tongue diagnosis. It not only reveals the general health conditions of a patient, but also provides necessary information for computerized tongue diagnosis [1], [2]. For the CTES, HSL color model [16] is chosed to represent the color of tongue. HSL values are expressed in the range of (0, 100) in this paper. The decision boundaries for color space in CTES are listed in Table 1 and graphical representation of these decision boundaries is shown in Fig. 3. The

Algorithm for determining the textural-related property of tongue

In this section, the formulae of some SGTDM-based textural features such as angular second moment (ASM), contrast, correlation, variance and entropy are introduced. These features are employed to determine the grimy coating of a tongue. As the diagrams of our proposed recognition approaches shown in Fig. 2, the monochrome image with 256 graylevels is used to calculate the texture features. The spatial gray-tone dependency matrices are based on the estimation of the co-occurrence probability p(i

Implementation issues

As discussed in Section 1, the correctness of tongue diagnosis is easily affected by environmental factors, such as light source and brightness. Therefore, it is essential to set up a stable and consistent sampling environment for the purpose of providing an objective basis for tongue diagnosis. The components include light source, color CCD camera, frame grabber and computer. Note that each component of the hardware has its own properties even compared to the same specification, i.e. the

Conclusions

In this research, a novel approach based on chromatic and spatial textural properties for traditional Chinese medical diagnosis of the tongue has been presented. The general structure of the computerized tongue diagnosis system had also been implemented. Based on this system, the effect of environmental factors on tongue diagnosis has been diminished by the well-controlled hardware. Several important tongue properties such as colors of substance and coating, thickness of coating and the

Acknowledgements

The author would like to thank the National Science Council, Taiwan, ROC, for partially supporting this work under the grant number NSC-88-2213-E-035-033. Additionally, the author also wish to thank Professor Yung-Hsien Chang at the China Medical College Hospital, Taichung, Taiwan, ROC, for his counsel in regard to the work presented herein.

References (16)

  • T.T. Deng

    Basic theory of traditional Chinese medicine

  • C.H. Horng

    The principles and methods of tongue diagnosis

  • S.L. Huang et al.

    The fundamentals of palpation

    The Study of Palpation in Chinese Medicine

    (1995)
  • D.W. Beaven et al.

    Examination of the mouth and tongue

    A Colour Atlas of the Tongue in Clinical Diagnosis

    (1988)
  • U. Mattsson et al.

    Computer analysis in oral linchenoid reactions

    Acta Odontol. Scand.

    (1994)
  • M. Takeichi et al.

    Computerized color analysis of ‘Xue Yu’ (blood stasis) in the sublingual vein using a new technology

    Am. J. Chinese Med.

    (1997)
  • J. Caprioli et al.

    Reproducibility of optic disc measurements with computerized analysis of stereoscopic video images

    Arch. Ophthalmol.

    (1986)
  • Y.N. Sun et al.

    Assessing liver tissue fibrosis with an automatic computer morphometry system

    IEEE Eng. Med. Biol. Magazine

    (1997)
There are more references available in the full text version of this article.

Cited by (161)

  • A multi-step approach for tongue image classification in patients with diabetes

    2022, Computers in Biology and Medicine
    Citation Excerpt :

    Tongue image features usually include color, texture, tongue coating proportions, and geometric features. Chuang-Chien Chiu established a computerized tongue examination system (CTES) to quantify the tongue color, the thickness of the tongue, the texture of the tongue, etc [40]. Yonggang Wang et al. introduced a computer-aider tongue examination system to quantify tongue color [41].

View all citing articles on Scopus
View full text