Tongue diagnosis is a standard expert technique of traditional Chinese medicine (TCM). This process may be automated through computerization yet the image capturing and tongue images segmentation upon which this depends is made difficult by the fact that it is sometimes difficult to display the tongue outside the mouth in a consistent manner or proportion or according to any recognised criteria. Further, the tongue is non-rigid and varies greatly in size, shape, color, and texture. Variations in the tongue include ravines and patches on the surface of tongue which can affect edge detection, and according to TCM, disease can cause variations in the color, plumpness, or slenderness of the tongue.

Most recent work has sought to capture images of the tongue using CCD cameras (Pang and Zhang, 2004; Zuo et al., 2004; Sun et al., 2003). However, this approach produces acceptable images only if they are acquired under highly controlled conditions, especially with regard to illumination and pose.

The spectroscope is a tool that may release us from these constraints. Spectral measurements of human tissue have been used for many years for characterizing and monitoring applications in biomedicine. In remote sensing, researchers have shown that hyperspectral data are effective for material identification in scenes where other sensing modalities are ineffective (Healey and Slater, 1999). The spectral properties of tissue are determined by the interaction of light with human tissue (Tuchin, 2000; Anderson and Parrish, 1981). The epidermal and dermal layers of human skin constitute a scattering medium that contains several pigments such as melanin, hemoglobin, bilirubin and β-carotene. Small changes in the distribution of these pigments led to obvious changes in the skin's spectral reflectance (Edwards and Duntley, 1939). Recent research has measured skin reflectance spectra over the visible wavelengths and proposed models for the spectra. Other researchers have used a skin reflectance model at the waveband of 0.3-0.8μm range to propose a method for skin detection under varying lighting conditions (Angelopoulou et al., 2001; Pan and Healey, 2003). Moreover, some progresses with hyperspectral technology have been done on biomedical engineering application (Vo-Dinh, 2004).

In this paper, we describe a compact device for tongue diagnosis that uses hyperspectral technology that is applied for the purposes of TCM.

The remainder of this paper is organized as the following: in Section 2, we introduce the setup of the device. Section 3 provides and discusses some data collected by this device. Section 4 offers a brief conclusion.

The tongue image capturing sensor device is based on the theory of the push-scan hyperspectral imager which is shown in Figure 1. It consists of a spectrometer, a matrix CCD, an instrument translation module, and a data collection module.

In system, we use the Kohler illumination light source. To get the hyperspectral images, a reflectance stripe of the observed subject is firstly mapped onto the entrance slit of the imaging spectrometer. Then, through the grating and prism, we can get the dispersive spectrum in the strip's vertical direction. At last, it will be imaged on the matrix CCD. For the CCD imaging plan, we call its dimension which is parallel to the slit as “special dimension” and the vertical dimension as “spectrum dimension”. Each photoconductive-element of the special dimension can get an image of a certain wave band of sample stripe. Thus, each image frame of the CCD plan is corresponding to a multi-spectrum image of the sampled stripe.

We need a push-scan module in the spectrum dimension to get the whole image of the object. In the application of aeronautic remote sensing, the push-scan hyperspectral imager is implemented by the front movement of plane to complete a push-scan procedure. In our device, we designed the instrument translation model to do this work. To get a satisfactory hyperspectral images requires precise control of this model, so we use a high precision stepping motor to drive the push-scan under the control of a single chip microcomputer.

Using this hyperspectral tongue image capture device, we can get a sequence of images. In another words, the HSI approach provides a “data cube” of spectral information, which consists of a series of optical images recorded at various wavelengths of interest. An example image cube is shown by Figure 2. Each pixel of the image has two properties: the spectrum property and the luminance property. We analyze these “image cube” in the spectrum range to segment the tongue area.

The spectrum range of our system is 400-900 nm. The number of the efficient pixels are 652 × 488 including 120 wave band. The resolution of the spectrum is better than 5 nm.

This paper presents a novel medical TCM tongue diagnosis device that makes use of hyperspectral imaging technology. This sensor system can overcome some limitations of the conventional image capture method. Experimental results using 1,522 clinical tongue images demonstrate that our new sensor is able to interpret tongue images effectively. This novel use of a hyperspectral medical sensor provides a considerable advance on the usual image capturing method.