Diagnosis of uremia patients based on color space

WANG Xue-min, SHAO Na, WANG Rui-yun, SUN Ying-ying,YU Zhi-feng, JIANG Zhi-hao, ZHOU Peng

(1. School of Precision Instruments and Optical Electronics Engineering， Tianjin University， Tianjin 300072， China；2. College of Traditional Chinese Medicine, Tianjin University of Traditional Chinese Medicine， Tianjin 300193， China)

Abstract： Since traditional nail diagnosis is susceptible to objective factors such as illumination, medical experience, etc., a nail diagnosis system of traditional Chinese medicine which combines objective nail image acquisition with information analysis is proposed and applied to the clinical research of nail diagnosis of uremia patients. Fifteen nail pictures of uremic patients were collected and segmented. The color information of nails was extracted. The relationship between the hemoglobin values of uremic patients and the values of color space before and after maintenance hemodialysis was analyzed by correlation analysis and multiple regression analysis. The experimental results show that the hemoglobin value of uremic patients have certain correlations with multiple color channels; before and after dialysis, there are significant changes in multiple channels; and the related multiple regression equation is established.

Key words： nail classification; nail color; uremia; correlation analysis; paired t test; multiple regression

0 Introduction

Traditional Chinese medicine holds that the human body is an organic whole. The boom-and-bust of the functional activities of viscera is reflected in the muscles, skin and nails of the human body through the meridian system. It is based on the principle[1]of “depending on its external response, to know its internal, to know the disease”. The fingernail diagnosis is of great practical significance in many medical and health fields, such as cardiovascular and cerebrovascular diseases, hepatitis, constitution, and inheritance of traditional Chinese medicine. The large number of characteristics of fingernail images and lack of objective indicators are the key factors that restrict the development of objectivity of fingernail diagnosis. Current research methods mostly use modern physical and biochemical technology to compare and analyze the nail components of different observation objects. It has been reported that the relationship between renal failure and nail free edge creatinine content has been studied by nail creatinine assay[2]; the difference of infrared spectra between healthy people and cancer patients has been compared by infrared spectroscopy[3]; and the influence of malignant tumors on nail trace elements has been explored[4].

In order to promote the study of the objectification of nail diagnosis, a method of nail color analysis based on color space is proposed, which combines the theory of nail diagnosis with computer image processing and information analysis technology, to study the objectification of fingernail images of uremic patients. The image system was used to collect and segment the frontal and lateral images of the image which meets the requirements of the diagnosis of fingernail. The unique image information of uremia patients was explored by using statistics and numerical fitting method, and compared by means of modern medical theory.

1 Image acquisition and fingernail segmentation

The nail image acquisition system is a bidirectional imaging system composed of hardware and software. As shown in Fig.1(a), the hardware system mainly completes the collection of fingernail images, which consists of three parts: dark box, white LED lamp and camera positioned vertically. Dark box settings can ensure a stable nail image acquisition environment; white LED lights with color index >80, color temperature of 5 500 K and illumination of 4 000 Lx are used to simulate natural light, and the light path is shown in Fig.1(b); vertical camera and horizontal camera are responsible for collecting the front and side photos of nails. Canon 700D micro-range camera is used to shoot frontal image. The camera is equipped with DIGIC5 image processing system. The size of the image is 5 180×3 456 pixels. The side image is taken by Bainaotong camera. The size of the image is 1 280×1 024 pixels. The focus mode is manual focusing. The software system is composed of the upper computer software of Canon camera and Bainaotong camera.

Fig.1 Fingernail image acquisition system

Fifteen nail images of uremia patients were collected by this instrument, as shown in Fig.2(a). The images were processed by pretreatment, convex hull extraction[5]and Snake segmentation, as shown in Fig.2(b). The non-meniscus regions of nails are extracted by manual segmentation method. The segmentation effect is shown in Fig.2(c).

Fig.2 Nail images

2 Color space

Feature information extraction is the final result of image processing. In the process of nail diagnosis objectification, it is an important step to extract the feature information of nails after effective segmentation[6]. Doctors often diagnose the location of the lesion through the shape, size, color, blood gas symbols, half-moon mark area and so on. In the process of objectification of A diagnosis, it is particularly important to extract feature information and assist doctors in diagnosis. In this paper, we mainly extract the characteristic information of color, which is mainly presented in the way of color channel in the process of objectification of A diagnosis. Each image has many color channels, and the color pattern of the image determines the number of color channels of the image. In this paper, we extractRGBcolor space,RGBcolor space,Labcolor space,Hsicolor space andYCbCrcolor space.

2.1 RGB color space

InRGBcolor space[7], “R” is red, “G” is green and “B” is blue. According to the principle of three primary colors, light of any color can be mixed byR,GandB. When the color of the three primary colors is 0 (the weakest), the mixed color is black; when the color of the three primary colors is 1 (the strongest), the mixed color is white. Any color light is the coordinate point of the optical vertical cube. Adjusting the weighting factors inR,GandBof the tricolor light will change the coordinate value of the compound color light, thus changing the color value of the mixed light. The RGB color space is shown in Fig.3.

Fig.3 RGB color space model

2.2 rgb color space

Thergbcolor space is derived fromRGBspace, which has certain significance only in mathematical statistics. The conversion betweenRGBcolor space andRGBcolor space is expressed as

(1)

2.3 Hsi color space

InHsicolor space[8], “H” is Hue, which determines the spectral composition of the color light; “s” is Saturation, which determines the purity of the color light; “i” is intensity, which determines the intensity of the color light. The top part of the cone represents the tone, and the range of color represents the gradual change of six standard colors: red, yellow, green, cyan, blue and magenta. The axis coordinate system of the cone represents the brightness. The value increases gradually from bottom to top, indicating the process of gradual brightness enhancement. The horizontal direction of the cone represents the saturation, and the value increases gradually from inside to outside. The depth of the color representation is gradually strengthened. TheHsicolor space is shown in Fig.4.

Fig.4 Hsi color space model

The conversion betweenHsicolor space andRGBcolor space is expressed as

(2)

2.4 Lab color space

InLabcolor space[9], “L” is luminance, and the numeric value is black to white from small to large; “a” is red to green; “b” is yellow to blue. TheLabcolor space belongs to an uncommon color space. This color space is based on the international standard of color measurement established by the International Lighting Commission (CIE) in 1931. Since its modification in 1976, it has been officially named CIELab. This color space is a device-independent color system and a color system based on physiological characteristics. It also means that it describes people’s vision in the form of numbers. TheLabcolor space is shown in Fig.5.

Fig.5 Lab color model

The conversion betweenRGBspace andLabspace needsXYZspace. First,RGBspace is converted toXYZcolor space, and thenXYZcolor space is converted toLabcolor space. The conversion betweenRGBspace andXYZcolor space is expressed as

(3)

The conversion relationship betweenXYZcolor space andLabcolor space is expressed as

(4)

Thef(t) function is expressed as

(5)

2.5 YCbCr color space

InYCbCrcolor space[10], “Y” refers to brightness component, “Cb” refers to blue color component, and “Cr” refers to red color component. If only the signal component ofYchannel is present in an image, the image is a gray image. The conversion formula betweenRGBcolor space andYCbCrcolor space is expressed as

3 Experimental analysis

Fifteen patients with uremia were treated with maintenance hemodialysis (MHD) in Tianjin Heping District Hospital around December 2016. Among them, seven patients were male and eight patients were female. The average age was 60.7±13.1 years. The duration of maintenance hemodialysis was 10.5±4.6 years. All patients were fully dialyzed. The primary disease was renal cyst in one case, hypertension in two cases, chronic glomerulonephritis in four cases, polycystic kidney in two cases, diabetic nephropathy in two cases and unknown primary disease in four cases.

The color space of nails in non-meniscus area (non-meniscus area) was studied by analyzing the image of nails before and after nail defect in 15 cases. The average value of each color space channel in the non-meniscus area of fingernail area was traversed and normalized. The correlation analysis, paired t-test analysis and multiple linear regression analysis were carried out by SPSS 25.0 software to explore the changes of uremic patients’ fingernail before and after dialysis.

In order to prevent tedious description, in the following data analysis, “RGB-R(-)” is used to instead of the R channel in the pre-dialysisRGBcolor space, “Hsi-s(-)” is used to instead of theschannel in the pre-dialysisHsicolor space, and “RGB-R(+)” is used to instead of theRchannel in the post-dialysisRGBcolor space, and “Hsi-s(+)” is used to instead of theschannel in the post-dialysisHsicolor space.

3.1 Relevance analysis

The purpose of correlation analysis is to study whether there is correlation between data and its degree of closeness[11]. The correlations between the hemoglobin values of 15 uremic patients and the values of all channels before and after dialysis were analyzed, and the correlation coefficients were counted. Some data were shown in Fig.6.

It can be seen from Fig.6 that the correlation between hemoglobin value andHsi-H(-) is 0.530,Lab-a(+) is 0.529, andHsi-H(+) is 0.605, showing a significant level of 0.05, which indicates that there is a significant positive correlation between hemoglobin value andHsi-H(-),Lab-a(+) andHsi-H(+). The correlation value between hemoglobin value andRGB-g(+) is -0.535,Lab-b(+) is -0.627, showing a significant level of 0.05, which indicates that there is a significant negative correlation between hemoglobin value andrgb-g(+) andLab-b(+).

Note: Grey column is p>0.05 and black column is pFig.6 Pearson correlation coefficients between hemoglobin values and partial channels

3.2 Paired sample t test

Paired samplettest is a special case of single samplettest[12]. There are many situations in pairedttest: two paired subjects receive two different treatments; the same subject receives two different treatments; the same subject receives two different treatments before and after the treatment results are compared (i.e. self-pairing); two parts of the same object are given different treatments. In this paper, paired samplesttest of color channels of uremic patients before and after dialysis was carried out. The results are shown in Table 1.

It can be seen that before and after dialysis,RGB-Rchannel,Lab-Lchannel andYCbCr-Ychannel show a significant difference at the level of 0.01, and the value of pre-dialysis channel is significantly higher than that of post-dialysis channel.Hsi-schannel shows a significant difference at the level of 0.01, and the channel value before dialysis is significantly lower than that after dialysis.RGB-Gchannel andHsi-ichannel shows a significant difference at the level of 0.05, and the value of pre-dialysis channel is significantly higher than that of post-dialysis channel.

Table 1 Matched sample t test statistical results

3.3 Multivariate linear regression analysis

By using SPSS23.0 software and stepwise regression method, the data were analyzed by multiple linear regression analysis[13]. The relationships between hemoglobin values and the values of color channels of uremic patients before and after dialysis were analyzed. The results of the multiple linear regression model of hemoglobin values are shown in Table 2.

Table 2 Results of hemoglobin values of multivariate linear regression model

The multivariate regression equation is expressed as

y=6 587.278-611.624x1-

13 110.176x2+159.077x3.

(7)

Among them,x1is the component ofLab-b(+), that is, the component ofBchannel in Lab color space after dialysis, andx2is the component ofYCbCr-Cb(+), that is, the component ofCbchannel inYCbCrcolor space after dialysis, andx3is the component ofRGB-B(-), that is, the component ofBchannel inRGBcolor space before dialysis.

Table 3 Fitting test of regression equation of hemoglobin values

OrdernumberActualvaluePredictedvalueThe relativeerror (%)1118118.620.522115113.860.993114117.823.354121121.830.695126123.901.676102100.221.7578893.716.498113111.811.0599499.235.5610126116.577.491110199.801.1912123123.710.5813115123.347.2514113101.1910.45159194.383.71Average3.52

4 Conclusion

Aiming at the problem that traditional nail diagnosis is susceptible to objective factors such as illumination and medical experience, a nail diagnosis system of traditional Chinese medicine, which combines objective nail image acquisition with information analysis, is proposed and applied to the clinical research of nail diagnosis in uremia patients. Fifteen nail pictures of uremic patients were collected and segmented. The color information of nails was extracted. The relationships between the hemoglobin values of uremic patients and the values of color space before and after maintenance hemodialysis were analyzed by correlation analysis and multiple regression analysis. In correlation analysis, hemoglobin value andHsi-H(-),Lab-a(+) andHsi-H(+) show a significant level of positive correlation at 0.05, and hemoglobin value,rgb-g(+) andLab-b(+) show a significant negative correlation at the level of 0.05. In the paired test, before and after dialysis,RGB-Rchannel,Lab-Lchannel,YCbCr-Ychannel andHsi-schannel show a significant difference at the level of 0.01.RGB-Gchannel andHsi-ichannel show a significant difference at the level of 0.05. In multivariate analysis, the multiple regression equations of hemoglobin value andLab-b(+),YCbCr-Cb(+),RGB-B(-) are established.