Study on the difference and regularity of tongue images in 309 patients with different pathological stages of non-small cell lung cancer

1.Introduction

Malignant tumors are significant diseases that seriously threaten human life and health. The latest statistics show that China ranks first in the world in terms of new cancer incidence and mortality, with 4.57 million new cancer cases in China, accounting for 23.7% of new cases worldwide (19.29 million new patients worldwide), and 3 million cancer deaths, accounting for 30% of total cancer deaths (9.96 million cancer deaths worldwide). Among these, lung cancer is China’s most common malignant tumor, with 820,000 new cases and 710,000 deaths. According to pathology, lung cancer can be classified into two categories: non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC), of which NSCLC is the most common [1, 2, 3]. There are an increasing numbers of studies on lung cancer prevention, diagnosis, and treatment, especially NSCLC. Tongue diagnosis is a crucial traditional Chinese medicine (TCM) inspection method for TCM syndrome differentiation and treatment. Tongue diagnosis identifies the location and nature of the disease, speculates on the cause and mechanism of the disease, and suggests the progression and prognosis of the disease [4]. In the past 30 years, many researchers have observed the tumor of the tongue characteristics. In 1987, the research results of the Cooperative Group of Traditional Chinese Medicine and Western Medicine Cancer Professional Committee recommended specific rules for the changes in tongue pictures of patients with tumors [5]. However, the clinical application of tongue diagnosis is affected by the external environment and doctors’ experience. Therefore, the theory and clinical experience of tongue diagnosis have been systematically summarized. The use of modern science and technology to establish an objective standard for tongue diagnosis is a significant challenge in expanding its application scope of tongue diagnosis. Since the 1990s, with the development of computer digital image processing technology and the progress of objectification of tongue diagnosis, many achievements have been made in the standardization of tongue image acquisition in Traditional Chinese Medicine Quantification [6, 7, 8, 9, 10, 11, 12, 13, 14, 15]. Researchers have increasingly applied objective tongue diagnosis to diagnose and treat clinical diseases and evaluate the curative effect based on research on classification and recognition.

This study describes the tongue image characteristics of non-small cell lung cancer with datable tongue image indices. Further, it analyzes the tongue image characteristics of NSCLC patients with different tumor node metastasis (TNM) stages. We introduced image indices that objectively reflect the diagnostic information of tongue images based on clinical observations. We provided technical support for establishing a combined Chinese and Western medicine lung cancer diagnosis and treatment evaluation system incorporating data-based information on tongue diagnosis.

2.Materials and methods

2.1Clinical information

2.1.1Study design and participants

The NSCLC group included 316 patients hospitalized in the Yueyang Hospital of Integrated Traditional Chinese and Western Medicine oncology ward between October 2016 and November 2018. Additionally, a total of 553 people were screened from September to December 2017 at the health check-up center of Shuguang Hospital as a control group. Tongue images of the patients were collected using a uniform tongue diagnostic device. A flowchart of the process is shown in Fig. 1.

Figure 1.

Flowchart.

2.1.2Diagnostic criteria

(1) Diagnostic criteria for non-small cell lung cancer

The National Comprehensive cancer network (NCCN) published guidelines for non-small cell lung cancer in 2017 and the principles of diagnosis and evaluation of clinical guidelines for diagnosis and treatment (2011 edition) [16, 17].

(2) Diagnostic criteria for the TNM staging of non-small cell lung cancer

In this study, according to the TNM staging standard (8th edition), the Union for International Cancer Control (UICC) issued in 2018 [18], only patients with NSCLC were observed, and no clinical intervention was carried out [16]. The American Joint Committee on Cancer (AJCC) cancer staging manual [19]. Patients with non-small cell tumors were classified into I, II, III, and IV stages without further detailed division.

2.1.3Inclusion criteria and exclusion criteria

Inclusion and exclusion criteria in the control group

(1) Inclusion criteria for the control group:

• a. There was no history of cancer diagnosis, no acute disease diagnosis within three months, and no treatment according to the clinical diagnostic criteria of the disease.

• b. Age between 25–85 years old;

• c. Unlimited gender.

(2) Exclusion criteria for the control group:

• a. Those who did not meet the inclusion criteria of the control group;

• b. Pregnant or lactating women;

• c. Unable to cooperate with researchers because of subjective and objective reasons.

Inclusion and exclusion criteria in the NSCLC group

(1) Inclusion criteria for patients with non-small cell lung cancer:

• a. The diagnosis is precise;

• b. The age is between 25–85;

• c. There is no restriction on gender.

(2) Exclusion criteria for patients with non-small cell lung cancer:

• a. Those who did not meet the inclusion criteria for NSCLC;

• b. Lung cancer is complicated with other significant diseases;

• c. Pregnant or lactating women;

• d. Unable to cooperate with researchers because of subjective and objective reasons.

2.2Experimental instruments and methods

2.2.1Experimental apparatus

The TDA-1 digital tongue diagnosis instrument was used as the tongue image acquisition equipment (developed by the Laboratory of Intelligent Processing of TCM Diagnostic Information, Shanghai University of Traditional Chinese Medicine, patent number: CN201814562U) [20]. The tongue image diagnosis analysis system (TDAS v2.0) developed by the TCM Diagnostic Information Intelligent Processing Laboratory, Shanghai University of Traditional Chinese Medicine (Software Copyright:2018SR033451) was used to obtain relevant data. All tongue image collection and inquiry studies were completed by professional TCM or integrated TCM and Western medicine personnel who had received standardized training to ensure the consistency and authenticity of data collection and interpretation, and to minimize deviation. The TDA-1 digital tongue diagnosis instrument is shown in Fig. 2, and the corresponding tongue image analysis system TDAS v2.0 is shown in Fig. 3.

Figure 2.

TDA-1 digital tongue diagnosis instrument: (a) front view; (b) profile view.

Figure 3.

Tongue image diagnosis analysis system (TDAS v2.0) of TDA-1 digital tongue diagnosis instrument.

2.2.2Experimental methods

Tongue image acquisition

• (1) The tongue images were collected before breakfast from 6:15 to 7:15 in the morning.

• (2) The process of tongue image acquisition was based on the standard procedure of tongue image acquisition developed by the Laboratory of Intelligent Diagnosis Technology, Shanghai University of Traditional Chinese Medicine.

Feature extraction of tongue image

The special tongue analysis software TDAS v2.0 extracts all tongue indices. The tongue image color index is derived from the International Commission on Illumination Lab color space (CIELab). The pixel’s brightness is represented with “L,” and its value range is [0–100], representing pure black to pure white. The range from red to green is indicated with “a,” and its value range is [127–128]. “b” denotes the range from yellow to blue, the value range is [127–128]; the diagrams of Lab color space are shown in Fig. 4. Lab color space is a type of uniform color space that is a color space independent of equipment. It is a color system based on physiological characteristics. In other words, Lab color space uses a digital method to describe human visual perception. It is a color space suitable for the clinical description of tongue features [21]. It has been used in tongue diagnosis by several researchers [22, 23]. Previous studies have shown that the Lab color space best describes and distinguishes the tongue color of Traditional Chinese Medicine. Previous studies have shown that the lab color space best describes and differentiates tongue colors in TCM. The four tongue colors (pale red, pale white, red-red, and blue-violet) can be distinguished by the a-value and b-value. It is recommended that the L*a*b color space is most suitable for TCM’s description and differentiation of tongue colors [24]. In this study, we used the Lab color space index as the primary reference index to describe tongue color, in which L (lightness), a (red-green axis), b (yellow-blue axis), tongue coating index: perAll (perAll = the ratio of coated tongue area to total tongue area), and perPart (the ratio of effective pixels of tongue body to the effective pixel of tongue coating). Texture indices include CON (Contrast), ASM (Angular Second Moment), ENT(Entropy), and MEAN. The image index of the tongue body (TB) includes the color index (TB-L, TB-a, and TB-b) and texture index (TB-CON, TB-ENT, TB-ASM, and TB-MEAN). The image indexes of tongue coating (TC) included the color index (TC-L, TC-a, TC-b), texture index (TC-CON, TC-ENT, TC-ASM, TC-MEAN), and tongue coating index (perAll, perPart).

Figure 4.

Lab space color.

2.2.3Statistical methods

Data conforming to a normal distribution were expressed as mean (standard deviation) and Mean (SD). Data not conforming to a normal distribution were expressed as median (25th percentile, 75th percentile) and Median (IQR). Measures were compared between three or two groups for normally distributed data using analysis of variance (ANOVA), multiple comparison tests, or t-tests. Kruskal-Wallis analysis of variance (ANOVA) or Mann-Whitney U-test was used for non-normally distributed data. Count data were expressed as percentages (%) and compared using Pearson’s χ2 test or Fisher’s exact test. One-way conditional logistic regression analysis was performed on statistically different factors by the t-test and χ2 test. A factor was considered a protective factor when the dominance ratio (odds ratio, OR) was less than one and a risk factor when the OR value was > 1. Factors that were statistically different by the t-test and χ2 test were subjected to one-way conditional logistic regression analysis. Excel and python 3.0 were used for data matching, merging, and sorting; SPSS version 26.0 (IBM Corp., Armonk, NY, USA) statistical software was used for processing. All test results were two-tailed (P< 0.05). GraphPad Prism 8 software was used to draw violin plots of the tongue data.

3.Experimental results

3.1General information

Age is an essential factor affecting lung cancer incidence [19]. Therefore, according to the inclusion criteria and physical examination results of the control group, 51 cases of the elderly control group older than 55 years were selected, and each was sampled once. At the same time, a total of 316 NSCLC patients were sampled, a total of 1257 images of NSCLC patients, excluding 6 cases of unclear stage classification, two times of surgery, finally including 309 people, for a total of 1235 tongue image data. The three groups’ basic information is shown in Tables 1 and 2.

Table 1

Comparison of essential information between NSCLC and control group [n (%), median (IQR)]

Project		Healthy control group n= 163		Elderly control group n= 51		NSCLC group n= 309		Total
Gender n%	Female	84	(51.5)	23	(45.1)	169	(54.7)	192 (53.3)
	Male	79	(48.5)	28	(54.9)	140	(45.3)	168 (46.7)
Age median (IQR)		32	(28–40)	61	(58–65)	64	(57.50–70.00)*

^*Statistically significant difference (P< 0.05). Pearson χ2 test was used for categorical data, and the Mann-Whitney U test was used for measurement data.

Table 2

History of NSCLC patients

Project		Number
Total (n)		309
Age, median (IQR)		64	(57.50–70.00)
The course of lung cancer, median (IQR)		14.317	(5.233–24.567)
Gender, n%	Female	169	(54.7)
	Male	140	(45.3)
Pathological type, n%	Adenocarcinoma	264	(85.4)
	Squamous cell carcinoma	45	(14.6)
Pathological stage, n%	I	137	(44.3)
	II	20	(6.5)
	III	35	(11.3)
	IV	117	(37.9)
At the same time, n		58	(18.77)
EGFR (+), n%		62	(20.06)
Smoking history, n%	No	235	(79.9)
	Yes	59	(20.1)
Family history, n%	Nothing	272	(92.5)
	Yes	22	(7.5)
Hypertension, n%	No	223	(72.9)
	Yes	83	(27.1)
Diabetes, n%	No	275	(90.2)
	Yes	30	(9.8)
Hyperlipidemia, n%	No	297	(97.1)
	Yes	9	(2.9)
Operation condition, n%	No operation	81	(26.2)
	Operation	228	(73.8)
Chemotherapy, n%	Not used	151	(48.9)
	Used	158	(51.1)
	Used in a month	21	(6.9)
Targeted therapy, n%	No	256	(82.8)
	Yes	53	(17.2)

Note: Owing to the lack of clinical data, the number of people in some items is not equal to the total number.

Table 1 shows that the gender and age distribution of the NSCLC group was not statistically different (P> 0.05) compared with the elderly control group. However, compared with the healthy control group, the age of the NSCLC group was different (P< 0.05), and the gender distribution was not different (P> 0.05).

Figure 5.

Violin plot of tongue data between the Elder Control and NSCLC.

Figure 6.

Violin plot of tongue data among the Health Control and NSCLC.

3.3Analysis of tongue image characteristics between the NSCLC group and the healthy control group

The presence of characteristic differences in tongue image indexes between the NSCLC group (309 people) and the healthy control group (163 persons) was analyzed to control the influence of baseline factors on tongue image because of the difference in age distribution between the NSCLC group and the healthy control group. The differences in tongue image characteristics between the NSCLC and healthy control groups were analyzed using binary logistic regression. The results are shown in Tables 5 and 7, respectively.

Table 5

Logistic analysis of tongue image index between the NSCLC and healthy control

Index	B		OR		95% CI.				P
					Lower limit		Upper limit
Gender	-0.	520	0.	594	0.	250	1.	411	0.238
Age	0.	242	1.	274	1.	212	1.	340	0.000
perAll	5.	982	396.	041	16.	060	9766.	617	0.000
TC-L	-0.	161	0.	851	0.	731	0.	991	0.038
TC-a	-0.	389	0.	678	0.	539	0.	852	0.001

Note: The dependent variables were the NSCLC group and healthy control group (0 = healthy control group, 1 = NSCLC group), the control variables included gender (1 = female, 2 = male) and age, and the independent variables included TB-L, TB-a, TB-b, TCL, TC-a, TC-b, TBASM, TCASM, and perAll. None of the variables had collinearity problems (VIF < 10). Only tongue image indices that could enter the regression equation were included in the table (P< 0.05). Forward: The LR method was used.

Figure 7.

Forrest plot of the NSCLC and healthy control.

The results in Table 5 show that after controlling for the influence of sex and age, the tongue image indicators perAll, TC-L, and TC-a in the NSCLC and healthy control groups were specific between the two groups (P< 0.05). Compared to the healthy control group, perAll was a risk factor for the NSCLC group (OR > 1), and TC-L and TC-a were protective factors (OR < 1). The ranking of the influence of the three tongue image indices on the classification of NSCLC and healthy control group was perAll > TC-L > TC-a by calculating the OR value, indicating that perAll is the most specific index for classifying NSCLC patients’ tongue images and healthy tongue images.

The results showed that the tongue coating index was the main difference in tongue image features between the NSCLC and healthy control groups. The tongue coating in the NSCLC group was thicker, the red component of the tongue coating was reduced, and the brightness was lower compared with the healthy control group. The above results showed that the tongue image index could reflect the difference between the tongue image of patients with NSCLC and normal tongue image data; therefore, it could be used to represent the change in tongue image characteristics of NSCLC patients.

4.Discussion

NSCLC is all epithelial cancer except SCLC. All types of NSCLC are likely to occur. The most common types of NSCLC are adenocarcinomas, squamous cell carcinomas, and large cell carcinomas. The etiology and pathogenesis of tumors are complex, and their etiology remains unknown. The top risk factors for lung cancer are age, smoking history, past cancer history, family history, and occupational exposure [16].

The results showed that compared with the elderly control group, the tongue body and tongue coating brightness of NSCLC patients increased, the red component decreased, the tongue coating thickened, and the yellow component increased. In addition, the texture of the tongue coating was more delicate, showing a tendency for the thick and greasy coating to change. 1235 tongue images of NSCLC were compared with those of 163 healthy controls. The results showed that the brightness of the NSCLC tongue increased, the red component decreased, the tongue coating thickened, and the yellow component increased. This conclusion is similar to those of previous studies.

Preliminary findings show that a thick tongue with reduced brightness distinguishes a tongue image of NSCLC from that of a healthy tongue. A binary logistic stepwise regression was used to control for gender and age since there was a significant difference in the age distribution between the NSCLC group and the healthy control group (P< 0.05) to exclude the influence of age and gender on the tongue image. The results showed that the three variables that could enter the regression equation were all tongue-related indicators. According to the team’s previous study, perAll had a higher diagnostic value for a thick tongue, with a higher value indicating a thicker tongue. Logistic regression results also showed that a thick tongue was a risk factor for NSCLC, and the intensity of the effect was higher, indicating that a thicker tongue indicated an increased likelihood of NSCLC.

Generally, the development of tumor disease from stage I to stage IV is a gradual process from excessive pathogenic factors to positive deficiency. However, because of chronic diseases, phlegm and blood stasis are often produced. In the later clinical stage, the increase in TCM deficiency syndrome and mixed syndromes, and the appearance of the tongue changes from pale to dark and even bruised. The results of this experiment showed that there was no significant difference between the phase I group and the stage II + III index. The stage IV group showed a decrease in TB-b compared to the stage I and II + III groups. A decrease in TB-b indicates an increase in the blue component of the tongue. When the blue component of the tongue increased, the tongue became greener, and the brightness decreased. In the binary logistic regression analysis, after eliminating the influence of other confounding factors, TB-b in stage IV was positively correlated with stages I and II + III, which also showed that the blue component of the tongue image in stage IV increased. Comprehensively showing that patients in the stage IV group are bluer than those in stage I and II + III. The results showed that the change in tongue image in NSCLC patients was not significant from the beginning of stage I to the deepening of tongue color and the appearance of blue and purple tongues with the progression of the disease. This change was not obvious in the comparison between stage II and III and stage I patients. Compared with the previous three stages, patients in stage IV changed significantly. It also reflects that with the course of NSCLC, phlegm and blood stasis intermingled, and blood stasis appeared in the tongue. A clinical research report showed that NSCLC was mostly a blue-purple tongue, and the degree of bluish-purple coloration was aggravated with the progression of the disease [25]. Li Xiaojuan believed in the research on the pathogenesis of lung cancer in TCM that there are more patients with Qi deficiency and blood stasis syndrome in the middle and late stages of lung cancer, and they are accompanied by coagulation dysfunction [26]. The change in the coagulation index is related to tumor staging. In this study, tongue images of patients with NSCLC at different stages were described using objective tongue image indicators. The results showed that the tongue color of patients with advanced lung cancer was bluish and purple in the early and middle stages, consistent with previous research. It is proposed that tongue image features can reflect the characteristics of different stages of NSCLC, especially the tongue features of stage IV, which also means that tongue image features can be used as a reference for judging the course of NSCLC.

5.Conclusion

The results of this study show that tongue imaging features can provide an objective basis for the diagnosis and efficacy evaluation of NSCLC and can also provide new ideas and methods for the diagnosis and treatment of NSCLC using a combination of traditional Chinese and Western Medicine. The construction of this study was based on a clinical cross-section and belongs to a retrospective case-control design, which is a more commonly used research method for risk prediction models. But models constructed on the basis of disease outcome are not predictive models in the strict sense, while controlled studies have certain limitations, such as the inability to specify risk factors for the outcome and the possibility of bias in the selection of study subjects, etc., which to a certain extent have an impact on the accuracy and reliability of the study.

However, in the future, we will further expand the sample size for research because the clinical conditions are limited, the number of samples is too small, it is based on a cross-sectional study, and there is not enough observation time for continuous observation of pathological changes in samples. Moreover, this study shows that tongue coating changes have important implications in the progression of NSCLC; therefore, further studies will be conducted in the future.

Author contributions

Ling-zhi Zeng and Ji Cui contributed equally to this work. Ling-zhi Zeng designed the study and drafted the manuscript. Ji Cui was responsible for preprocessing the data and checking the results. Tao Jiang and Hai-dan Liu enrolled study subjects and collected data. Ji Cui and Li-ping Tu provided support on data analysis and interpretation. Ya-bin Gong and Ling Xu offered suggestions. Ji Cui and Jia-tuo Xu are the corresponding authors. All authors have reviewed and approved the final manuscript.

Funding

The work was supported by the National Natural Science Foundation of China (82104736, 82104738).

Data availability statement

The datasets generated and analyzed during the current study are not publicly available due to the confidentiality of the data, which is an important component of the National Key Technology R&D Program of the 13th five-year plan (No. 2017YFC1703301) in China. However, the datasets are available from the corresponding author upon reasonable request.