Increasing the accuracy in the diagnosis of stomach cancer based on color and lint features of tongue
Introduction
Cancer is a major health concern in human societies, which is associated with the highest mortality after cardiac disorders. Cancer could be caused by unhealthy foods, air pollution, smoking, and talc exposure [1,2]. Gastric cancer is the second common cancer in Asia [3]. According to statistics, about 800,000 new cases of gastric cancer are detected, and the yearly mortality of the disease has been reported to be 650,000 [4]. Gastric cancer is the most common cancer in Iran, and the prevalence rate has increased more significantly compared to other countries within the past two decades. The largest number of the gastric cancer patients is reported in the north and northwest of Iran. The most important risk factor for the disease is Helicobacter pylori bacterial infection, which leads to ulcers and gastritis [5]. Moreover, several environmental and genetic factors are known to increase the risk of gastric cancer.
Endoscopy is considered to be the optimal diagnostic measure for gastric cancer, which enables several biopsy removals of the wound margin [6]. The detection accuracy of endoscopy reaches 98 % by several biopsy removals. However, it is an invasive and costly procedure requiring anesthesia. If endoscopy is not performed correctly, the patient may experience numerous complications. Laparoscopy and gastroscopy are also used for gastric cancer diagnosis through the double-contrast imaging of the stomach by inserting a camera into the throat and stomach [7]. Other less common examinations in this regard include barium swallow, X-ray imaging of the upper digestive system, and endoscopic ultrasound.
In the absence of disease symptoms, the patient would not agree to undergo invasive procedures. Respiratory analysis is a novel, cost-efficient technique, which is also faster and simpler compared to the current diagnostic methods of gastric cancer [8]. In this test, the patient should blow on a pipe, and the physician examines the reaction of some sensors using the respiratory sample. Notably, serological methods (especially urea breathing test) are more common owing to lower invasiveness and cost-efficiency. However, these tests remain positive within six months to two years after treatment.
Incorrect diagnosis is a major obstacle to the treatment of gastric cancer in Iran [9]. Unfortunately, no common tests are available in Iran for the screening of gastric cancer patients. In some cases, the disease may remain undiagnosed despite prolonged symptoms such as stomachache, weight loss, dysphagia, and vomiting [10]. Gastric cancer is treatable if diagnosed early. In case of disease progression, treatment probability may decrease. Given the high prevalence of gastric cancer, some patients refer to health centers in the advanced stages of the disease. Therefore, effective diagnostic tools are essential to the early diagnosis of gastric cancer, so that the survival of the patient could be estimated accurately [11]. Gastric cancer treatment is often performed by using a combination of stomach surgery, chemotherapy, and radiotherapy, along with auxiliary treatments [12,13]. Gastrectomy (tumor removal) is considered to be the most efficient surgical protocol for gastric cancer, especially in the early stages of the disease.
Chinese traditional medicine is a natural and holistic health system, which dates back to 3–5 thousand years ago [14]. Iranian traditional medicine is a fine and valuable heritage, which dates back to 5,000 years BC in the Achaemenid era [15]. Iranian traditional medicine is an amalgamation of the Persian culture, community, and religious sciences, which emphasizes balance in the human body. The principles of Iranian traditional medicine could be incorporated into the age of technology given their effectiveness in the improvement of human health and disease elimination. Chinese traditional medicine has a long history of disease treatment in eastern Asia and is known as a complementary system and medical alternative in western countries [16].
Today, traditional medicine is well-received by common individuals, as well as scientific and medical centers across the world. Moreover, the World Health Organization (WHO) encourages the health policymakers of different countries to integrate traditional medicine into modern medicine. The scientific departments of traditional medicine are growing in number each year. According to the WHO strategy (2014–2023), Chinese traditional medicine will be expanded in the universities of more than 100 countries and subsequently converted into an industry [17].
Despite the widespread practice of Chinese traditional medicine in China and western countries, the precise scientific observations of its effects are limited. Disease diagnosis studies based on Chinese traditional medicine are also challenging due to the differences of the notions with modern western medicine [18]. In Chinese traditional medicine, disease diagnose is based on the obtained information from four diagnostic processes, including seeing, hearing, smelling, and touching. The diagnoses are primarily based on examining the pulse rate and tongue [19].
Examination of the tongue surface plays a key role in the diagnostic methods of Chinese traditional medicine. The importance of the tongue in Chinese traditional medicine is due to the fact that it is in palate, far from external and environmental factors [20]. According to the principles of Chinese traditional medicine, the tongue is divided into four sections, which are the tongue tip, tongue margin, tongue center, and end of the tongue (tongue growth). Fig. 1 shows the connections between different sections of the tongue and the internal organs of the body [21]. Correspondingly, tongue tip shows the pathological changes of the heart and lungs, and the two sides of the tongue show the changes in the liver and bladder sac. The center of the tongue reflects the pathological changes of the stomach and spleen, and the end of the tongue shows the changes of the kidneys, intestines, and bladder.
In Chinese traditional medicine, physicians diagnose different diseases based on the shape and size of the tongue, as well as the cover on the tongue, tongue color, and moisture on the tongue [22]. A healthy tongue is oval and pink, with a very thin cover. Some of the disease symptoms reflected in the tongue include color change, thickness increment, cover change, geometrical shape change, and signs of grooves on the tongue.
Currently, the methods that are commonly used for the diagnosis of gastric cancer and H. pylori infections are classified as invasive and noninvasive [23].
- A.
Endoscopy: It is a method involving anesthesia and high costs, which may yield false negative results in case of active/recent bleeding or in the process of anti-secretion treatment [24].
- B.
Polymerase chain reaction: In this method, H. pylori infections are diagnosed in small tissue samples that harbor the bacteria. The main limitation of this method is the DNA of bacteria, which remain in the stomach mucus from dead bacteria after the treatment and cause false positive results [25].
- C.
Rapid urease test: It is a primary method for the diagnosis of infections by using H. pylori capability in the generation of high urea levels. Although it is a faster and less costly method similar to tissue culture and investigation, its sensitivity depends on organism density. In other words, if the number of organisms is low, the test sensitivity could decrease to 30 %.
- A.
Serological tests: It is a cost-efficient method. Given the differences in H. pylori strains in various geographical areas, local antigenes of each area must be used in the laboratory diagnostic kits, so that the sensitivity of the test would not decrease. Moreover, many false positive tests have been observed due to the generated antibodies of other infections and cross-reactions [26].
- B.
Urea breathing test: This method is based on H. pylori urease enzyme activity, which is a time-consuming process and requires costly equipment. Moreover, the presence of other positive urease bacteria in the mouth and stomach holes may lead to false positive results [27].
- C.
H. pylori antigen detection in feces: This method is utilized for the detection of the H. pylori antigen in fecal samples using the ELISA assay and is more expensive compared to other tests [28].
According to Chinese traditional medicine, the cover on the tongue shows the physiological and pathological changes in the digestive organs, especially the stomach [29]. The reports of clinical studies in this regard indicate that tongue appearance provides essential information to physicians for the diagnosis, treatment, and prediction of chronic stomach diseases, gastrointestinal ulcers, and gastric/colon cancer [30]. Researchers have been attempting to find simple and cost-efficient methods for the early diagnose of cancer by investigating the association between coating metabolic symptoms and the cover of the tongue surface in order to diagnose chronic diseases such as gastric cancer.
In this article, Section 2 includes the introduction and investigation of the previous studies in this regard. The proposed method has been explained in Section 3, and the experimental results have been presented in Section 4. Section 5 concludes the paper and proposes some research implications.
Section snippets
Literature review
Various methods have been proposed to diagnose cancer and other chronic diseases based on artificial intelligence and different algorithms in the fields of machine learning, evolutionary methods, and neural networks. In general, each of the proposed methods has advantages and disadvantages. In this section, the advantages and disadvantages of the recent studies in this regard have been discussed.
Zhang et al. (2017) in [31] have proposed Diabet diagnoses based on tongue standard images using the
Proposed method
In this section, the proposed method has been explained to increase the precision of gastric cancer diagnosis using the features of coating and color of tongue based on convolutional deep neural networks and support vector machine. Fig. 2 illustrates the diagram of the proposed method.
Results
The results of the proposed method were evaluated in seven architectures of CNN. Table 1 shows the characteristics of the CNN architectures. Accordingly, Alexnet and DenseNet networks had the least and most depth, respectively. Moreover, GoogleNet and VGG-16 had the lowest and highest number of parameters (weight and bias of each layer), respectively.
Table 2 shows the diagnostic accuracy of various architectures applied to the primary dataset in CNN. As can be seen, DensNet network had the
Conclusion and research implications
Cancer is a severe, chronic disease that affects the human life, and gastric cancer is reported to be highly prevalent across the world. With technological advancement and the use of artificial intelligence algorithms, the diagnosis of chronic diseases has been facilitated. In this study, a method was proposed to increase the diagnostic accuracy of gastric cancer based on the surface and color features of the using a combination of deep neural network, support vector machine, and deep
Attachment
CRediT authorship contribution statement
Elham Gholami: Instructor, first author, methodology and implementations. Seyed Reza Kamel Tabbakh: Assistant professor, ideas and visualization. Maryam kheirabadi: Assistant professor, ideas, editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (43)
- et al.
Talc exposure and risk of stomach cancer: systematic review and meta-analysis of occupational cohort studies
J. Formos. Med. Assoc.
(2020) - et al.
Surgical treatment for peritoneal carcinomatosis from gastric cancer
Eur. J. Surg. Oncol.
(2010) A novel approach based on computerized image analysis for traditional Chinese medical diagnosis of the tongue
Comput. Methods Programs Biomed.
(2000)- et al.
TCM tongue diagnosis index of early-stage breast cancer
Complement. Ther. Med.
(2015) - et al.
RMDL: recalibrated multi-instance deep learning for whole slide gastric image classification
Med. Image Anal.
(2019) - et al.
Convolutional neural network for the diagnosis of early gastric cancer based on magnifying narrow band imaging
Gastric Cancer
(2020) - et al.
Gastric carcinoma
N. Engl. J. Med.
(1995) - et al.
Schwartz’s Principles of Surgery
(2005) - et al.
Current Surgical Diagnosis & Treatment
(2003) - et al.
Effects of Helicobacter pylori treatment on gastric cancer incidence and mortality in subgroups
J. Natl. Cancer Inst.
(2014)
Sabiston Textbook of Surgery
Epidemiology of gastric cancer
World J. Gastroenterol.
A nanomaterial-based breath test for distinguishing gastric cancer from benign gastric conditions
Br. J. Cancer
Survival rate of esophageal carcinoma in Iran - a systematic review and meta-analysis
Iran J Cancer Prevent
Gastric cancer in Iran 1966-2006
Asian Pac. J. Cancer Prev.
Smoking and histological factors influencing long-term survival of gastric carcinoma in consecutive patient series
Middle East J. Cancer
Survival rate estimation and its associated factors in patients with stomach cancer in Tohid Hospital in Sanandaj
Sci. J. Kurdistan Univ. Med. Sci.
Comparison of traditional chinese medicine and traditional iranian medicine in diagnostic aspect
Trad. Integr. Med.
The Canon of Medicine
Bridging the gap between traditional chinese medicine and systems biology: the connection of cold syndrome and NEI network
Mol. Biosyst.
WHO Traditional Medicine Strategy: 2014–2023
Cited by (17)
A multi-step approach for tongue image classification in patients with diabetes
2022, Computers in Biology and MedicineCitation Excerpt :We compare the clustering results of various features, including TDAS features, VQ-VAE features, LBP features, and HOG features. The LBP and HOG features are calculated by the Local Binary Pattern algorithm [43,55] and the Histogram of Oriented Gradient algorithm [56,57], commonly used in tongue diagnosis analysis. K-means performs clustering analysis based on all data without being split into train and test.
A novel tongue segmentation method based on improved U-Net
2022, NeurocomputingCitation Excerpt :For example, Strawberry tongues are associated with scarlet fever and staphylococcal sepsis [4]; Map tongue is related to tongue papillitis [5]; Black hairy tongue is quite unusual condition of a fungal infection [6]. As a non-invasive, intuitive, and rapid diagnostic method, Tongue diagnosis has been widely used in medical diagnosis [7-9]. With the advancement of intelligent medical care, some achievements have been made in the intelligent algorithm processing of tongue diagnosis by its intuitionistic characteristics [10-12].
A two-stage deep transfer learning model and its application for medical image processing in Traditional Chinese Medicine
2022, Knowledge-Based SystemsCitation Excerpt :Especially, as one of the TCM basic diagnosis theories, tongue diagnosis has attracted extensive attention among TCM practitioners and computer engineers. Recently, DNN based tongue diagnosis has been scientifically studied and integrated with western medicine [7,8], computer-based detection methods such as early diagnosis of stomach cancer [9] and nonalcoholic fatty liver disease (NAFLD) [10] under the help of TCM tongue diagnosis have been applied. Emerging technologies represented by attention networks [11] are being used to help build reliable models, providing clinical decision-making assistance for young medical workers [12].
Establishing and validating a spotted tongue recognition and extraction model based on multiscale convolutional neural network
2022, Digital Chinese MedicineCitation Excerpt :With the development of artificial intelligence (AI), new technology has been provided for tongue image recognition and extraction [6-8]. In recent years, deep learning and convolutional neural network (CNN) have been applied to the TCM feature analysis method of tongue images [9-16] and achieved better recognition accuracy than image algorithms and machine learning. However, the irregular distribution of spots on the tongue, the large difference between the size of spots and the tongue, and the small difference between the color of the spots and the tongue under natural light conditions, etc., are the difficulties and challenge of automatic recognition for spotted tongue [17].
A Systematic Review on Detection of Gastric Cancer in Endoscopic Imaging System in Artificial Intelligence Applications
2024, Lecture Notes in Networks and Systems