Development of multilayer fuzzy inference system for diagnosis of renal cancer

Abstract

In this research work, a new multilayer fuzzy inference system is proposed for diagnosis of renal cancer. This proposed automated diagnosis of renal cancer using multilayer Mamdani fuzzy inference system can help to classify the different stages of renal cancer such as no cancer, stage 1, stage 2, stage 3 or stage 4 cancer. This expert system has four input variables at layer 1 and similarly seven input variables at layer 2. At layer 1, the input variables are smoking, dialysis, occupational exposure and genetic or hereditary that recognize the output conditions of renal or kidney to be normal or to have renal cancer. The further input variables for layer 2 are haematuria (blood in urine), red blood cell count, flank pain, tumor size, Von Hippel-Lindau gene, high blood pressure and trichloroethylene exposure that reveal the output condition of kidney such as stage 1 cancer, stage 2 cancer, stage 3 cancer or stage 4 cancer. The novelty in this research work is development of multilayer fuzzy inference system that deals with fuzzy values, uncertain and ambiguous data to detect the stage of renal cancer by using two layers. This paper presents an analysis of results accurately using the proposed expert system to model the renal cancer process with medical expert advice. The confidence indicator for this proposed expert system is 95%.

Keywords

Artificial intelligence fuzzy inference system renal cancer

1 Introduction

The analysis of disease is very essential in the domain of healthcare and medicine. Also, the detection of disease at an early stage needs more attention and undoubtedly it becomes important [1, 2]. Similarly, the detection of renal or kidney cancer at an early stage is also important so that it can help patients to get the right cure and care at the right time which will lead to better health. Otherwise, it will affect the paired organs [3].

As the kidney helps to maintain the homeostatic balance of fluids and solute in the body of human beings, also it helps to remove the waste products from the blood and to control blood pressure in the human body. These main tasks make kidneys the most essential and vital organs of the human body [4]. There are certain questions like What are the major biomarkers for renal cancer and the risk factors of kidney cancer? How the identification and verifications of these biomarkers can be done? What are the main causes of renal cancer? To answer these types of questions, it needs much knowledge about renal or kidney cancer and also the past report or history of an individual patient. Hence, it is very crucial to detect and analyze renal cancer at the right time and to identify those conditions that cause renal cancer [5].

Renal cancer originates from the lining in the small tubes of kidneys which transport the urine. The cells of the kidney become cancerous and it grows out of control with time and age. These cancerous cells are then formed into the tumor or lump if the unwanted growth of kidney cells is not controlled at a proper time. This type of renal cancer is known as renal cell carcinoma (RCC). Renal cell carcinoma often stays in the kidney, bit if the growth will not be terminated at the time, it can spread to bones, brains and other organs of the body that are near to the kidneys.

In worldwide, kidney cancer has the 10th position among all the cancers in men as well as in women. The main cause of kidney cancer is smoking which concludes that men have a higher risk of being suffering from this disease than women. The other symptoms can be obesity, excessive alcohol consumption, hypertension, and low physical activity [6]. The kidney failure can be caused due to hypertension and high blood pressure and these symptoms are also known as ‘silent killer’ [7]. Additionally, the renal tumor is one of the main causes of renal cell carcinoma (RCC) [8]. Renal cell carcinoma is a category of renal cancer and out of 10, 9 are those cases in which the type of kidney cancer is renal cell carcinoma (RCC) [9]. Kidney cancer can be caused due to genetic or hereditary genes. There are 3% to 5% cases, where kidney cancer is caused due to hereditary genes. The result of identification is that there are 12 different types of the gene that cause kidney cancer and this number increases with time[10, 11]. Re-occurrence of renal cancer is also possible after a couple of years have been passed after diagnosis. Hence, it is necessary to monitor the health of patient with regular check-ups [12].

To diagnose renal cancer, many techniques have been developed by researchers. The technique which is frequently used for the detection of renal cancer is by using concept of image processing, In which the magnetic resonance image or computed tomography images were acquired and after using various image processing techniques such as segmentation, edge detection, fusion, visualization, etc the identification of renal cancer has been done [13 –15]. The image features can be seized by using a histogram. After capturing features of images, the classification is done by using data mining technique[16, 17]. There are computer-assisted clinical tools, which were used to assess and classify the tumors in renal with captured computed tomography images [18]. The age and birth also affect the renal and leads to renal cancer as well as pancreatic cancer [19]. The monitoring of tumor growth in kidney or renal is mandatory. Tumor tissues are responsible for the growth of renal tumors [20]. The behavior of tumor in the kidney is also detected by Particle Automata Model. This model allows the identification and detection of tumor growth factors and this model also helps in monitoring the expansion of renal cancer [21, 22]. The Adaptive Neuro-Fuzzy Inference System is also used to predict kidney movement from skin markers having more than 94% accuracy [23].

Artificial intelligence is basically the power of the computer program to act, think and learn like a human being. It is a domain that tries to make computers “smart” so that computers or machines can also do various complex tasks without any command. It is a wide-ranging branch of computer science that makes smart machines which are capable of performing various operations that are complex and requires human intelligence to solve. It is also used to diagnose numerous types of medical problems in these days[24, 25]. Medical issues can be fixed by using intelligent systems [26]. To predict the problem and provide their solution, a fuzzy inference system is a powerful tool and an effective method to diagnose the disease [27]. At present, it is used in every domain either it is a medical domain or automatic robotics or industry [28]. The fuzzy set theory deals with the uncertainty or ambiguity in the gathered data and in the measurements of errors, hence fuzzy set theory is a powerful and effective tool to solve these types of problems [29, 30]. The ambiguity can arise during the process of gathering data and implementation. The fuzzy differential equations are also used to deal with this type of data [31, 32]. Fuzzy logic is many-valued logic, addressing the ambiguity phenomenon and modeling of developing tool is done by truth values captured from an ordered scale [33]. The knowledge of an expert is moved from an expert to an expert system. After that, this knowledge is used to conclude the solution to a particular problem. 70% problems of the medical domain are resolved by using an expert system at the present time [34].

The fuzzy logic-based systems have four main parts. These blocks of fuzzy logic-based system or fuzzy inference system are discussed below:

Fuzzification Module: The fuzzification module helps to transform the given input in the form of crisp values into the fuzzy values.

Knowledge Base: The rules generated by the experts in the form of IF-THEN are stored in the knowledge base.

Inference Engine: The inference engine maps the rules stored in the knowledge base with the inputs given by the user to the system. It stimulates the process of human reasoning.

Defuzzification Module: The defuzzification module is used to convert or transform the obtained output by the system in the form of a fuzzy set into the crisp numbers.

The research work explained above gave effective and profitable knowledge about diagnosis of renal or kidney cancer. The main intention of this research paper is to develop an automatic diagnostic system for renal cancer by adopting a multilayer Mamdani fuzzy inference system.

2 Method

The proposed automated diagnosis of renal cancer using a multi-layered Mamdani fuzzy inference system is elaborated in this section. Figure 1 presents the methodology for the proposed expert system.

Fig. 1

Methodology of proposed expert system.

The expert system has two layers as shown in Fig. 2. In layer 1, renal cancer is identified using four input variables (smoking, dialysis, occupational exposure, genetic or hereditary) and the output of this layer will depict either there is cancer in renal or not.

Fig. 2

Developed an ADHB-ML-MFIS expert system.

The values of these input variables are also used in Table 8 to determine the condition of renal of patients. If layer 1 of the developed expert system gives the output as there is cancer in renal only then layer 2 will be activated. Layer 2 detects the stage of renal cancer or kidney cancer on the seven input variables as shown in Fig. 2. The input and output variables of Layer 2 are shown in Table 7. The algorithm designed for the proposed work is stated below:

Algorithm:

Define the linguistic terms and variables used in the developed model.

Construct the membership functions for all variables.

Create a knowledge base in which all the rules are stored.

Layer 1

Transform the crisp data (inputs at layer 1) into the fuzzy data by using membership functions.

Map the given inputs with the rules in the rule base or knowledge base.

Combine the final output from each rule.

Transform the calculated output from fuzzy to crisp values.

Layer 2

Use the output of layer 1 as the input of layer 2 with the rest of the decided inputs.

Convert all the inputs from crisp values to fuzzy values in the Fuzzification module.

Evaluate the rules stored in the knowledge base.

Merge all the results observed from each rule and calculate the final output.

Convert the final output given by layer 2 from fuzzy value to non-fuzzy value.

Mathematically, the layer 1 of the proposed expert system is explained as $μ_{RC, layer 1} = MFIS [μ_{smoking}, μ_{dialysis}, μ_{OE}, μ_{GOR}]$ (1) where RC = renal cancer, OE = occupational exposure, GOR = genetic or hereditary and MFIS =multilayered fuzzy inference system

and similarly, the layer 2 of ADRC-ML-MFIS (automated diagnosis of renal cancer using multilayer Mamdani fuzzy inference system) expert system can be written as follows

$\begin{matrix} μ_{RC, layer 2} = & MFIS [μ_{RC, layer 1}, μ_{hematuria}, μ_{RBCC}, \\ μ_{FP}, μ_{TS}, μ_{VHLG}, μ_{HBP}, μ_{TCEE}] \end{matrix}$ (2) where RC = renal cancer, RBCC = red blood cell count, FP = flank pain, TS = tumor size, VHLG = Von Hippel-Lindau gene, HBP = high blood pressure and TCEE = trichloroethylene exposure.

As it is difficult to study the entire population. Therefore, the tests are applied on a sample of 30 patients out of the total population to gain some useful and fruitful information from the data. Figure 3 –5 show the normality test applied by using the software MINITAB 19. The normality test used for this analysis is Anderson Darling and Ryan-Joiner’s normality test.

Fig. 3

Normality test for input variable “Tumor size”.

Fig. 4

Normality test for input variable “Trichloroethylene Exposure”.

Fig. 5

Normality test for input variable “Hematuria”.

Anderson Darling normality test: The Anderson–Darling test is a statistical test to check whether a given sample of data is drawn from a given probability distribution. The test rejects the hypothesis of normality when the p-value is less than or equal to 0.05.

$\begin{matrix} AD = & - n - \frac{1}{n} \sum_{i = 1}^{n} (2 i - 1) [ln F (X_{i}) \\ + ln (1 - F (X_{n - i + 1}))] \end{matrix}$ (3)

Where n = sample size, F(X) = Cumulative distribution function for the specified distribution and i = the i^th sample when the data is sorted in ascending order.

Ryan-Joiner normality test: The Ryan-Joiner statistic measures how well the data follows a normal distribution by calculating the correlation between data and the normal scores of data. If the p-value is greater than 0.100, the population is likely to be normal. $RJ = \frac{\sum Y_{i} b_{i}}{\sqrt{s^{2} (n - 1) \sum b_{i}^{2}}}$ (4)

Where Y_i = ordered observation, b_i= normal scores of ordered data and s² = sample variance.

Table 1 shows the required p-values using normality tests and Table 2 shows the obtained p-values by applying the normality tests. In Table 3, the various parameters for Tumor size are mentioned

Table 1

Required p-values in the normality tests

Normality test	p-value for acceptance of hypothesis	p-value for rejection of hypothesis
Anderson Darling	>0.05	< = 0.05
Ryan-Joiner	>0.100	< = 0.100

Table 2

Obtained p-values by applying the normality tests

Normality tests	p-value for Anderson Darling	p-value for Ryan-Joiner
Tumor Size	0.165	>0.100
Trichloroethylene Exposure	0.090	>0.100
Hematuria	0.095	>0.100

Table 3

Parameters for tumor size

Linguistic Labels	n	Mean	σ	α	tα,n-1	Zα=	LL	ULL	UUL	UL
Smaller	40	3.4	1.49	0.1	1.65	–0.1	0	5.304	6.94
Larger	30	7.4	0.88	0.1	2.045	5.94	7.103	10.4	13.6

Abbreviations of Table 1: n - Sample size; LL - Lower limit; ULL - Unilateral lower limit; UUL - Unilateral upper limit; UL -Upper limit.

2.1 Input variables

Input variables are fuzzy statistical values that are used to diagnose renal cancer. In this research, there are total of eleven different kinds of input variables that are used in the detection of kidney or renal cancer on both layers. Out of these eleven input variables, four variables are used in layer 1 and the rest of the seven input variables are used in layer 2. The proper detail of these input variables with their ranges is described in Tables 4 and 5.

Table 4
Input variables of layer 1 of the proposed expert system

Sr no. Input parameter Ranges Semantic sign Reference Ranges/Cutoff values

1. Smoking LT < 3.85 Acute 0–3

B/W

3–7 Moderate

GT > 6.2 Chronic

2. Dialysis LT < 3 No 0–2

GT > 2 Yes

3. OE LT < 3.8 LittleBit 0–2.5

GT > 2.5 Excessive

4. GOH LT < 3.51 Low 0–2.42

B/W

2.42–6.68 Moderate

GT > 5.8 High

Sr no.	Input parameter	Ranges	Semantic sign	Reference Ranges/Cutoff values
1.	Smoking	LT < 3.85	Acute	0–3
		B/W
		3–7	Moderate
		GT > 6.2	Chronic
2.	Dialysis	LT < 3	No	0–2
		GT > 2	Yes
3.	OE	LT < 3.8	LittleBit	0–2.5
		GT > 2.5	Excessive
4.	GOH	LT < 3.51	Low	0–2.42
		B/W
		2.42–6.68	Moderate
		GT > 5.8	High

LT = less than; GT = greater than; B/W = between.

Table 5

Input variables of layer 2 of the proposed expert system

Sr. no	Input Parameters	Ranges	Semantic sign	Reference Ranges/Cutoff values
1.	Hematuria	LT < 10.9	1+	0–9
		B/W
		6.94–25.46	2+
		B/W
		17.5–57.2	3+
		GT > 53.5	4+
2.	RBCC	LT < 0.35	Low	0.35–0.52
		B/W
		0.3–0.56	Normal
		GT > 0.52	High
3.	FP	LT < 0.36	Low	0–0.29
		B/W
		0.29–0.69	Medium
		GT > 0.64	High
4.	TS	LT < 6.9	Smaller	0–5.9
		GT > 5.9	Larger
5.	VHLG	LT < 0.36	No	0–0.3
		GT > 0.3	Yes
6.	HBP	LT < 85.05	Normal	60–85
		GT > 78.4	High
7.	TCEE	LT < 4.24	Low	0–3.7
		B/W
		0.73–7.01	Moderate
		GT > 6.76	High

LT = Less than; GT = Greater than; B/W = between.

2.2 Output variables

As the proposed system is based on multi-layered architecture, therefore, Number of outputs is two. The layer 1 will give output as yes or no. If the output of layer 1 is yes, then layer 2 will be activated. The output of layer 2 will be the stage of renal cancer. The output variables of layer1 and layer 2 are described in Table 6.

Table 6
Output variables of Layer 1 and Layer 2

Sr. no. Output variables Semantic sign

1. Layer 1 Renal cancer Yes

No

2. Layer 2 Cancer Stage Stage 1

Stage 2

Stage 3

Stage 4

Sr. no.	Output variables	Semantic sign
1.	Layer 1	Renal cancer	Yes
			No
2.	Layer 2	Cancer Stage	Stage 1
			Stage 2
			Stage 3
			Stage 4

2.3 Membership function

The statistical values of input and output variables of both layers are provided by mathematical functions. Table 7 represents the graphical and analytical views of membership functions of input and output variables of both layers.

Table 7
Input and output variables membership function used in the proposed expert system

Sr. Input/Output Membership function (MF) Graphical representation of MF

no Variables

1. Hematuria H(μ_H,(h)) μ_H,1 +(h) = max(min(1,10.91-h/0.1),0) μ_H,2 +(h) = max(min(h6.94/0.1,17.5,25.46-h/0.1),0) μ_H,3 +(h) = max(min(h-17.5/0.1,53.5,57.2-h/0.1),0) μ_H,4 +(h) = max(min(h-53.5/0.1,1),0)

2. Red Blood cell count (RBCC) R(μ_R,(r)) μ_R,L(r) = max(min(1,0.351-r/0.1),0) μ_R,M(r) = max(min(r-0.306/0.1, 0.52,0.565-r/0.1),0) μ_R,H(r) = max(min(r-0.52/0.1,1),0)

3. Flank Pain (FP) F(μ_F,(f)) μ_F,L(f) = max(min(1,0.36-f/0.1),0) μ_F,M(f) = max(min(f-0.295/0.1,0.64, 0.69-f/0.1),0) μ_F,H(f) = max(min(f-0.64/0.1,1),0)

4. Tumor Size (TS) T(μ_T,(t)) μ_T,S(t) = max(min(1,6.94-t/0.1),0) μ_T,L(t) = max(min(t-5.94/0.1,1),0)

5. Von Hippel-Lindau Gene (VHLG) V(μ_V,(v)) μ_V,N(v) = max(min(1,0.36-v/0.1),0) μ_V,Y(v) = max(min(v-0.3/0.1,1),0)

6. High Blood Pressure (HBP) B(μ_B,(b)) μ_B,N(b) = max(min(1,85.05-b/0.1),0) μ_B,H(b) = max(min(b-78.4/0.1,1),0)

7. Trichloroethylene Exposure (TCEE) C(μ_C,(c)) μ_C,L(c) = max(min(1,4.24-c/0.1),0) μ_C,M(c) = max(min(c-3.73/0.1,6.76, 7.01-c/0.1),0) μ_C,H(c) = max(min(c-6.76/0.1,1),0)

8. Cancer stage CS(μ_CS,(cs)) $μ_{cs, s 1} (cs) = {\begin{matrix} 1, cs < 1.88 \\ . \\ \frac{3.43 - cs}{1.55}, 1.88 < cs ⩽ 0.35 \end{matrix}$ $μ_{cs, s 3} (cs) = {\begin{matrix} \frac{cs - 3.435}{1.365} 3.435 < cs ⩽ 4.8 \\ 1 cs = 4.2 \\ \frac{5.5 - cs}{1.3} 4.2 < cs ⩽ 5.5 \end{matrix}$ $μ_{cs, s 3} (cs) = {\begin{matrix} \frac{cs - 5.5}{1.1} 5.5 < cs ⩽ 6.6 \\ 1 cs = 5.7 \\ \frac{7 - cs}{1.3} 5.7 < cs ⩽ 7 \end{matrix}$ $μ_{cs, s 4} (cs) = {\begin{matrix} \frac{cs - 6.59}{1.03} 6.59 < cs ⩽ 7.62 \\ 1 cs ⩾ 7.62 \end{matrix}$

Sr.	Input/Output	Membership function (MF)
1.	Hematuria H(μ_H,(h))	μ_H,1 +(h) = max(min(1,10.91-h/0.1),0) μ_H,2 +(h) = max(min(h6.94/0.1,17.5,25.46-h/0.1),0) μ_H,3 +(h) = max(min(h-17.5/0.1,53.5,57.2-h/0.1),0) μ_H,4 +(h) = max(min(h-53.5/0.1,1),0)
2.	Red Blood cell count (RBCC) R(μ_R,(r))	μ_R,L(r) = max(min(1,0.351-r/0.1),0) μ_R,M(r) = max(min(r-0.306/0.1, 0.52,0.565-r/0.1),0) μ_R,H(r) = max(min(r-0.52/0.1,1),0)
3.	Flank Pain (FP) F(μ_F,(f))	μ_F,L(f) = max(min(1,0.36-f/0.1),0) μ_F,M(f) = max(min(f-0.295/0.1,0.64, 0.69-f/0.1),0) μ_F,H(f) = max(min(f-0.64/0.1,1),0)
4.	Tumor Size (TS) T(μ_T,(t))	μ_T,S(t) = max(min(1,6.94-t/0.1),0) μ_T,L(t) = max(min(t-5.94/0.1,1),0)
5.	Von Hippel-Lindau Gene (VHLG) V(μ_V,(v))	μ_V,N(v) = max(min(1,0.36-v/0.1),0) μ_V,Y(v) = max(min(v-0.3/0.1,1),0)
6.	High Blood Pressure (HBP) B(μ_B,(b))	μ_B,N(b) = max(min(1,85.05-b/0.1),0) μ_B,H(b) = max(min(b-78.4/0.1,1),0)
7.	Trichloroethylene Exposure (TCEE) C(μ_C,(c))	μ_C,L(c) = max(min(1,4.24-c/0.1),0) μ_C,M(c) = max(min(c-3.73/0.1,6.76, 7.01-c/0.1),0) μ_C,H(c) = max(min(c-6.76/0.1,1),0)
8.	Cancer stage CS(μ_CS,(cs))	$μ_{cs, s 1} (cs) = {\begin{matrix} 1, cs < 1.88 \\ . \\ \frac{3.43 - cs}{1.55}, 1.88 < cs ⩽ 0.35 \end{matrix}$ $μ_{cs, s 3} (cs) = {\begin{matrix} \frac{cs - 3.435}{1.365} 3.435 < cs ⩽ 4.8 \\ 1 cs = 4.2 \\ \frac{5.5 - cs}{1.3} 4.2 < cs ⩽ 5.5 \end{matrix}$ $μ_{cs, s 3} (cs) = {\begin{matrix} \frac{cs - 5.5}{1.1} 5.5 < cs ⩽ 6.6 \\ 1 cs = 5.7 \\ \frac{7 - cs}{1.3} 5.7 < cs ⩽ 7 \end{matrix}$ $μ_{cs, s 4} (cs) = {\begin{matrix} \frac{cs - 6.59}{1.03} 6.59 < cs ⩽ 7.62 \\ 1 cs ⩾ 7.62 \end{matrix}$

2.4 Lookup table

The lookup table for the proposed expert system contains the total number of 96 input-output combinations at both layers. Some of the input-output rules are shown in Table 8. This table is constructed with the help of advice from medical experts or doctors from the field of renal cancer.

Table 8
Lookup table for the proposed expert system

Rules Hematuria RBCC Flank pain(FP) Tumor size (TS) VHL gene High blood pressure Trichloroethylene exposure Results

1 1+ Low Low Smaller Yes Normal High Stage 1

2 3+ Normal High Smaller Yes High Low Stage 1

3 2+ Normal Medium Smaller No Normal Moderate Stage 2

4 3+ High Low Smaller Yes High Moderate Stage 2

5 2+ Normal Low Larger Yes High Moderate Stage 3

6 4+ High Low Smaller No Normal Moderate Stage 3

7 4+ High Medium Larger No Normal High Stage 3

8 3+ Normal Low Smaller No High High Stage 3

9 1+ Low Medium Larger No High High Stage 3

10 1+ Normal Low Larger No High High Stage 4

11 2+ Normal High Larger Yes High High Stage 4

12 4+ Normal High Larger Yes High High Stage 4

Rules	Hematuria	RBCC	Flank pain(FP)	Tumor size (TS)	VHL gene	High blood pressure	Trichloroethylene exposure	Results
1	1+	Low	Low	Smaller	Yes	Normal	High	Stage 1
2	3+	Normal	High	Smaller	Yes	High	Low	Stage 1
3	2+	Normal	Medium	Smaller	No	Normal	Moderate	Stage 2
4	3+	High	Low	Smaller	Yes	High	Moderate	Stage 2
5	2+	Normal	Low	Larger	Yes	High	Moderate	Stage 3
6	4+	High	Low	Smaller	No	Normal	Moderate	Stage 3
7	4+	High	Medium	Larger	No	Normal	High	Stage 3
8	3+	Normal	Low	Smaller	No	High	High	Stage 3
9	1+	Low	Medium	Larger	No	High	High	Stage 3
10	1+	Normal	Low	Larger	No	High	High	Stage 4
11	2+	Normal	High	Larger	Yes	High	High	Stage 4
12	4+	Normal	High	Larger	Yes	High	High	Stage 4

2.5 I/O Rules

The input-output rules play a vital role in development of fuzzy inference system (FIS). The performance of every expert system depends on I/O rules. A sample of I/O rules used in the developed expert system are shown in Table 8. Figure 6 and 7 also describe the I/O rules for the proposed multi-layer fuzzy expert system.

Fig. 6

I/O rules for layer 1 of an expert system.

Fig. 7

I/O rules for layer 2 of an expert system.

2.6 Inference Engine

As the inference engine is the core part of every expert system. Hence, in this developed system, the Mamdani inference engine is used at layer 1 and layer 2.

2.7 Defuzzifier

Defuzzifier is also a crucial part of an expert system. There are different types of defuzzifiers used to develop various expert systems. In this research work, the defuzzifer used is the centroid method. Figure 8 shows the graphical view of defuzzifier of layer 1 and in Fig. 9(a)-9(c), the defuzzifier graphical representation of layer 2 expert system is shown.

Figure 8 represents the surface area of rules between two input variables used in layer 1. These two variables are smoking and genetic or hereditary. If the value of smoking is less than the value of another input variable then the result is no cancer. Hence, it is observed that if the value of smoking is less than the probability of having cancer to a patient is also less in that case. Similarly, if the value of the input variable ‘smoking’ is more than the probability of having cancer automatically increases.

Fig. 8

Layer 1 surface area of smoking with the genetic or hereditary symptom.

Figure 9(a) shows the cancer stage based on hematuria and tumor size. It is a three-dimensional rule surface. Two axes of this surface are representing the inputs given to the model and the third one represents the output given by the system corresponding to given inputs. There are a number of colors in the surface region which are representing the stages of renal cancer. It is observed that if the value of tumor size (y-axis) is more than the probability of renal cancer increases with an increase in the size of a tumor.

Fig. 9

Rule surface for (a) Tumor size and Hematuria (b) Tumor size and Trichloroethylene exposure (c) Trichloroethylene exposure and Hematuria.

Similarly, Figs. 9(b) and 9(c) show the surface region of the renal cancer stage based on input variables such as tumor size with trichloroethylene exposure and hematuria with trichloroethylene exposure. As there are four colors in Fig. 9(b), these colors represent the stages of renal cancer. If the tumor size is low and also the inhalation of trichloroethylene is low then there is a blue color which means at stage 1. If the amount of the inhalation of trichloroethylene and the size of the tumor is slightly more then there is a cyan color that depicts the stage 2 of renal cancer. Similarly, the light cyan color and yellow color depicts the stage 3 and stage 4 of renal cancer respectively. Similarly, in Fig. 9(c), there is a 3-dimensional curve that presents the mapping from trichloroethylene exposure and haematuria to the output of the system i.e. stage of renal cancer. Two axes of this curve speak for the inputs and another one is for the stage of renal cancer. These surface regions represent the chance of occurrence by using two input variables from given seven input variables used in layer 2. Hence, the result of stages of renal cancer is based on the combination of basically three input variables.

3 Result

MATLAB R2014a tool is used for this simulation. This tool is basically used for many domains such as the development of algorithms, prototyping and many others. It is also a very powerful tool for programming, visualization, and analysis of data. In this research work, Layer 1 shows the result that either the patient is suffering from cancer or not, and Layer2 shows stage of cancer such as stage 1, stage 2, stage 3 or stage 4. This research work is not only for the detection of renal cancer, but the proposed expert system can also depict different stages of renal cancer.

Eighty tests are captured to scrutinize renal cancer using a multilayer fuzzy inference system. The observed results from the proposed system are compared with the result given by experts or specialists. From this comparison, it was observed that out of 160 tests, 152 tests are correctly classified into the correct class by the developed inference system.

The confusion matrix is shown in Table 9. 160 different cases of renal cancer have been carried out for these tests. In the first column, 40 cases of stage 1 of renal cancer are considered and all of them are correctly classified i.e. stage 1. In the second column, the data of 40 patients of stage 2 of renal cancer is considered, out of which 37 are classified into stage 2 and the rest of 3 are classified into incorrect class i.e. stage 3 of renal cancer. Similarly, in the third column, 40 cases of stage 3 are considered and from these 40 cases, 36 are classified correctly in the class of stage 3 but 4 are classified into the class of stage 4 of renal cancer which is wrong. Likewise, fourth column shows that there are 40 cases of stage 4 of renal cancer, but out of 40, 1 is misclassified and 39 are classified into the correct and exact class of stage 4 of renal cancer.

Table 9
Performance analysis of the proposed expert system

Patient Hematuria RBCC FP TS VHLG HBP TCEE Human expert decision Proposed expert system decision Success

1. 16.3 0.05 0.12 2.64 0.8 120.3 6.20 Stage 1 Stage 1 Yes

2. 5.4 0.16 0.27 3.21 0.63 104.3 5.95 Stage3 Stage3 Yes

3. 23.55 0.37 0.32 6.21 0.32 81.16 3.5 Stage 2 Stage 2 Yes

4. 38.04 0.4 0.35 6.78 0.31 91.3 4.15 Stage 1 Stage 1 Yes

5. 45.29 0.59 0.32 6.35 0.5 110 7.6 Stage 4 Stage 4 Yes

6. 45.29 0.56 0.19 5.64 0.31 102 5.95 Stage 2 Stage 2 Yes

7. 9.05 0.3 0.29 6.21 0.32 81 7.36 Stage 3 Stage 3 Yes

8. 5.43 0.25 0.19 3.6 0.21 59 7.75 Stage 1 Stage 1 Yes

9. 9.05 0.3 0.43 6.35 0.36 82 1.96 Stage 1 Stage 1 Yes

10. 5.43 0.19 0.49 7.07 0.47 82.61 9.16 Stage 3 Stage 3 Yes

11. 9.058 0.27 0.72 6.64 0.39 102 7.49 Stage 4 Stage 1 No

12. 34.4 0.32 0.33 5.64 0.26 72 2.99 Stage 1 Stage 1 Yes

13. 5.43 0.59 0.17 6.6 0.32 76 7.7 Stage 3 Stage 3 Yes

14. 70.65 0.25 0.10 5.78 0.28 75 3.3 Stage 1 Stage 1 Yes

15. 45.29 0.7 0.26 8.92 0.74 97 9.67 Stage 4 Stage 4 Yes

16. 38.04 0.33 0.55 7.64 0.44 91 6.46 Stage 3 Stage 3 Yes

17. 48.91 0.6 0.27 5.5 0.19 82 4.4 Stage 2 Stage 2 Yes

18. 143.1 0.3 0.23 6.35 0.26 131 7.62 Stage 4 Stage 4 Yes

19. 88.77 0.2 0.16 6.64 0.18 98 6.33 Stage 3 Stage 3 Yes

20. 30.8 0.57 0.27 6.35 0.26 71 3.36 Stage 3 Stage 4 No

21. 85.14 0.83 0.13 2.92 0.51 85 2.4 Stage 1 Stage 1 Yes

22. 16.3 0.46 0.22 4.64 0.28 75.3 3.5 Stage 1 Stage 1 Yes

23. 9.06 0.39 0.5 5.92 0.35 69 5.17 Stage 2 Stage 2 Yes

24. 19.93 0.55 0.7 6.92 0.29 91.3 9.55 Stage 4 Stage 4 Yes

25. 74.28 0.83 0.29 5.78 0.5 66.6 4.9 Stage 2 Stage 2 Yes

26. 77.9 0.57 0.25 6.35 0.38 84 3.12 Stage 1 Stage 1 Yes

27. 12.69 0.39 0.55 7.78 0.21 85.5 6.07 Stage 3 Stage 3 Yes

28. 41.67 0.23 0.50 8.07 0.64 91 8.26 Stage 4 Stage 4 Yes

29. 12.68 0.45 0.49 6.7 0.26 98 5.21 Stage 3 Stage 3 Yes

30. 13.5 0.57 0.26 5.78 0.36 66 3.12 Stage 1 Stage 1 Yes

Patient	Hematuria	RBCC	FP	TS	VHLG	HBP	TCEE	Human expert decision	Proposed expert system decision	Success
1.	16.3	0.05	0.12	2.64	0.8	120.3	6.20	Stage 1	Stage 1	Yes
2.	5.4	0.16	0.27	3.21	0.63	104.3	5.95	Stage3	Stage3	Yes
3.	23.55	0.37	0.32	6.21	0.32	81.16	3.5	Stage 2	Stage 2	Yes
4.	38.04	0.4	0.35	6.78	0.31	91.3	4.15	Stage 1	Stage 1	Yes
5.	45.29	0.59	0.32	6.35	0.5	110	7.6	Stage 4	Stage 4	Yes
6.	45.29	0.56	0.19	5.64	0.31	102	5.95	Stage 2	Stage 2	Yes
7.	9.05	0.3	0.29	6.21	0.32	81	7.36	Stage 3	Stage 3	Yes
8.	5.43	0.25	0.19	3.6	0.21	59	7.75	Stage 1	Stage 1	Yes
9.	9.05	0.3	0.43	6.35	0.36	82	1.96	Stage 1	Stage 1	Yes
10.	5.43	0.19	0.49	7.07	0.47	82.61	9.16	Stage 3	Stage 3	Yes
11.	9.058	0.27	0.72	6.64	0.39	102	7.49	Stage 4	Stage 1	No
12.	34.4	0.32	0.33	5.64	0.26	72	2.99	Stage 1	Stage 1	Yes
13.	5.43	0.59	0.17	6.6	0.32	76	7.7	Stage 3	Stage 3	Yes
14.	70.65	0.25	0.10	5.78	0.28	75	3.3	Stage 1	Stage 1	Yes
15.	45.29	0.7	0.26	8.92	0.74	97	9.67	Stage 4	Stage 4	Yes
16.	38.04	0.33	0.55	7.64	0.44	91	6.46	Stage 3	Stage 3	Yes
17.	48.91	0.6	0.27	5.5	0.19	82	4.4	Stage 2	Stage 2	Yes
18.	143.1	0.3	0.23	6.35	0.26	131	7.62	Stage 4	Stage 4	Yes
19.	88.77	0.2	0.16	6.64	0.18	98	6.33	Stage 3	Stage 3	Yes
20.	30.8	0.57	0.27	6.35	0.26	71	3.36	Stage 3	Stage 4	No
21.	85.14	0.83	0.13	2.92	0.51	85	2.4	Stage 1	Stage 1	Yes
22.	16.3	0.46	0.22	4.64	0.28	75.3	3.5	Stage 1	Stage 1	Yes
23.	9.06	0.39	0.5	5.92	0.35	69	5.17	Stage 2	Stage 2	Yes
24.	19.93	0.55	0.7	6.92	0.29	91.3	9.55	Stage 4	Stage 4	Yes
25.	74.28	0.83	0.29	5.78	0.5	66.6	4.9	Stage 2	Stage 2	Yes
26.	77.9	0.57	0.25	6.35	0.38	84	3.12	Stage 1	Stage 1	Yes
27.	12.69	0.39	0.55	7.78	0.21	85.5	6.07	Stage 3	Stage 3	Yes
28.	41.67	0.23	0.50	8.07	0.64	91	8.26	Stage 4	Stage 4	Yes
29.	12.68	0.45	0.49	6.7	0.26	98	5.21	Stage 3	Stage 3	Yes
30.	13.5	0.57	0.26	5.78	0.36	66	3.12	Stage 1	Stage 1	Yes

$\begin{matrix} Confidence indicator \\ = (\frac{Success number}{Total number of tests} * 100) \end{matrix}$ (5)

Where:

Number of successes = 152

Total number of tests = 160

Hence, The confidence indicator for the proposed expert system is 95%.

Now, the first two stages (stage 1 and stage 2) are taken as ‘No’. Similarly, the rest of the two stages are considered as ‘Yes’. Hence, the dimensionality of confusion matrix is reduced and this reduced 2*2 matrix in shown in Table 10.

Table 10

Matrix with reduced dimensionality

No	Yes	Class Name
77	03	No
01	79	Yes

The performance of a multilayered fuzzy inference system for diagnosis of renal cancer is calculated by considering various parameters like accuracy, precision, specificity, and sensitivity.

From Table 11,

Table 11

Confusion matrix for renal cancer

Stage 1	Stage 2	Stage 3	Stage 4	Class Name
40	00	00	00	Stage 1
00	37	03	00	Stage 2
00	00	36	04	Stage 3
01	00	00	39	Stage 4

TP: True Positive is 79

FN: False Negative is 01

FP: False positive is 03

TN: True Negative is 77 $Sensitivity = \frac{(TP)}{(TP + FN)} = \frac{79}{79 + 01} = 98.75$ $Specificity = \frac{(TN)}{(TN + FP)} = \frac{77}{77 + 03} = 96.25$ $Precision = \frac{(TP)}{(TP + FP)} = \frac{79}{79 + 03} = 96.34$ $\begin{matrix} Classification Accuracy \\ = \frac{(TP + TN)}{(TP + FP + TN + FN)} \\ = \frac{79 + 77}{79 + 03 + 77 + 01} = 97.5 \end{matrix}$ These parameters are also shown in the form of the graph in Fig. 10.

Fig. 10

Performance parameters of proposed system.

Figure 11(a) and 11(b) show the sample of output obtained from developed expert system for renal or kidney cancer.

Fig. 11

Layer 1 lookup diagram for (a) no cancer (b) presence of renal cancer.

Figure 11(a) shows that if all the input variables have normal range, then there is no renal cancer. Figure 11(b) shows if the range of input variables is more then normal then there are chances of occurrence of cancer in renal due to smoking, dialysis and other input variables used in layer 1.

4 Conclusion and future work

The main aim of this research is to develop an expert system that can be able to diagnose renal cancer. The proposed expert system fulfills the primary objective and also this expert system is easy to use for medical and other professionals. An unexperienced person can also use this expert system to detect the stage of renal cancer by giving the required inputs. The confidence indicator achieved from this expert system is 95%.

The main limitation of this developed model is that the updation or change in rules is very difficult. The system is not able to learn by its experience and it is also not able to adapt itself according to the environment.

Therefore, in the future, performance and productivity can be improved by using other techniques like neural network and neuro-fuzzy expert system. This research can also be continued by adding more risk factors or input variables and can be used in those areas where there is a lack of resources.

References

Batra

, Batra

and Singh

, A review to predictive methodology to diagnose chronic kidney disease. In 2016 3rd International Conference on Computing for Sustainable Global Development (INDIACom) (2016), (pp. 2760-2763). IEEE.

Sobrinho

, da Silva

L.D.

, Pinheiro

M.E.

, Cunha

, Perkusich

and Medeiros

, Formal specification of a tool to aid the early diagnosis of the Chronic Kidney Disease. In 2015 CHILEAN Conference on Electrical, Electronics Engineering, Information and Communication Technologies (CHILECON) (2015), (pp. 173-178). IEEE.

Vasilescu

G.M.

, Cauni

, Kacso

, Stănculescu

, Marin

M.E.

, Maricaru

and Hănţilă

I.F.

, Issues on early cancer detection using thermographic methods. In 2017 10th International Symposium on Advanced Topics in Electrical Engineering (ATEE) (2017), (pp. 283-286). IEEE.

Chow

W.H.

, Dong

L.M.

and Devesa

S.S.

, Epidemiology and risk factors for kidney cancer, Nature Reviews Urology 7(5) (2010), 245.

Basappa

N.S.

, Basiuk

, Canil

, Al-Asaaed

, Heng

D.Y.

, Wood...

and Jewett

M.A.

, Setting research priorities for kidney cancer, European Urology 72 (2017), 861–864.

Borisova

, Mantareva

, Bliznakova

, Angelov

, Avramov

and Pavlova

, Photodiagnosis and photodynamic therapy of cutaneous melanoma, Current Management of Malignant Melanoma (2011), 141–156.

Signorini

and Greussay

, Silent Killing: an Object-Oriented View of Hypertension and Kidney Failures Part I. In 2007 IEEE 7th International Symposium on BioInformatics and BioEngineering (2007), (pp. 635–640). IEEE.

De Leon

A.D.

, Kapur

and Pedrosa

, Radiomics in kidney Cancer: MR imaging, Magnetic Resonance Imaging Clinics 27(1) (2019), 1–13.

Deng

S.P.

, Cao

, Huang

D.S.

and Wang

Y.P.

, Identifying stages of kidney renal cell carcinoma by combining gene expression and DNA methylation data, IEEE/ACM Transactions on Computational Biology and Bioinformatics 14(5) (2016), 1147–1153.

10.

Weiss

R.H.

, Metabolomics and metabolic reprogramming in kidney cancer. In Seminars in nephrology (2018), (Vol. 38, No. 2, pp. 175–182). WB Saunders.

11.

Haas

N.B.

and Nathanson

K.L.

, Hereditary kidney cancer syndromes, Advances in Chronic Kidney Disease 21(1) (2014), 81–90.

12.

Tander

, Özmen

and Özden

, Neural network design for the recurrence prediction of post-operative non-metastatic kidney cancer patients. In 2015 9th International Conference on Electrical and Electronics Engineering (ELECO) (2015, November), (pp. 162–165). IEEE.

13.

Golodetz

, Voiculescu

and Cameron

, A proposed decision-support systemfor (renal) cancer imaging. In 2007 Frontiers in the Convergence of Bioscience and Information Technologies (2007, October), (pp. 361–366). IEEE.

14.

Tuncer

S.A.

and Alkan

, A decision support system for detection of the renal cell cancer in the kidney, Measurement 123 (2018), 298–303.

15.

Sridharan

, Kasoji

, Chang

and Dayton

P.A.

, Contrast-enhanced ultrasound (CEUS) in patients with chronic kidney disease (CKD). In 2017 IEEE International Ultrasonics Symposium (IUS) (2017), (pp. 1–4). IEEE.

16.

Kothari

, Phan

J.H.

, Young

A.N.

and Wang

M.D.

, Histological image feature mining reveals emergent diagnostic properties for renal cancer. In 2011 IEEE International Conference on Bioinformatics and Biomedicine (2011), (pp. 422–425). IEEE.

17.

Charleonnan

, Fufaung

, Niyomwong

, Chokchueypattanakit

, Suwannawach

and Ninchawee

, Predictive analytics for chronic kidney disease using machine learning techniques. In 2016 Management and Innovation Technology International Conference (MITicon) (2016), (pp. MIT-80). IEEE.

18.

Linguraru

M.G.

, Wang

, Shah

, Gautam

, Peterson

, Linehan

W.M.

and Summers

R.M.

, Computer-aided renal cancer quantification and classification from contrast-enhanced CT via histograms of curvature-related features. In 2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society (2009), (pp. 6679–6682). IEEE.

19.

Mdzinarishvili

, Gleason

M.X.

and Sherman

, Influence of Time Period and Birth Cohort Effects on Age-Specific Incidence Rates of Pancreatic and Kidney Cancer. In 2010 43rd Hawaii International Conference on System Sciences (2010), (pp. 1–9). IEEE.

20.

Saribudak

, Dong

, Gundry

, Hsieh

and Uyar

M.Ü.

, Mathematical models of tumor growth using Voronoi tessellations in pathology slides of kidney cancer. In 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (2015), (pp. 4454–4457). IEEE.

21.

Panuszewska

, Minch

and Dzwinel

, Particle automata model of renal cancer progression, 2018 International Interdisciplinary PhD Workshop (IIPhDW), (2018), pp. 254–256.

22.

Meng

, Sheng

, Lv

, Sun

and Wu

, Differential geometry images: remeshing and morphing with local shape preservation, The Visual Computer 26(1) (2010), 51.

23.

Abhilash

R.H.

and Chauhan

, Respiration-induced movement correlation for synchronous noninvasive renal cancer surgery, IEEE transactions on ultrasonics, ferroelectrics, and frequency control 59(7) (2012), 1478–1486.

24.

Grossi

, How artificial intelligence tools can be used to assess individual patient risk in cardiovascular disease: problems with the current methods, BMC Cardiovascular Disorders 6(1) (2006), 20.

25.

Kazemi

and Mirroshandel

S.A.

, A novel method for predicting kidney stone type using ensemble learning, Artificial Intelligence in Medicine 84 (2018), 117–126.

26.

Ahmad

, Khan

M.A.

, Abbas

, Athar

, Khan

B.S.

and Aslam

M.S.

, Automated diagnosis of hepatitis b using multilayer mamdani fuzzy inference system, Journal of Healthcare Engineering, 2019.

27.

Ahmadi

, Gholamzadeh

, Shahmoradi

, Nilashi

and Rashvand

, Diseases diagnosis using fuzzy logic methods: A systematic and meta-analysis review, Computer Methods and Programs in Biomedicine 161 (2018), 145–172.

28.

Sivathasan

, Cecelja

and Balachandran

, ECG Diagnosis using neural network and fuzzy expert system. In Proceedings of the 17th IEEE Instrumentation and Measurement Technology Conference [Cat. No. 00CH37066] (2000), (Vol. 2, pp. 988–992). IEEE.

29.

Arqub

O.A.

, Adaptation of reproducing kernel algorithm for solving fuzzy Fredholm–Volterra integrodifferential equations, Neural Computing and Applications 28(7) (2017), 1591–1610.

30.

Yang

and Wang

, Robust adaptive fault-tolerant control for uncertain nonlinear system with unmodeled dynamics based on fuzzy approximation, Neurocomputing 173 (2016), 1660–1670.

31.

Arqub

O.A.

, Mohammed

A.S.

, Momani

and Hayat

, Numerical solutions of fuzzy differential equations using reproducing kernel Hilbert space method, Soft Computing 20(8) (2016), 3283–3302.

32.

Arqub

O.A.

, Al-Smadi

, Momani

and Hayat

, Application of reproducing kernel algorithm for solving second-order, two-point fuzzy boundary value problems, Soft Computing 21(23) (2017), 7191–7206.

33.

Novák

, Which logic is the real fuzzy logic? Fuzzy Sets and Systems 157(5) (2006), 635–641.

34.

Roventa

and Rosu

, The diagnosis of some kidney diseases in a small prolog Expert System. In 2009 3rd International Workshop on Soft Computing Applications (2009), (pp. 219–224). IEEE.

Development of multilayer fuzzy inference system for diagnosis of renal cancer

Abstract

Keywords

1 Introduction

2 Method

Table 6 Output variables of Layer 1 and Layer 2 Sr. no. Output variables Semantic sign 1. Layer 1 Renal cancer Yes No 2. Layer 2 Cancer Stage Stage 1 Stage 2 Stage 3 Stage 4

2.7 Defuzzifier

References

Table 6
Output variables of Layer 1 and Layer 2

Sr. no. Output variables Semantic sign

1. Layer 1 Renal cancer Yes

No

2. Layer 2 Cancer Stage Stage 1

Stage 2

Stage 3

Stage 4