Design a FPGA,fuzzy based,insolent method for prediction of multi-diseases in rural area

Abstract

Nowadays there are lots of fatal diseases are growing at a rapid rate. We consider about four primary diseases like jaundice, diabetes mellitus, yellow fever, and cholera. In this paper, we design a novel method with the help of fuzzy and FPGA system for prediction multiple diseases in a rural area. Association rule mining technique helps to define fuzzy rules, which implemented on both Spartan3-E and Artix-7 FPGA kit. Due to this implementation, it is easier to design a cost-effective and portable system for multi-disease prediction. The innovation lies in design a low power FPGA and meticulousness method for identification, prediction of four fatal diseases. The whole plan has tested on Xilinx and Cadence tool for generating RTL model and Layout design.

Keywords

Fuzzy logic medical diagnosis association rule mining Artix-7 FPGA

1 Introduction

During the last lustrum, medical field beheld a remarkable development in research [1, 2]. Various institutes and research laboratories formed an inestimable amount of data. Need to develop more accessible, effectual techniques and approaches to analyze and epitomize the biomedical data. Analysis of past data developed new theories, and these things are getting when made literature survey. The inevitability to isolate and process this wide range of data has helped to the research and development in Data Mining [3, 4].

Now a day fatal diseases are capital of the world. These diseases are increasing in younger group with a significant threat to both male and female. We consider four of them named jaundice, diabetes mellitus, yellow fever and cholera, in which rural area people are mostly suffering [5, 6]. Disease caused due to less function of the liver and produced high bilirubin level in human body. The symptoms are yellowish skin, liver and kidney; brain tends to damage. From survey report in India, every one person from ten is suffering from diabetes mellitus. It is a metabolic disorder having signs of weight loss, thirst, heart stroke, kidney disease. Yellow fever is a viral disease. Its symptoms are fever, chills, abdominal pain, chances of liver damage, kidney problem [7].

Lotfi Zadeh was introduced fuzzy set theory for solving the problem like yes or no, true or false [8]. But the recent trend, fuzzy logic is a heart of problem-solving in control system problem and meticulously used in critical decision -making problems [9]. Try to implement fuzzy logic for defining all parameters of four diseases like jaundice, diabetes mellitus, yellow fever and cholera and predict it before a patient will critically suffer.

This paper is systematized as follows: Section 2, 3 pronounces literature survey and the dataset respectively, which are used for design the predicted method, the detailed analysis of four diseases and what constraints we consider. Section 4 describes, fuzzy rules define base on association rule mining, passes the constraint value through weighted sum algorithm and comparing the predicted value with doctors original report. Section 5 and 6 states FPGA implementation, power consumption comparison and layout design.

2 Literature survey

Here are many works done by Chowdhury and Das on this research topic.

Chowdhury et al. 2008 designed a hardware system, which has the capabilities of medical diagnosis. For fast design cycle, the method implemented in an FPGA chip. The whole system realized on Altera cyclone FPGA board. The main aim behind the design is to develop an inexpensive, portable system for rural area application [10].

Chowdhury et al. 2008 develop an FPGA based system for Spiro-metric data application. In this system smart agent record data of the patient. Parallel data processing architecture is using here for fast the computational speed. The whole system is tested on Altera cyclone EP1K6Q240C8 FPGAchip [11].

Das et al. 2010 design a decision making the fuzzy expert system and its application in a medical diagnostic system. This attempt improves the accuracy of fuzzy expert decision-making system, using type-2 fuzzy set. Improve decision-making accuracy verified by medical diagnostic decision-making system for the renal diagnostic application [12].

Chowdhury et al. 2011 design ASIC digital chip of fuzzy logic circuit for diagnostic applications. The system on the chip used fuzzification, defuzzification of membership function to make the output decision [13].

Das et al. 2012 try to increase the accuracy of the fuzzy expert decision- making system by changing the parameters of the type-2 membership function. Accuracy estimation applied to physiological parameters like body mass index (BMI), glucose, urea, creatinine, systolic, diastolic pressure. Type -1 fuzzy set is used to develop an FPGA based smart processor [14].

3 Material and method

From survey report, it observed that 69% of the total residents are in rural area and semi-rural area, but 25% of full skilled doctors are present there. These things make doctor-patient ratio 1 : 4000, but in the urban area, its rate improves and comes to 1 : 640 [15]. Above statistical data clarify, in India availability of doctor is less and experience, qualified doctors are few. We design a fuzzy and FPGA based method for detection and prediction of fatal disease. Analysis purpose takes the medical data from Pima Indians, Physionet, world health organization (WHO) [16, 17].

The Fig. 1 consists of four main disease name Jaundice, diabetes mellitus, yellow fever, cholera, in which rural area people are regularly suffering. We need some constrains of the given disease for its prediction. Jaundice constrains are sbt, sbd, sgpt, Sgot and urine. Diabetes mellitus constrains fasting sugar level, sugar level after taking food, alpha hemoglobin. Yellow fever parameters are IgG, IgM. Cholera disease constrain is stool sample.

Fig.1

Disease with its parameters.

Figure 2. Explain about the disease prediction methodology, starting from medical database collection to layout design leading to product level development. First collect the medical data of disease as mention above. Pass the constraints of illness to the fuzzy system in which association rule mining defines fuzzy rules. Prediction is calculating by weighted sum fuzzy algorithm. Compare the weighted sum algorithm result with the original doctor’s report. If both are not match up to 95% then again initiate from starting point. Otherwise, implement on Artix-7 and Spartan3-e FPGA kit. Compare the power consumption of both FPGA kit and choose the better one. Leading to product level design, make the layout of the whole system by cadence tool.

Fig.2

Flow chart on multi- disease prediction methodology.

4 Define fuzzy rules

There are five fuzzy rules; based on association rule mining and statistical analysis data for enhanced forecasts of the diseases and fatal cases [18]. Parameters for disease prediction are coming from the blood, serum, stool and urine test. Jaundice constraints are sbt (serum bilirubin total), Sbd (serum bilirubin direct), Sgpt (serum glutamic pyruvic transaminase), sgot (serum glutamic-oxaloacetic transaminase) and urine colour test. Number ranging from sbt (0-15), Sbd (0-15), Sgpt (0-400), Sgot (0-400) and urine (0-10) and output value (0-10) [19]. Diabetes mellitus constrains fasting sugar level, sugar level after taking food, alpha hemoglobin level (HbA1c). Its value ranging from fasting sugar level(0-200), sugar level after taking food (0-200), alpha hemoglobin level (0-10). Yellow fever is the third disease that we consider for prediction having parameters Immunoglobulin-G (IgG), Immunoglobulin-M (IgM).Its constraints value range from IgG (231-1600), IgM (0-230). Cholera is a viral effected disease. To constrain we put the sample to thiosulfate citrate bile salts sucrose (TCBS) solution. If the virus will form the yellow colony, then it is the sign of cholera disease. Below fuzzy rules are defined with the help of association rule mining.

4.1 Association rules

An association rule R has the form P ⇒ Q, for P, Q ⊆ Z, where Z is a set of all Element and P and Q are element sets. If frequency (P) denotes the number of Transactions that are supersets of item set P, and N the number of all relations, then Support (p) = frequency (P)/N. Each rule is associated with its confidence and support: $Confidance (R) = \frac{frequency (P \cup Q)}{frequency (p)}$ (1) $Support (R) = \frac{frequency (P \cup Q)}{N}$ (2)

Above one is a popular method for discovering a relationship between two variables. Here we consider total 1000 patients data. Take an example, from this database and original medical report, and we got that the critical condition of jaundice disease is occurring when sbt is primary, and sgpt is critical. The occurrence of above statement is 150 times, sbt comes 250 times out of 1000 data analysis. The confidence and support are finding 0.6, 0.15 respectively. The value of confidence is 0.6 which is closer to 1. It means more chance of a patient to suffer in critical condition of jaundice disease if both sbt (primary) and sgpt (critical) state are satisfying. Below we write five fuzzy rules according to association rulemining.

R1: If (sbt is high) and (sbd is high) and (sgpt is high) and (sgot is high) and (urine is high (reddish)) then (jaundice output is critical).

R2: If (sbt is moderate) and (sbd is moderate) and (sgpt is moderate) and (sgot is moderate) and (urine is high (yellow)) then (jaundice output is primary).

R3: If (sbt is low) and (sbd is low) and (sgpt is low) and (sgot is low) and (urine is low (light_yellow)) then (jaundice output is normal).

R4: If (sbt is low) and (sbd is low) and (sgpt is moderate) and (sgot is moderate) and (urine is low (light_yellow)) then (jaundice_output is primary).

R5: If (sbt is moderate) and (sbd is moderate) and (sgpt is high) and (sgot is high) and (urine is high (yellowish)) then (jaundice output is critical) [9].

R6: If (fasting sugar is > 110 mg/dl) and (alpha hemoglobin > 6 dcct) then (Diabetes Mellitus output is primary).

R7: If (fasting sugar is < 110 mg/dl) and (alpha hemoglobin < 6 dcct) then (Diabetes Mellitus output is normal).

R8: If (sugar after taking food > 140 mg/dl) and (alpha hemoglobin > 6 dcct) then (Diabetes Mellitus output is primary).

R9: If (sugar is > 200 mg/dl) and (alpha hemoglobin > 6.5 dcct) then (Diabetes Mellitus output is critical).

R10: If (age 0-19) and (IgG = 231-1584 mg/dl) and (IgM = 0-259 mg/dl) then (Yellow Fever output is primary).

R11: If (age 0-19) and (IgG = 231-1000 mg/dl) and (IgM = 0-180 mg/dl) then (Yellow Fever output is normal).

R12: If (age > 19) and (IgG = 700-1600 mg/dl) and (IgM = 40-230 mg/dl) then (Yellow Fever output is primary).

R13: If (age > 19) and (IgG >1600 mg/dl) and (IgM >230 mg/dl) then (Yellow Fever output is critical).

R14: If (sample+TCBS = yellow colony) then (Cholera output is critical).

R15: If (sample+TCBS=green colony) then (Cholera output is normal).

R16: If (sample+TCBS=greenish yellow colony) then (Cholera output is primary)

Figure 3 denotes different medical constraints and its outcomes. Sixteen Fuzzy rules defined by the help of association rule mining. Out of which three states are proof above test bench bypassing parameters value having given the range. In test bench simulation at the output, we used three different colour like green, yellow and red. These colour are green, yellow and red represent for the normal, primary and critical state of diseases. Simulation result for 0 to 100ns input constraints sbt, sbd, sgpt, sgot, urine values are 3, 3, 78, 86, 3 respectively. This input values satisfied the fuzzy rule R2, and we got the primary condition jaundice disease as output. Similarly fasting sugar = 130, alphahe3moglobin=7, which contented the fuzzy rule R6 and predict as the primary condition of diabetes. The critical state of yellow fever is satisfied by rule 12 for which input parameters are age = 20, IgG = 1650, IgM = 240. For cholera, disease prediction gets the parameter TCBS = 7, which combine with the stool sample an gives greenish yellow colony and it satisfied rule 16 having predicted primary condition of given disease.

Fig.3

Test bench generated by xilinx tool for different value of medical constraints.

Fig.4

Schematic diagram of fuzzy rules having thirteen inputs and twelve outputs.

Fig.5

Hardware implementation using Artix-7 kit.

4.2 RTL model using Xilinx EDA tool

Vivado is an EDA tool, which labels the schematics demonstration of the fuzzy system. Schematic display describes visuals symbol rather than original pictures [20]. In Figure. Four left-hand side shows thirteen input name sbt, sbd, sgpt, sgot, urine, fasting sugar, alpha hemoglobin, sugar after food, sugar, age, IgG, IgM, TCBS. Corresponding twelve outputs which are normalJ, primaryJ, criticalJ, normalD, primaryD, criticalD, normalY, primaryY, criticalY, normalC, primaryC, criticalC shown on the right-hand side. RTL modeling of the fuzzy system is making by digital component like universal logic gates, multiplexers, flip-flop and adder. In between input and output, these components are present. Fuzzification and defuzzification are two essential parts of the system. In fuzzification process, input values are converting into crisp value, and defuzzification process the outcomes values are again converting into original scalar value. Through this scalar output value, predict the disease on which a patient is suffering [11].

4.3 Weighted Sum Algorithm for medical diagnosis

The algorithm for analysis the weighted mean of the membership functions of the patient’s past medical data. It is helpful to predict the disease and its medical condition will normal or primary or critical is computed as $P_{r} (y) = \frac{\sum_{i = 1}^{n} i μ_{i} (y)}{\sum_{i = 1}^{n} i}$ (3)

Where the summation is done from i = 1 to n, and the value of n is the total of time at which clinical data are taken into consideration. Help of membership function it define normal, primary and critical value of medical condition. The predicted value is considered as P(y) [16], which is defined below equation.

5 Experiments and results:

Thirteen input constraints are taken two consecutive days for getting the more accurate prediction of disease. Membership functions, and are referred to fuzzy value having low, moderate and high respectively. These membership functions are calculating bypassing the constraints.

The scrutiny of membership function will finalize the prediction of risk parameter normal, primary, critical. According to the possibility parameter, the probable next physiological state of a patient will find out.

In Xilinx software creates a code that permits us to provide a repeatable set of inputs. It consists of a clock (for synchronization if needed), input data, and output. Test bench check our program is working correctly or not by passing some input constraint and got desired output. In Figs three and four, we consider thirteen different inputs and twelve probable outputs which associate with sixteen different fuzzy rules based on association rule mining. Out of sixteen states, five rules are defining for normal condition of jaundice, diabetes mellitus, yellow fever and cholera. Five rules are for critical, and six rules are for the primary state of abovedisease [21].

Table 1
Comparison of Spartan-3e and artix-7

SPARTAN3-E ARTIX-7

No Logic Utilization No of used Available Utilization No of used Available

1 Number of slice flip-flop 18 9312 1% 35 63400

2 Number of 4 I/P LUTS 242 9312 2.6% 184 9312

3 Number of I/O blocks 180 232 77% 63 210

4 Power Consumption 0.097 w 0.042w

	SPARTAN3-E	ARTIX-7
1	Number of slice flip-flop	18	9312	1%	35	63400
2	Number of 4 I/P LUTS	242	9312	2.6%	184	9312
3	Number of I/O blocks	180	232	77%	63	210
4	Power Consumption	0.097 w	0.042w

5.1 Layout of implemented module:

Layout design of the fuzzy system is drawing in Fig. 6 using Cadence Innovus tool [21]. The speed-enhancing architecture having featured on reduce repetition and provide the runtime boost. This technology has pipeline architecture which helps to build, integrate lots of application and system. Development time and reconfigurable cost of the system with very less.

Fig.6

Layout of fuzzy rules module.

6 Comparison

Check the error between fuzzy system outcomes and doctors’ real decision. As fuzzy system outcomes are 95.23% match, and then we proceed for FPGA implementation. The fuzzy system is implemented on FPGA using Verilog coding on Artix-7 and Spartan3-E kit. Comparing both the package according to resources utilisation and power consumption Artix-7 is found most efficient than spartan-3e. For same module implementation, consumption using Artix-7 it is 0.042 watt, and spartan3e is 0.097 watt.

7 Conclusion

This paper deals with two measure aims. One is detection and prediction of multi-disease like jaundice, diabetes mellitus, yellow fever and cholera with more accuracy. The second one is low power FPGA operation, leading to chip and product level design. After collecting the blood, urine, stool sample from a person five liver tests, sugar, hemoglobin, TCBS agar to stool test are done. The data are delivered, through the fuzzy system that analyses and predict the chance of a person suffering from above four diseases or not. As our fuzzy system outcomes are 95% matched with doctors’ real medical report. Hardware implementation is doing by dump the Verilog code on Artix-7 and Spartan3-e kit, which accomplishes Artix-7 are efficient regarding power consumption and resource utilization. Last part is to design layout of the whole fuzzy system, which leads to chip design and product level design.

References

Zhu

, Liu

A.A.

, Chen

, Tasdizen

and Su

, “Special Issue on Biomedical Big Data: Understanding, Learning and Applications,”, IEEE Transactions on Big Data 3(4) (2017), 375–377.

Jovic

, Kukolja

, Jozic

and Horvat

, “A web platform for analysis of multivariate heterogeneous biomedical time-series — A preliminary report,” 2016 International Conference on Systems, Signals and Image Processing (IWSSIP), Bratislava, 2016, pp. 1–4.

Leandrou

, Petroudi

, Reyes-Aldasoro

C.C.

, Kyriacou

P.A.

and Pattichis

C.S.

, “Quantitative MRI Brain Studies in Mild Cognitive Impairment and Alzheimer’s disease: A Methodological Review,”, IEEE Reviews in Biomedical Engineering PP(99), 1–1.

Venkatesan

, Karthigaikumar

, Paul

, Satheeskumaran

and Kumar

, “Ecg Signal Preprocessing And Svm Classifier Based Abnormality Detection In Remote Healthcare Applications,”, IEEE Access PP(99), 1–1.

Mark,

Patrick B.

, Winocour

Peter

, and Day

Caroline

. “Management of lipids in adults with diabetes mellitus and nephropathy and/or chronic kidney disease: Summary of joint guidance from the Association of British Clinical Diabetologists (ABCD) and the Renal Association (RA).”, British Journal of Diabetes 17.2 (2017), 64–72.

Mane

V.M

, Jadhav

D V

, Shirbahadurkar

S D

, “Hybrid classifier and region-dependent integrated features for detection of diabetic retinopathy”, Journal of Intelligent & Fuzzy Systems 32(4) (2017), 2837–2845.

Sun

, Lu

, Zhang

, Salathé

and Cao

, “Infectious Disease Containment Based on a Wireless Sensor System,”, IEEE Access 4 (2016), 1558–1569.

Zadeh

L.A.

, “Fuzzy logic computing with words,”, IEEE Transactions on Fuzzy Systems 4(2) (1996), 103–111.

Zahedi

K.A

, Salleh

K. A

, Razak

, “Application of a preference relationship in decision-making based on intuitionistic fuzzy soft sets”, Journal of Intelligent & Fuzzy Systems 34(1) (2018), 123–139.

10.

RoyChowdhury

, Chakrabarti

, Saha

, “FPGA realization of a Smart Processing System for Clinical Diagnostic Applications using Pipelined Datapath Architectures”, Microprocessors and Microsystems 32(2) (2008), 107–120.

11.

Roy Chowdhury

, Chakrabarti

Saha

, “Development of an FPGA based smart diagnostic system for Spirometric Data Processing Applications”, International Journal of Smart Sensing and Intelligent Systems 1(4) (2008), 985–1018.

12.

Roy Chowdhury

and Saha

, “Development of an FPGA based Fuzzy Neural Network System for early diagnosis of critical health condition of a patient”, Computers in Biology and Medicine 40(2) (2010), 190–200.

13.

Roy Chowdhury

, Roy

and Saha

, “ASIC design of a digital fuzzy system on chip for medical diagnostic applications”, Journal of Medical Systems 35(2) (2011), 221–235.

14.

Das

, Roy Chowdhury

and Saha

, “Accuracy enhancement in a fuzzy expert decision making system through optimization of membership functions and its application in a medical diagnostic decision making system”, Journal of Medical Systems, Journal of Medical Systems 36(3) (2012), 1607–1620.

15.

Deo

Madhav G

, Doctor population ratio for India – The reality, Indian Journal of Medical Research 137(4) (2013), 632–635.

16.

Diabetes mellitus disease and other fatal disease database https://archive.ics.uci.edu/ml/datasets/pima+indians+diabetes.

17.

Clinical data base for several disease like heart, respiratory, ECG https://physionet.org/physiobank/database/#clinical.

18.

Hong,

Tzung-Pei

, Lin

Kuei-Ying

and Wang

Shyue-Liang

, “Fuzzy data mining for interesting generalized association rules.”, Fuzzy sets and systems 138.2 (2003), 255–269.

19.

Mitra

and Hayashi

, “Neuro-fuzzy rule generation: Survey in soft computing framework,”, IEEE Transactions on Neural Networks 11(3) (2000), 748–768.

20.

Xilinx Technology Xilinx, Matlab combine applica-tion on dsp https://in.mathworks.com/solutions/fpga-design/simulink-with-xilinx-system-generator-for-dsp.html.

21.

Hunt

and Rowson

J.A.

, “Blocking in a system on a chip,”, IEEE Spectrum 33(11) (1996), 35–41.

	SPARTAN3-E				ARTIX-7
No	Logic Utilization	No of used	Available	Utilization	No of used	Available
1	Number of slice flip-flop	18	9312	1%	35	63400
2	Number of 4 I/P LUTS	242	9312	2.6%	184	9312
3	Number of I/O blocks	180	232	77%	63	210
4	Power Consumption	0.097 w	0.042w