E-Anfis To Diagnose The Progression Of Chronic Kidney

2y ago
27 Views
2 Downloads
715.32 KB
10 Pages
Last View : Today
Last Download : 2m ago
Upload by : Harley Spears
Transcription

RESEARCHE-Anfis to diagnose theprogression of chronic kidneydiseaseAbstractChronic renal failure is not well explored. In this study, an artificial intelligence technique is proposed for overcoming theoccurrence of local minima and local maxima in diagnosing the progression of kidney disease. An AI technique, a mixtureof ALO and ANFIS, E-ANFIS (Enhanced Adaptive Neuro-fuzzy Inference Systems) is introduced. Normally back propagationis used in ANFIS but in proposed using new optimizer ALO. The performance of ANFIS is improved by utilizing the AntLion Optimizer. This enhanced ANFIS used to diagnose the progression stage of the CKD. The proposed technique wasexecuted in MATLAB/Simulink platform and compared with the existing techniques ANFIS, fuzzy, and ANN. Performanceevaluation is assessed in terms of accuracy, recall, precision, F-measure and specificity. The obtained results showed thatthe newly introduced E-ANFIS is the best algorithm when compared to other involved existing algorithms.Keywords: Chronic renal disease, data mining, GFR, ant lion optimizer, adaptive neuro-fuzzy inference system,microalbuminuria, fuzzy, artificial neural network, E-ANFISAbbreviationsCKD: Chronic Kidney Disease; GFR:Glomerular Filtration Rate; ALO-Ant LionOptimizer; ANFIS: Adaptive NeurofuzzyInference System; E-ANFIS: EnhancedAdaptive Neurofuzzy Inference System; CRF:Chronic Renal Failure; BMI: Body Mass Index;MDRD: Modification of Diet in Renal Disease;ANN: Artificial Neural NetworkIntroductionChronic Kidney Disease is a decline in kidneyfunction due to any type of diabetes mellitus,abnormal blood pressure, glomerulonephritis,congenital abnormalities in the kidneys, orgenetic reasons [1]. CKD leads to a lack ofremoving wastes and extra fluids from the body.The hormone level is imbalanced in the bodyand not able to maintain the body’s balance ofacid and base. According to the severity of thedisease, CRF is classified into five stages basedon the Glomerular Filtration Rate (GFR). Stage1 represents kidney damage with normal orincreased GFR and final stage 5 referred to askidney failure wherein there is a total loss ofkidney function. At this stage, most of the peopleneed dialysis [1,2]. The problems relevant toCRF may happen gradually, over a long period1235of time without symptoms and they may end atend-stage renal failure. Early identification andtreatment are therefore useful in preventing theprogression of the disease. The kidney diseaseprogression considered as a function of variousfactors including GFR, urine microalbumin,serum sodium, serum potassium, serum uricacid, blood urea, total protein, serum albumin[1-4]. Among these, microalbuminuria (30300 mg/day) is an earlier sign of chronic kidneydisease [5].Subhashini R1* andMK Jeyakumar2Department of Computer Applications,Noorul Islam Centre for HigherEducation, Kumaracoil, India2Additional Controller of Examination,Noorul Islam Centre for HigherEducation, Kumaracoil, India1*Author for correspondence:baskisubha24@gmail.comIn recent years, early diagnosis of thedisease determines the appropriate time toapply medical treatments for CKD receivedgreat attention among physicians. Researchersthrough studies try to diagnose CKD in patientsas early as possible and to control the riskfactors of the disease progression like high bloodpressure, proteinuria, and hyperphosphatemia[6,7]. Based on the evaluation, different modelswere developed to predict progression. However,they cannot accurately predict the variations ofGFR [8].This paper is further organized as: Section2 represents related work, Section 3 providesmaterials and methods, Section 4 is modeling ofE-ANFIS technique based on ant lion optimizer,Section 5 is the result and discussion, and finally,Section 6 provides the conclusion.Clin. Pract. (2019) 16(5), 1235-1244ISSN 2044-9038

RESEARCHSubhashini & Jeyakumar Survey of related research worksAguilar et al. analyzed the factors associatedwith CKD on 105 patients. The CKD relatedfactors were age more than 65, sex, the presenceof cardiovascular disease, anemia, and overweightwith BMI 30. Their work showed that age andanemia both were the strongest factors relevantto CKD [9]. There has been more number ofstudies on GFR variations among differentCKD patients. Artificial Intelligent and machinelearning techniques have been increasingly usedin disease forecasting.Gaspari et al. derived from 12 predictionequations by plasma iohexol clearance in a groupof 91 renal transplant patients [10]. They foundthat all models overestimate renal function.Brier et al. compared neural networks withlogistic regression in the prediction of DelayedGraft Function (DGF) in renal transplantpatients. They evaluated the results of a neuralnetwork with logistic regression and foundedhigher sensitivity of logistic regression in theprediction of DGF (91 versus 80%), the neuralnetwork was sensitive to the prediction for DGF(66 versus 47%) [11].Hussain et al. given a tool for detectingcancer using Support Vector Machines (SVM).They evaluated the performance of the newmethod with the remaining classificationmethods. Accordingly, SVM improved in itsperformance [12].Recently, fuzzy methods, especially expertsystems have been increasingly used in predictionof diseases. It seems like employing this methodwith clinical tools for diagnosis of diseases andthe condition may reduce diagnostic errors.Fuzzy inference technique is accurate. ANFISis based on neural networks concepts. ANFISnetwork is proposed by Ojugo, et al. [13]. Thisis a network equivalent to a Takagi-Sugenofuzzy system. Learning is a continuous updateof parameters. ANFIS is a hybrid algorithm inwhich back propagation algorithm is used toupdate fundamental factors [14].If we can predict the renal functionworsening, we can manage this disorder. Anappropriate parameter should be consideredfor disease worsening. The microalbuminuria isthe parameter, which detects the progression ofkidney disease at an earlier stage [15]. The otheradditional parameters considered in this paper is1236GFR, serum sodium, serum potassium, serumuric acid, blood urea, total protein, and serumalbumin. No other efficient method proposedin the past for predicting CKD worsening time.The objective is to provide a reliable methodwith good accuracy in the healthcare system.Materials and Methods Data collectionThe data of the present study were therenal failure test records of diagnosed CKDpatients from Dr. Jeyasekharan Medical Trust,Kanniyakumari during January 2014-December2017. The new parameter included is urinemicroalbumin. All the procedures wereapproved by the committee of Dr. JeyasekharanMedical Trust. A total of 900 CKD patient’s labdata were collected. Input selectionE-ANFIS is used in the proposed study topredict GFR values. The GFR value is calculatedby the MDRD equation.All variables were used as continuous to havegood training. Seven variables were influencingparameters of GFR. These variables includedurine microalbumin, serum sodium, serumpotassium, serum uric acid, total protein, bloodurea, and serum albumin. These variables weretaken as the inputs of the predicting model. Theexisting work excluded the urine microalbuminlab data. In this proposed study we includedthis as an additional attribute. The correlationbetween the considered variables and GFRvalues were calculated using Pearson correlationcoefficients technique. Pearson correlationcoefficients test was used to determine the mostsignificant input variables and this was usedbecause of the continuous nature of the variables(FIGURE 1).FIGURE 1 and TABLE 1 represent thecorrelation coefficients between the inputs andoutput GFR at a 4-month interval. Of the 7inputs, microalbumin is more correlated withthe output. Therefore, we considered the urinemicroalbumin as a new input for modeling thetechnique for diagnosing the progression.In the next step, the GFR values werepredicted at 4-, 8-, and 12-month intervalsusing E-ANFIS network model. The real dataduring a four-year period were collected at4-month intervals. Therefore, the GFR values10.4172/clinical-practice.1000470Clin. Pract. (2019) 16(5)

RESEARCHSubhashini & Jeyakumar700Urine MicroalbuminSerum SodiumSerum PotassiumSerum UricAcid600Total ProteinBlood UreaSerum AlbuminNo of samples500FIGURE 1. GFR ranges withother 5May2015Sep2015Jan2016May2016Sep2016DurationTABLE 1. Correlation between GFR and the attribute variables.Jan 2014May 2014Sep 2014Jan 2015May 2015Sep 2015Jan 2016May 2016Sep rum 163159172181159176182179were predicted for three sequential 4-monthintervals at 4-, 8-, and 12-month intervals.a) Building training and test datasets:The first step is to train all neural networks intotraining and test datasets. Training data usedfor optimization of weights. Testing data usedfor quality and forecasts. The test datasets arenormally selected among 25% to 35% of theoriginal data. In this work, 30% of the data wereselected for test data, the remaining 70% wereused for trainingb) Fuzzification of input variables:Neurofuzzy classifier in MATLAB was used tofuzzify input variables and to establish the rulebasec) Creating a fuzzy rule base forE-ANFIS: The fuzzy rules are generated usingthe membership functions of input variables.Total 200 rules (6 8 3 8 3 8 3 8 2 8 2 8 2 8 2 8 2 8) are created inthe rule base and used to estimate GFR values.TABLE 2 shows the membership function forthe considered variables in this study. FIGURE2 shows the E-ANFIS architecture of thepredicting model used in the proposed work1237Serum Uricacid315398451462421434403415442Total ProteinBlood UreaSerum 98401418405493504513486511506514526501Proposed E-ANFIS Technique Step 1Renal Failure dataset collected from Dr.Jeyasekharan Medical Trust, KanniyakumariDistrict Total instances: 900 Train data: 600 (430 progression, 170non-progression) Test data: 300 (220 progression, 80 nonprogression) Step 2Load the training data in MATLAB. Thefuzzy logic toolbox in Matlab provides anenvironment to build and evaluate fuzzy systemsusing a graphical user interface. It consistsof a FIS editor, the rule editor, a membershipfunction editor, fuzzy inference viewer, andthe output surface viewer. The neural networkis introduced in the rule editor phase to assignweights for the inputs. ALO optimizer findsthe global minimum (non-progression), wherethe weight should be minimum and globalmaximum (progression) where the weight10.4172/clinical-practice.1000470Clin. Pract. (2019) 16(5)

RESEARCHE-Anfis to diagnose the progression of chronic kidney diseaseshould be maximum using its exploratory andexploitation behavior, so the existing backpropagation algorithm is replaced by ALO inthe membership function editor phase. Step 3Modeling(E-ANFIS):oftheproposedmethodi) Crisp data converted into fuzzy valuesii) Eight input variables and one outputvariable are introduced to the fuzzy toolbox.ANN will assign a membership function foreach variable using Gaussian membershipfunctioniii) The fuzzy inference rule on each variableis determinediv) The number of membership functionand their locations are found for each inputv) Fuzzification is performed by assigningmembership valuesTABLE 2. Description of the data attributes and membership function.GFR/ml/min/1.73 m2Urine Microalbumin/mcg/min (ur ma)Serum sodium/mEq/l (sr Na)Serum Potassium/mEq/l (sr K)Serum uric acid/mg/dL (sr ua)Blood urea/mg/dL (bl ur)Total protein/mg/dL (tp)Serum albumin/mg/dL (sr al)Normal 90 rate Less 0--------------------------- Step 4ALO optimization algorithm:a. Initialization. Randomly initialize thepositions of input variablesb. Calculate the cumulative sum of amaximum number of iteration, where iterationrepresents the steps taken in a random walk. Thelocation of each input’s value is stored in onematrix. The corresponding objective values arestored in another matrix. One more matrix iscreated for saving the position and fitness valuec. Update the position of the input value byassigning random weight (random walk)d. Create two vectors, one with a minimumof all variables of one input source and otherwith a maximum of all variables of the sameinput. This gives the fitter input weight for thedesired output value (building traps)Severe15-29---------------Failure 15 300 120, 160 7, 3.5 6 14-23 8.0 5.4Membership Function63332222FIGURE 2. Proposed architecture ofE-ANFIS.Clin. Pract. (2019) 16(5)10.4172/clinical-practice.10004701238

RESEARCHE-Anfis to diagnose the progression of chronic kidney diseasee. Replace the position of all 8 inputvariables with the corresponding fit of the otherinput variables, if it becomes fitter. (entrapmentof ants in traps)f. Finally, update weight and position.(catching preys)g. Check termination criteria, if terminationis satisfied, return the optimal solution,otherwise back to update position. (rebuildingtraps) Step 5After the inference, automatic rules aregenerated. Elitism (remember the best solutionfound) is the important character of a natureinspired algorithm that allows maintainingthe best solution obtained at any stage of theoptimization process. In this study, the bestoutput obtained in each iteration is saved andconsidered as an Elite. Since the Elite is thefittest output, it will affect the movements ofall the other variable’s weight during iteration.The obtained overall result is a fuzzy value. Thisresult is converted into final crisp output bydefuzzification. Step 6Defuzzification is performed according tothe membership function of the output variable. Step 7After training the model, the test data isloaded and the final output is given by E-ANFIS(TABLE 2).TABLE 3 shows the attribute details and itsmembership function with output variable. TheGaussian membership function is used in allthe fields. The final output variable is GFR. Anumber of rules framed for the system are 200.The Mamdani system is used for designing theE-ANFIS.TABLE 3. E-ANFIS information.TypeNumber of inputsInput labelsNumber of outputsOutput labelsNumber of rulesOptimizerDefuzzification methodClin. Pract. (2019) 16(5)Mamdani8serum microalbumin (sr ma), serum sodium (sr Na), serumpotassium (sr K), serum uric acid (sr ua), total protein (tp),blood urea (bl ur), urine albumin (ur al), glomerular filtrationrate (gfr)1Progression/Nonprogression200ALOCentre of Sums (COS)E-ANFIS technique disallowed the localminima and local maxima for receiving anoptimal solution. The ALO algorithm is usedto compute the input for E-ANFIS algorithm.ANFIS is a grouping of ANN and fuzzy logic.Fuzzy logic has visions into the developmentof precise quantitative analysis. The parametersdecide the shape of the membership function.Fuzzy Logic Toolbox in Matlab gives informationabout the membership function. Fuzzy modelingprocedure learns information about the data set.From the given data set, the proposed algorithmE-ANFIS system is tuned using ALO optimizer.This makes the fuzzy to learn from the model.ALO is applied to adjust the occurrence of localminima and local maxima.Identifying the progression of renal diseaseis helpful in maintaining graft construction fordialysis. The E-ANFIS technique is an effectiveoptimization technique.T he trained E-ANFIS was used for estimatingGFR at 4, 8, and 12 months. The GFR changesare followed by E-ANFIS. Then the GFRfunction was estimated for the 4-month period.The results are presented as the value of GFR ata 4-month interval. This shows the relationshipbetween input and output variables. In the sameway, GFR values for 8 months and 12 monthswere predicted. E-ANFIS is able to predict GFRand find out the progression with improvedaccuracy.Result and DiscussionThis paper introduced an innovativemethodology for diagnosing the progression ofchronic renal disease. This method is developedby hybridizing ALO and ANFIS. Matlab 2016aversion is used for the proposed methodology.The parameters used for validation are accuracy,precision, recall, specificity, and F-Measure. AccuracyNumber of correct output from all outputmade. Accuracy is not the be-all and end-allmetric to use when selecting the best model.Accuracy is the traditional way to measure theperformance of a system but equally weighs thepositive and negative results, which may not bedesirable in an informal retrieval system, as thenumber of negative results can vastly outweighthe number of positive results. Thus otherparameters also considered.10.4172/clinical-practice.10004701239

RESEARCHAccuracy Subhashini & JeyakumarTP TNTP FP TN FN 100From FIGURE 3 and TABLE 4 it is knownthat the proposed algorithm is given betteraccuracy. This is because existing ANFIS usesback propagation algorithm. Back propagationalgorithm considers the nearby best fit and theproblem is local minima, but there may be abetter solution at a distant point. Proposedalgorithm rectifies the problem by its exploratorybehavior which helps in local optima and localmaxima avoidance and with its exploitationbehavior, it converges rapidly towards the globalminimum and global maximum. PrecisionIt can be very precise but inaccurate, alsobe accurate but imprecise. Precision talks abouthow precise the model is out of those predictedpositive, how many of them are actuallypositive. The progression of the disease shouldbe identified as a progression. A minor error inidentifying this will create unwanted chaos intreating the patient. So, the proposed algorithmshould be more précised.TPPr ecision 100TP FPFrom FIGURE 4 and TABLE 5, it is knownthat the proposed algorithm gives more numberof correct answers. The existing ANFIS fails toidentify more correct answers because of theissue in back propagation. The newly addedattribute urine microalbumin helps in takinga decision, thus the proposed E-ANFIS givesmore precision rate. RecallIf a disease progressive patient diagnosedas non-progressive, the cost associated withfalse negative will be high as the patient isleft untreated. It is necessary to prove that theproposed algorithm improves the percentage ofrecall.Re call / Sensitivity TPTP FN 100FIGURE 5 and TABLE 6 shows the recallpercentage. Network paralysis occurs in existingalgorithms when the weights are adjustedfrom very low to very high and vice versa.The proposed E-ANFIS takes many repeatedpresentations of the input patterns and theweights are needed to be adjusted before thenetwork is able to settle down into an optimalsolution. This improves the percentage of recallin the proposed algorithm.100ANN without MicroalbuminFuzzy without MicroAlbumin91.3390ANFIS without roposed) with MicroAlbumin76.3373.677271.6769.337073.33Accuracy (%)6050FIGURE 3. Accuracy.403020100Jan 2017May 2017Sep 2017Duration (months)TABLE 4. e.1000470Sep71.6773.3376.3389.33Clin. Pract. (2019) 16(5)

RESEARCHSubhashini & Jeyakumar10095.7590ANN without Microalbumin84.7895.63Fuzzy without MicroAlbumin89.58987.6287.394.69ANFIS without MicroAlbumin86.185.7987.82ALO-ANFIS(Proposed) with MicroAlbumin85.358070Precision (%)6050FIGURE 4. Precision.403020100Jan 2017May 2017Sep 2017Duration (months)TABLE 5. uzzy without MicroAlbumin89.0990ANFIS without MicroAlbuminALO-ANFIS(Proposed) output with .6473.18Recall(%)6050FIGURE 5. Recall.403020100 SpecificityIn medical diagnosis, specificity is the ability1241of a test to correctly identify those without thedisease (true negative rate). In this proposedwork true negative is the numbe

progression of the disease. The kidney disease progression considered as a function of various factors including GFR, urine microalbumin, serum sodium, serum potassium, serum uric acid, blood urea, total protein, serum albumin [1-4]. Among these, microalbuminuria (30-300 mg/day) is an ear

Related Documents:

May 02, 2018 · D. Program Evaluation ͟The organization has provided a description of the framework for how each program will be evaluated. The framework should include all the elements below: ͟The evaluation methods are cost-effective for the organization ͟Quantitative and qualitative data is being collected (at Basics tier, data collection must have begun)

Silat is a combative art of self-defense and survival rooted from Matay archipelago. It was traced at thé early of Langkasuka Kingdom (2nd century CE) till thé reign of Melaka (Malaysia) Sultanate era (13th century). Silat has now evolved to become part of social culture and tradition with thé appearance of a fine physical and spiritual .

On an exceptional basis, Member States may request UNESCO to provide thé candidates with access to thé platform so they can complète thé form by themselves. Thèse requests must be addressed to esd rize unesco. or by 15 A ril 2021 UNESCO will provide thé nomineewith accessto thé platform via their émail address.

̶The leading indicator of employee engagement is based on the quality of the relationship between employee and supervisor Empower your managers! ̶Help them understand the impact on the organization ̶Share important changes, plan options, tasks, and deadlines ̶Provide key messages and talking points ̶Prepare them to answer employee questions

Dr. Sunita Bharatwal** Dr. Pawan Garga*** Abstract Customer satisfaction is derived from thè functionalities and values, a product or Service can provide. The current study aims to segregate thè dimensions of ordine Service quality and gather insights on its impact on web shopping. The trends of purchases have

In this paper the design of an ANFIS estimator of permanent magnet brushless DC motor in PV pumping system is presented. The Adaptive Neuro Fuzzy Inference System (ANFIS), is a neural network that is functionally the same as a Takagi-Sugeno type inference model. The ANFIS is a hybrid

Chính Văn.- Còn đức Thế tôn thì tuệ giác cực kỳ trong sạch 8: hiện hành bất nhị 9, đạt đến vô tướng 10, đứng vào chỗ đứng của các đức Thế tôn 11, thể hiện tính bình đẳng của các Ngài, đến chỗ không còn chướng ngại 12, giáo pháp không thể khuynh đảo, tâm thức không bị cản trở, cái được

Le genou de Lucy. Odile Jacob. 1999. Coppens Y. Pré-textes. L’homme préhistorique en morceaux. Eds Odile Jacob. 2011. Costentin J., Delaveau P. Café, thé, chocolat, les bons effets sur le cerveau et pour le corps. Editions Odile Jacob. 2010. Crawford M., Marsh D. The driving force : food in human evolution and the future.