Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning

GRD Journals- Global Research and Development Journal for Engineering | Volume 6 | Issue 5 | April 2021 ISSN- 2455-5703

Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning Dr. Sheshang Degadwala Associate Professor Department of Computer Engineering

Dhairya Vyas Managing Director Shree Drashti Infotech LLP, Nizampura, Vadodara, Gujarat, India

Sigma Institute of Engineering, Vadodara, Gujarat

Abstract The CKD is the worldwide phenomenon with high morbidity and death rates. Chronic renal disease (CKD). Since the early stages of the CKD do not have any symptoms, patients frequently struggle to recognize their condition. HIV-patients are most likely to suffer from critically compromised kidney failure. Early diagnosis of CKD allows patients to get prompt medication to improve the disease's development. The suggested CNN deep learning model for the organization of the CKD phases observed with HIV is presented in this article. The credits of CKD patients are carried out on site. In the Chronic Kidney Disease phase predicted, CNN is 99% accurate with the PCA model.

Keywords- Chronic Kidney Disease, Stage, Machine Learning, Deep Learning, Convolution Nural Network, Principle Componend Analysis

I. INTRODUCTION Chronic kidney disease is an increasing global health concern (CKD). This is an incurable illness linked to a rise in morbidity and death, an increased risk for many other illnesses, including cardiac failure and higher costs for health care. More than two million patients worldwide undergo dialysis or kidney transplants to remain alive, but this figure may only account for 10% of people requiring survival therapy [2]. In only five rich nations, the bulk of the 2 million individuals who receive anti-kidney disease therapy constitute 12 percent of the world's population. In contrast, in about 100 developed countries just 20% of the world's population is handled, representing about half of the world's population. More than a million people die each year from unexplained kidney disease because of the enormous financial cost of dialysis and renal transplant care in 112 low-income countries [5].

When this form of circumstance occurs, filters do not function properly in order to allow HIV to infect the cells of the kidney, HIV may damage glomeruli (nephrons). If any drugs used to treat HIV are not closely controlled, the nephrons can damage the kidney. Since several studies were carried out to identify CKD or not and to detect CKD steps. The relationship between CKD and HIV is, however, novel.

The rate of incidence and prevalence of CKD in the context of HIV infection varies in all areas with significant differences and on the same continent. Variety depends on a number of factors such as kidney function tests, CKD definition, genetic variation, prevention program, access to a health care system and the implementation of integrated ART. The first obstacle to overcome is proper kidney function tests because no methods have been used to measure the glomerular filtration rate (eGFR) certified in PLWH.

Fig. 1: Causes of Chronic Kidney Disease [16]

Therefore, it is very important to detect, control, and control the disease early. It is necessary to predict the progression of CKD with due accuracy due to its strong and subtle nature in the early stages, as well as the heterogeneity of the patient. CKD is often described in stages of severity. Clinical decisions are influenced by the stage, whether the patient is progressing, and the level of progression. Also, defining the stage of the disease is very important as it provides many indications that support the determination of the necessary interventions and treatment.

All rights reserved by www.grdjournals.com

53

Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning (GRDJE/ Volume 6 / Issue 5 / 007)

Stages One Two Three a Three b Four Five

Table 1: Chronic Kidney Disease Stages Explanation GFR Normal damage of kidney function >90% Minor damage of kidney job 89-60% Minor to Modest damage 59-45% Modest to simple damage 44-30% Simple damage of kidney meaning 29-15% Kidney Stop Working <15%

Automated computer aided diagnose is process of getting stage information by input the patient data so according it will take action.

Algorithms for the prediction and differentiation of machine learning in healthcare is used. In order to identify and predict diabetic and predominant patients, Yu et al [2] have used the Support Vector Machine algorithm to indicate that SMV helps classify patients with common disorders. M N. Amin et al.[4] have now been diagnosed by the use of the Support Vector Machine (SVM) for the analysis of a full group of patients with anatomical magnetic resonance imaging (MRI) and the findings indicate that SVM is an encouraging means of diagnosing Alzheimer's disease at an early stage. PRNN is more effective than other cardiovascular prediction algorithms. E. Perumal et a cardiovascular disease prediction with the Algorithm for Probabilistic Neural Network, Algorithm for Decision Tree and the Algorithm for Naïve Bayes and PRNN. R. Shinde et al. [8] conducted HBV-induced hepatic cirrhosis predictors using the MLP and the findings indicate a satisfactory predictive result for liver disorders, in particular HBV patients with heptic insufficient disease with the MLP separator. R. Shinde et al.

II. LITERATURE PAPERS The thesis was conducted by Jack Edward, Corinne Isnard Bagnis, David M. Gracey et al (2020) The more regrettable prosperity of chronic kidney infection/disorder is synonymous with PLWWHIV, higher terribleness and mortality, and is strongly linked to CVD. If a person is in high risk of being chronically infected with the kidney or already has chronic kidney disease it should possibly prevent nephrotoxic ART. A analysis (2019) undertaken in Chinese by J. Qin, L. Chen, Y. Liu, C. Liu, C. Feng and B. Chen et al.[2], Chinese. The CKD diagnostic approach enables data imputation and sample diagnosis. For non-institute functionality using the KNN adjective, the integrated model may achieve ample consistency behind the Independent Loan. Open data research is mostly closed with only 400 models available due to contextual problems. Modeling approach. Likewise, the model's power could decline. In addition, the model does not explore the reality of chronic kidney disease since only two categories of data checks are used (Chronic kidney and Not Chronic kidney diseases).

E. H. A. Rady and A. S. Anwar et al. [3] On a clinical/lab dataset of 361 persistent kidney sickness patients, using PNN, SVM, RBF, and MLP mining calculations. The results of the kept an eye on computations were stood out all together from sort out which figuring conveyed the most exact results in gathering the reality period of Chronic kidney sickness. The Probabilistic Neural Networks computation is the best figuring for specialists to use to diminish assurance and treatment botches, as shown by this report. A. Al Imran, M. N. Amin, and F. T. Johora et al. [4] Experimental discoveries were analysed for two circumstances, model presentations on real (imbalanced) data and model displays on oversampled (changed) data, using Logistic Regression, Feedforward Neural Networks, and Large and Deep Learning. For both certifiable and oversampled data, feedforward neural associations made the best execution, with a 0.99 F1-score, 0.97 Precision, 0.99 Recall, and 0.99 AUC score.

N. Chetty, K. S. Vaisla, and S. D. Sudarsan et al. [5] On the Chronicle kidney disease dataset quality evaluator and arrangement models were utilized. Utilizing NB, SMO, and IBK Classifiers, bringing about better characterization precision on the decreased dataset than the first dataset.

P. Arulanthu and E. Perumal et al. [6] To lessen the ascribes of the first dataset, use JRip, SMO, Naive Bayes, and IBK Attributes evaluators, and analyse the after effects of the first and diminished datasets, as the classifier using the first dataset characterization results functioned admirably, yet the proposed classifier didn't woks for Jring for multiple data rows.

K. Shankar, P. Manickam, G. Devika, and M. Ilayaraja et al. [7] To order Chronic kidney illness, specialists used a streamlining model and a learning convention. The interaction chooses pertinent highlights of kidney information utilizing the Ant Lion Optimization (ALO) strategy to pick the best highlights for grouping. To accomplish high arrangement exactness, accuracy, F-measure, and affectability by utilizing DNN. Using the following: Maurya, R. Wable, R. Shinde, R. Jadhav, S. John and Dakshayani, R. [8]; Patients can be guided by the amount of potassium they have collected by using the levels of blood potassium to delay the movement of chronic kidney infection.

The author describes the importance of highlight preference procedures to improve the precise and implementing arrangements mechanisms, which are used to create and classify different techniques for determining or reducing their dimensionality.

III. PROPOSED SYSTEM As show in the below diagram of proposed CNN based deep learning system for CKD stage classification of HIV patient.

All rights reserved by www.grdjournals.com

54

Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning (GRDJE/ Volume 6 / Issue 5 / 007)

.

Fig. 2: Proposed System

A. CKD Dataset The dataset was purchased from the University of California, Irvine Machine Learning Repository. The dataset includes 26 functions, including 19 numerical attributes and 7 categorical ones. There were 158 separate instances in this dataset. The highlights include age, a blood pressure, explicit weight loss, egg whites, sugar, red platelets, discharge cell, bacteria, irregular blood glucose, blood urea, serum creatinine, sodium, hemoglobin, volumes of the pressed cells, white platelets, red platelets, hypertension and diagnosis.

Table 2: Dataset Colum of HIV CKD Patient Sr no Data Type 1 Age Numerical 2 Gender Categorical 3 ethnicity Numerical 4 Blood Pressure Numerical 5 Specific Gravity Numerical 6 Albumin Numerical 7 Sugar Numerical 8 Red Blood Cells Numerical 9 Pus Cell Numerical 10 Pus Cell clumps Numerical 11 Bacteria Numerical 12 Blood Glucose Random Numerical 13 Blood Urea Numerical 14 Serum Creatinine Numerical 15 Sodium Numerical 16 Potassium Numerical 17 Hemoglobin Numerical 18 Packed Cell Volume Numerical 19 White Blood Cell Count Numerical 20 Red Blood Cell Count Numerical 21 Hypertension Numerical 22 Diabetes Mellitus Categorical 23 Coronary Artery Disease Categorical 24 Appetite Categorical 25 Pedal Edema Categorical 26 Anemia Categorical 27 Class Categorical

Noise Removal Most standard Noise removal techniques is obtained because of the abuse of shortenings, data passage botches, copy records, missing esteems. Inside this specific circumstance, one key research subject is the de-duplication issue which is the recognition and expulsion of copy records from a database. The exploration challenge is that databases contain both correct and vague copies.

In the stage, the non-numerical data are cleared and procured the numerical dataset for proceeding with further.

B.

1) Shortenings [3] Data is not valuable in an array. Data is only valuable once information, insight or in other words knowledge is extracted from it and is used to make decisions. In data shorting data is divided into equal size raw and column.

2) Data Passage Botches [5] In this part data is process in badly manner so the data passage botches algorithm will remove bugs data.

All rights reserved by www.grdjournals.com

55

Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning (GRDJE/ Volume 6 / Issue 5 / 007)

3) Copy Records [2] It is better to construct a single statement that removes all unwanted copies in one go. Before removing the duplicate records, you must decide which instances you want to keep.

4) Missing Esteems [7] Missing value imputation is one of the important tasks in machine learning especially in the cases where the data is small and there is a need of using all available data. Therefore, it plays an important role in the classification performance of the models. As our dataset is small and all the features except the output feature contain missing values, we have an obvious need for imputing missing values. There are many missing value imputation techniques in machine learning, however, we have used the mean and mode imputation in this research. The numerical features were imputed using mean and the categorical features were imputed using mode.

C. eGFR Calculation Mostly, the phases of ongoing kidney illness are controlled by the assessed glomerular filtration rate (eGFR). There are five phases, with two sub-stages 3a and 3b in stage 3. Kidney work is normal at stage one. To analyses CKD, two examples should be required in any event 90 days separated, with recorded qualities utilized too. Creatinine computation, sex, identity, and age all impact the assessed Glomerular Filtration Rate (eGFR). Alteration of Diet in Renal Disease is quite possibly the most dependable techniques for ascertaining eGFR. Correct it and Calculate egfr with this equation. eGFR=175 x (Creatinine/88.4)^ 1.154 x (Age) ^ 0.203 x (0.742 if female) x (1.210 if dark) (1) Example we have value of Creatinine=4.1 Age=68 ethnicity=black Gender= Male then the value of eGFR= 175x (4.1/88.4) ^1.154x (68) ^0.203x1.210= 14.41 so stage is 5. As it is between 0 to 15.

D. Attributes Selection RFE Recursive Feature Elimination, or RFE for short, is a popular Attributes selection algorithm. RFE is popular because it is easy to configure and use and because it is effective at selecting those features (columns) in a training dataset that are more or most relevant in predicting the target variable. There are two important configuration options when using RFE: the choice in the number of features to select and the choice of the algorithm used to help choose features. Both of these hyperparameters can be explored, although the performance of the method is not strongly dependent on these hyperparameters being configured well. For our data 28 data column after applying RFE it will be reduced to 14 data columns E. Dimensition Reduction PCA In this research use orthogonal transformation in this process, which is a statistical technique for transforming a collection of potentially correlated values known as principal components into a single value (PCA). When using PCA as a feature selection method, the basic concept is to choose variables based on the magnitude (from largest to smallest in absolute values) of their coefficients (loadings). PCA aims to substitute pp (more or less correlated) variables with k<pk<p uncorrelated linear combinations (projections) of the original variables, as you may remember. Let's set aside the question of how to choose the best kk for the problem at hand. The value of the kk principal components is determined by their explained variance, and each variable contributes to each component to varying degrees. Using the largest variance criteria is similar to function extraction, in which the principal attribute is used as a new feature rather than the original variables. However, we can choose to keep only the first factor and pick the j<pj<p variables with the highest absolute coefficient; the number jj may be determined by the proportion of the total number of variables (e.g., keep only the top 10% of the pp variables) or by a fixed cut-off. (e.g., considering a threshold on the normalized coefficients). For our data 28 data column after applying PCA it will be reduced to 13 data columns.

F. CNN Layer Description 1) Conv In-depth study, the convolutional neural network is a group of deep neural networks, often used to analyse visual images. They are also known for fixed or fixed space exchange networks that are inserted, based on the construction of shared weights and translation signals.

Fig. 3: Convolution Example

All rights reserved by www.grdjournals.com

56

Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning (GRDJE/ Volume 6 / Issue 5 / 007)

2) MaxPool Max Pool is a blending capacity that chooses the most noteworthy item in the element map district covered by a channel. In this manner, the yield next the maximum pooling layer will be an element map that contains the best noticeable highlights of the older component map.

Fig. 4: Pooling Example

3) Flatten Flattening is changing over the information into a 1-dimensional cluster for contributing it to the following layer. We smooth the returns of the convolutional layers to make a lone long segment vector. Additionally, it is related with the last course of action model, which is known as a totally related layer.

Fig. 5: Flatten Example

4) Dense The dense layer is a neural association layer that is related significantly, which infers each neuron in the dense layer gets commitment from all neurons of its past layer. The dense layer is found to be the most ordinarily used layer in models. Far out, the thick layer plays out a grid vector increase.

5) Dropout Dropout can be utilized after convolutional layers (for example, Conv2D) and in the wake of pooling layers (for sample MaxPooling2D). Regularly, dropout is just utilized after the pooling layers, however, this is only a harsh heuristic. For this situation, dropout is applied to every component or cell inside the element maps.

Fig. 6: Dropout Example

6) FineTune Fine-tuning an organization is a methodology dependent on the idea of move learning [1,3]. We begin preparing CNN to learn highlights for a wide area with an order work focused at limiting blunder in that space.

All rights reserved by www.grdjournals.com

57

Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning (GRDJE/ Volume 6 / Issue 5 / 007)

Fig. 7: Fine-Tune Flow of Work

Fine-Tune is a process based on the concept of learning transfer. Here our proposed model is beginning to train CNN to learn the characteristics of a broader domain with a segmentation function aimed at reducing error on that domain. After that, it will replace the partition function and use the network and reduce the error in another domain. Under this setting, transfer network features and parameters from a broad domain to a specific one. Classification function uses SoftMax classification in this network learning.

IV. PARAMETERS, RESULTS AND ANALYSIS In this part discuss about different parameters definition and also showcase results. At the end comparative analysis is done with existing methods.

TP + TN Acccuracy = (2) TP + FP + TN + FN TP Precision = (3) TP + FP TP Recall = (4) TPFN

2∗TP

F1 − Score = (5) 2∗TP+FP+FN Where, true positive (TP), true negative (TN), false positive (FP) and false negative (FN) As visible in figure 8 CNN is train using all 24 attributes and accuracy report is plot for recall, precision and f1-score.

Fig. 8: Convolution Neural Network (CNN) Result

All rights reserved by www.grdjournals.com

58

Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning (GRDJE/ Volume 6 / Issue 5 / 007)

As visible in figure 9 CNN is train using all 13 attributes which is selected by PCA and accuracy report is plot for precision, recall, and f1-score.

Fig. 9: Convolution Neural Network (CNN)+ PCA Result Table 3: Result Analysis of HIV and non-HIV Datasets Classifier Attributes Non-HIV Accuracy (%) HIV Accuracy (%) SVM 14 89 93 KNN 14 99 97 DT 14 99 97 RF 14 98 95 Ada Boost 14 99 97 XgBoost 14 98 97 CNN 24 97.88 99 CNN-PCA 13 98.66 99

Above Table 3 shows Comparative study of Different Classifier of Machine Learning Approaches and Deep learning Approaches using Attribute and accuracy parameter. Due to Blood Pressure and Red Blood cell value fluctuation in non-HIV datasets affects the classifier accuracy. Therefore, non-HIV accuracy is less than HIV data. Below graph indicate CNN+PCA gives almost 99% accuracy for stage classification.

Fig. 10: Analysis graph of ML and DL methods

All rights reserved by www.grdjournals.com

59

Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning (GRDJE/ Volume 6 / Issue 5 / 007)

V. CONCLUSION For companies to make various choices, the classification of chronic renal diseases in patients afflicted with HIV is highly useful.

First, delete the noisy data and then apply the PCA algorithm for the collection of attributes. Used CN N for CKD designation, which will enhance the precision compared with chronic kidney disease stages after application of the attribute collection. With this proposed CNN + PCA method, reliable classification of CKD data would be enhanced. After calculating and splitting the information into five levels, eGFR is subsequently classified into a CKD formula. Early diagnosis and decision-making will also be effectively carried out. The move to a proposed model would produce a 99 percent accurate real-time classification. It will allow people to get early care and save lives. The x-ray image can be used for static data for the precision of the battery.

REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22]

C. I. Bagnis, J. E. Heron, and D. M. Gracey, “Contemporary issues and new challenges in chronic kidney disease amongst people living with HIV,” AIDS Res. Ther., vol. 17, no. 1, pp. 1–13, 2020, doi: 10.1186/s12981-020-00266-3.

Y. Liu, C. Liu, J. Qin, L. Chen, C. Feng, and eB. Chen, “A machine learning methodology for diagnosing chronic kidney disease,” IEEE Access, vol. 8, pp.

20991–21002, 2020, doi: 10.1109/ACCESS.2019.2963053.

A. S. Anwar and E. H. A. Rady , “Prediction of kidney disease stages using data mining algorithms,” Informatics Med. Unlocked, vol. 15, no. March, p.

100178, 2019, doi: 10.1016/j.imu.2019.100178.

M. N. Amin, A. Al Imran, and F. T. Johora, “Classification of Chronic Kidney Disease using Logistic Regression, Feedforward Neural Network and Wide Deep Learning,” 2018 Int. Conf. Innov. Eng. Technol. ICIET 2018, pp. 1–6, 2019, doi: 10.1109/CIET.2018.8660844.

S. D. Sudarsan,N. Chetty, K. S. Vaisla, “Role of attributes selection in classification of Chronic Kidney Disease patients,” pp. 1–6, 2016, doi: 10.1109/cccs.2015.7374193.

E. Perumal , P. Arulanthu , “Predicting the Chronic Kidney Disease using Various Classifiers,” 4th Int. Conf. Electr. Electron. Commun. Comput. Technol.

Optim. Tech. ICEECCOT 2019, pp. 70–75, 2019, doi: 10.1109/ICEECCOT46775.2019.9114653.

K. Shankar, G. Devika, P. Manickam, M. Ilayaraja, “Optimal Feature Selection for Chronic Kidney Disease Classification using Deep Learning Classifier,” 2018 IEEE Int. Conf. Comput. Intell. Comput. Res. ICCIC 2018, pp. 1–5, 2018, doi: 10.1109/ICCIC.2018.8782340.

R. Shinde, S. John, R. Jadhav, A. Maurya, R. Wable, and R. Dakshayani, “Chronic Kidney Disease Prediction and Recommendation of Suitable Diet Plan by using Machine Learning,” 2019 Int. Conf. Nascent Technol. Eng. ICNTE 2019 - Proc., no. Icnte, pp. 1–4, 2019, doi: 10.1109/ICNTE44896.2019.8946029.

S. C. Jat . R. Yadav, “Feature selection and dimensionality reduction methods for chronic disease prediction,” Int. J. Sci. Technol. Res., vol. 9, no. 4, pp.

2912–2918, 2020.

S. saraswat, S. Vashisth, I. dhall, CKD diagnosis using Multilayer Perceptron Classifier’ 10th International Conference on Cloud Computing, Data Science & Engineering (Confluence)” 2020.

K. Shankar, M. Elhoseny, J. Uthayakumar, “Intelligent Diagnostic Prediction and Classification System for Chronic Kidney Disease,” Sci. Rep., vol. 9, no.

1, pp. 1–14, 2019, doi: 10.1038/s41598-019-46074-2.

M. H. Abd Wahab, A. A. Johari, A. Mustapha, “Two-Class Classification: Comparative Experiments for Chronic Kidney Disease,” 2019 4th Int. Conf. Inf. Syst. Comput. Networks, ISCON 2019, pp. 789–792, 2019, doi: 10.1109/ISCON47742.2019.9036306.

G. Saini, B. Pandey, A. Khamparia, S. Tiwari, D. Gupta, and A. Khanna, “KDSAE: Chronic kidney disease classification with multimedia data learning using deep stacked autoencoder network,” Multimed. Tools Appl., 2019, doi: 10.1007/s11042-019-07839-z.

S. Maheshwari, S. Singhal, and M. Meena, Recent Findings in Intelligent Computing Techniques, vol. 707. Springer Singapore, 2019.

V. Sharma, S. Sharma, A. Sharma, “Performance Based Evaluation of Various Machine Learning Classification Techniques for Chronic Kidney Disease Diagnosis,” 2016, [Online]. Available: http://arxiv.org/abs/1606.09581.

"Learn More About Kidney Disease: Types, Symptoms, and Causes, Siemens-healthineers.com, 2021. [Online]. Available: https://www.siemenshealthineers.com/en-us/clinical-specialities/kidney-disease/about-kidney-disease. [Accessed: 04- Mar- 2021] Estimated Glomerular Filtration Rate (eGFR), National Kidney Foundation, 2021. [Online]. Available: https://www.kidney.org/atoz/content/gfr. [Accessed: 04- Mar- 2021] The dawn of a new era in chronic kidney disease, Astrazeneca.com, 2021. [Online]. Available: https://www.astrazeneca.com/what-science-can-do/stories/thedawn-of-a-new-era-in-chronic-kidney-disease.html. [Accessed: 04- Mar- 2021] 10 Applications of Deep Learning in Business - Algorithm-X Lab, Algorithm-X Lab, 2021. [Online]. Available: https://algorithmxlab.com/blog/10applications-deep-learning-in-business/. [Accessed: 04- Mar- 2021] Check out this amazing template to make your presentations look awesome at, Slideteam.net, 2021. [Online]. Available: https://www.slideteam.net/examplesof-deep-learning-applications-ppt-powerpoint-presentation-introduction.html. [Accessed: 04- Mar- 2021] Wang, Huafeng et al. ‘A Hybrid CNN Feature Model for Pulmonary Nodule Malignancy Risk Differentiation’. 1 Jan. 2018 : 171 – 187.

How Deep Learning Can Help Prevent Financial Fraud, Investopedia, 2021. [Online]. Available: https://www.investopedia.com/terms/d/deep-learning.asp.

[Accessed: 04- Mar- 2021]

All rights reserved by www.grdjournals.com

60