applied

sciences
Article
A Deep Learning Approach for Kidney Disease Recognition
and Prediction through Image Processing
Kailash Kumar 1 , M. Pradeepa 2 , Miroslav Mahdal 3, * , Shikha Verma 4 , M. V. L. N. RajaRao 5
and Janjhyam Venkata Naga Ramesh 6

1 College of Computing and Informatics, Saudi Electronic University, Riyadh 11673, Saudi Arabia
2 School of Information Technology and Engineering, Vellore Institute of Technology, Vellore 632014, India
3 Department of Control Systems and Instrumentation, Faculty of Mechanical Engineering,
VSB-Technical University of Ostrava, 17. Listopadu 2172/15, 708 00 Ostrava, Czech Republic
4 Department of Computer Applications, ABES Engineering College, Ghaziabad 201009, India
5 Department of Information Technology, Seshadri Rao Gudlavalleru Engineering College,
Vijayawada 521356, India
6 Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Guntur 522302, India
* Correspondence: [email protected]

Abstract: Chronic kidney disease (CKD) is a gradual decline in renal function that can lead to
kidney damage or failure. As the disease progresses, it becomes harder to diagnose. Using routine
doctor consultation data to evaluate various stages of CKD could aid in early detection and prompt
intervention. To this end, researchers propose a strategy for categorizing CKD using an optimization
technique inspired by the learning process. Artificial intelligence has the potential to make many
things in the world seem possible, even causing surprise with its capabilities. Some doctors are
looking forward to advancements in technology that can scan a patient’s body and analyse their
diseases. In this regard, advanced machine learning algorithms have been developed to detect the
presence of kidney disease. This research presents a novel deep learning model, which combines
a fuzzy deep neural network, for the recognition and prediction of kidney disease. The results
show that the proposed model has an accuracy of 99.23%, which is better than existing methods.
Citation: Kumar, K.; Pradeepa, M.; Furthermore, the accuracy of detecting chronic disease can be confirmed without doctor involvement
Mahdal, M.; Verma, S.; RajaRao, as future work. Compared to existing information mining classifications, the proposed approach
M.V.L.N.; Ramesh, J.V.N. A Deep shows improved accuracy in classification, precision, F-measure, and sensitivity metrics.
Learning Approach for Kidney
Disease Recognition and Prediction Keywords: kidney disease; image processing; fuzzy logic; deep neural network; hybrid of fuzzy and
through Image Processing. Appl. Sci.
deep neural network
2023, 13, 3621. https://doi.org/
10.3390/app13063621

Academic Editor: Jan Egger

1. Introduction
Received: 8 February 2023
Revised: 8 March 2023 Computer vision techniques that resemble video surveillance, such as image segmenta-
Accepted: 9 March 2023 tion, fall under the category of image saliency detection. The low-resolution image issue is
Published: 12 March 2023 addressed using a convolutional neural network (CNN) and deep learning CNN to enhance
the image quality and clarity of identification [1]. In this paper [2], the author explains the
issue-solving methods in three different aspects faced when using CNNs. Facial expression
recognition is a form of progress, but it does not provide accurate results in live detection.
Copyright: © 2023 by the authors.
By combining a first examination report (FER) with the CNN method, the quality of the
Licensee MDPI, Basel, Switzerland.
content becomes better and more understandable. Typically, machine vision is used to
This article is an open access article
create a laser spot energy that operates in a superposition area. The models used in this
distributed under the terms and
area involve a convolutional network with a deep learning (DL) concept [3]. If a possibility
conditions of the Creative Commons
exists to detect the presence of acute kidney disease in a human body through machine
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
learning (ML) concepts, it would be beneficial for both doctors and patients to solve this
4.0/).

Appl. Sci. 2023, 13, 3621. https://doi.org/10.3390/app13063621 https://www.mdpi.com/journal/applsci

Appl. Sci. 2023, 13, 3621 2 of 14

issue. The accuracy of the results shows that non-renal sequential organ failure assessment
(SOFA) prediction can occur even in the presence of acute kidney disease (AKD) [4].
The term health information categorization refers to the use of a classifier model
with datasets to improve the definition of healthcare. Clinical records are organized to
identify and predict objectives. This has a significant impact on how extraction outcomes
are predicted. These techniques assist medical professionals in making accurate analytical
conclusions during specific diagnoses. Although tree topologies in classifications can be
easily coupled [5], the decision tree is still widely used in decision-making as a classification
technique. To be more specific, information retrieval tools can be designed to uncover valid,
beneficial, and logical frameworks, instances, themes, or decision-making elements hidden
within health information [6]. Both children and adults can suffer from chronic kidney
disease (CKD), where the kidney’s ability to function consistently deteriorates [7].
CKD is widely studied due to the high risk of renal problems in certain groups of
individuals, including those with hypertension, obesity, or with a parent diagnosed with
CKD [8]. The decline in renal function is gradual and occurs over a long period, which
distinguishes it from severe renal disease [9]. Based on information from previous patients,
individuals with the same health conditions can be grouped and effective treatment options
can be provided [10]. Regular occurrence identification techniques extract important
features and functions that describe the entire sample [11,12]. Chronic kidney disease is a
major clinical emergency that is often ignored. End-stage renal disease occurs when the
kidneys are damaged and unable to remove harmful substances from the blood [13]. Some
recent works aim to identify life-threatening illnesses, including renal disease, by using
sequence models such as naive Bayes and artificial neural network (ANN) models such as
C4.5 to predict the occurrence of CKD [14]. Kidney ultrasound imaging for the prediction
of kidney function and chronic kidney disease (CKD) has long been regarded desirable in
clinical practise due to its safety, simplicity, and cost-effectiveness. It used kidney length
annotations to cut off the kidneys’ periphery and other data augmentation techniques to
provide more data with variances in order to further extract information from ultrasound
pictures. Moreover, bootstrap aggregation was used to reduce overfitting and enhance the
generalizability of the model [15].
The main contribution of the paper is:
• To begin, we gather CKD scan pictures and use a normalization technique to preprocess
the images.
• To identify the existence of renal illness, cutting-edge machine learning algorithms
have been created.
• To demonstrate a fuzzy deep neural network (FDNN) for the detection and prognosis
of renal illness.
The rest of the article is structured as follows: Section 2 presents a literature review;
Section 3 presents a suggested method; Section 4 presents the result and discussion; and
the last section presents a conclusion.

2. Literature Review
In order to effectively treat and control chronic kidney disease, early identification and
characterisation are believed to be essential components. In the study, effective data mining
methods are used to uncover and extract hidden information from clinical and laboratory
patient data. This information may help doctors identify disease severity stages with the
greatest accuracy [16]. Aljaaf et al. investigate the potential of several machine-learning
techniques for the early diagnosis of chronic kidney disease. While this topic has been
extensively investigated, we are using predictive analytics to assist our technique as it
looks at the link between the data parameters and the characteristic of the target class [17].
Using the Internet of Medical Things (IoMT), images captured by cameras are analyzed to
detect the presence of disease in the human body, as explained in this paper [18], which
reports a 96.88% verification rate for histopathological images. Deep learning (DL) and
machine learning (ML) occupy the fourth position among industrial revolutions, with
Appl. Sci. 2023, 13, 3621 3 of 14

DL being based on artificial neural networks. This paper [19] provides a comprehensive
overview of deep learning techniques and their applications. Due to the large amount of
data, data-parallel processing methods have been introduced, bringing many benefits to the
processing system through resource allocation. This paper [20] introduces a deep learning
interfacing model. Both ML and DL concepts rely on algorithms to function. This paper [21]
explains the proper use of algorithms in the machine learning concept and the ability to
automatically detect algorithms if necessary. Recently, web corruption has occurred, caused
by either the host or hackers. Bi-LSTM (bi-directional long short-term memory) is one tool
used to protect web-based applications. The output is checked in two modes, with the first
mode passing at 93.1% and the second mode reaching 93.91% [22].
When a person is affected by a disease, such as kidney disease, the risk of spreading
other diseases increases. For example, kidney disease can lead to cardiovascular or kidney
failure. Doctors or machines need to estimate the time frame for this to occur [23]. Mobile
phones, which use machine learning algorithms, may also allow users to unlock their
phone using facial recognition. Traditionally, facial recognition was accomplished through
a separate mechanism using neural concepts, as demonstrated in [24], where the author
used electrocardiography for testing. Both ML and DL concepts also aid in classifying and
analysing Indore identification, similar to crime and robbery, with an accuracy rate of up to
75% using hotspot connections and city information [25]. It can be challenging for machines
to differentiate images on a black and white surface, but if there is colour differentiation,
the results are clearer. The authors of [26] explore the possibility of identifying the presence
of stones in the human kidney through machine analysis.
It is anticipated that alternative or comprehensive treatments will focus on therapies
that aim to enhance quality of life, prevent illness, and address diseases that conventional
medicine has limited success in treating. The manufacturer demonstrated a preference for
using iridology to distinguish between different forms of kidney disease, either normal or
exceptional [27]. A total of 192 individuals with chronic kidney disease and 169 healthy
individuals were evaluated. A method for acquiring, processing, and characterizing iris
images using wavelet transformation and a flexible neuro-fuzzy inference system was
developed to reduce dependence on iridologists. The results showed, for both individuals
with kidney problems and healthy individuals, an accuracy rate of 81% and 92%, respec-
tively [28]. A CNN was constructed to identify 10 major crop diseases using a database
of 500 photos of healthy and diseased grain stems and leaves collected from agricultural
fields. The proposed convolution neural algorithm achieved an accuracy of 95.48% using a
10-fold cross-validation architecture. This accuracy is significantly greater than a traditional
classification model [29].

3. Proposed System
3.1. Problem Statement
The limitations of DL systems are another crucial factor to remember. These techniques
are often regarded as “black boxes”, making it challenging to “understand” them without
additional diverse data. The discussion in this paper is limited to the FDNN classification
issue and segmentation characteristics. To better educate and explain the DL findings to
doctors, clinical co-variables are also crucial. Moreover, it is critical to comprehend the
most effective communication tactics for conveying to doctors the significance of each
co-variable in the ultimate DL outcome.
Recent research allows for direct observation or imaging of the entire vascular system,
which is an easier task when performed on external body parts, but takes longer when it
comes to internal parts. Large-scale retinal studies have shown that concepts such as these
could be related to or combined with artificial intelligence and deep learning methodologies.
The main reason for using artificial intelligence in healthcare is the availability of medical
data, and the rise of complex algorithms that form the backbone of AI and ML. Images are
the only input for machines, and they function based on the questions asked by the user by
understanding errors and comparison concepts (Figure 1).
Appl. Sci. 2022, 12, x FOR PEER REVIEW 4 of 15

Appl. Sci. 2023, 13, 3621 comprehend the most effective communication tactics for conveying to doctors the sig- 4 of 14
nificance of each co-variable in the ultimate DL outcome.
Recent research allows for direct observation or imaging of the entire vascular sys-
tem, which is an easier task when performed on external body parts, but takes longer
when it comes to internal parts. Large-scale retinal studies have shown that concepts
such as these could be related to or combined with artificial intelligence and deep learn-
ing methodologies. The main reason for using artificial intelligence in healthcare is the
availability of medical data, and the rise of complex algorithms that form the backbone
of AI and ML. Images are the only input for machines, and they function based on the
questions asked by the user by understanding errors and comparison concepts (Figure
1).

Figure 1 FigureFigure
1. Proposed Model Model
1. Proposed for Kidney Disease Disease
for Kidney Prediction.
Prediction.

FigureFigure
2 shows the five
2 shows the stages of chronic
five stages kidney
of chronic disease
kidney in ain
disease human, which
a human, pro-progresses
which
gresses from
from Stage
Stage 1 to
1 to Stage
Stage 5,5,and
andaffects
affectsboth
boththe
theright
rightand
and left
left kidneys
kidneys equally. InIn this
this research,
research, the aim
the aim is toispredict
to predict
and and analyze
analyze these
these stages,
stages, andand numerical
numerical features
features areare ex-
extracted from
tracted from kidney
kidney imagesimages
to make tothemake the predictions.
predictions.

Normal Function Mid-loss of function Moderate loss of function Severe loss of function Kidney failure

FigureFigure 2. Stages
2. Stages ofKidney
of Renal Renal Kidney
Disease.Disease.

3.2. Dataset
The Changhua Christian Hospital in Taichung, Taiwan, is the source of the dataset.
With patients’ identifying information deleted, a total of 5617 records from 1 January 2000
to 27 July 2017 were obtained. Patients’ conditions were monitored throughout this time
or until they pass away. Patients’ monitoring times varied from less than a month to over
five years, with an average of about two and a half years. A flat CSV file was used to
Appl. Sci. 2023, 13, 3621 5 of 14

hold the dataset. The intended variable, “survival or not”, is among the 35 properties
that are listed in columns. Other factors can be broken down into two categories: general
information about the body or the CKD itself, such as the ID, diagnosis date, age, gender,
height, weight, BMI, stage, haemodialysis and its date; and whether or not the person was
taking special medications or had comorbid conditions, such as high blood pressure, heart
disease, chronic liver disease, or other conditions. Table 1 contains statistics regarding a
number of significant characteristics [30].

Table 1. Various feature of the dataset.

Attribute Detail Number Percentage

Gender Male 3151 56.1%
Female 2466 43.9%
y ≤ 18.5 239 4.3%
18.5 < y ≤ 25 2841 50.6%
BMI 25 < y ≤ 30 1862 33.1%
30 < y ≤ 40 642 11.4%
y > 40 33 0.6%
0 3722 66.4%
Hemodialysis 1 1326 23.6%
2 569 10.1%
High Blood Pressure No 1644 29.3%
Yes 3973 70.7%
Anemia No 4795 85.4%
Yes 822 14.6%
Non-survival No 4933 87.8%
Yes 684 12.2%
Chronic Liver Disease No 5059 90.1%
Yes 558 9.9%

As the dataset is mostly clean and comprehensive, minimal preparation was required.
The ID element and two date attributes were simply dropped, since the former was not
important and the latter were not full. Moreover, the absence of a haemodialysis date might
reflect one of two very different things: either the patient was receiving haemodialysis but
the date was not recorded, or the patient did not yet need it. As there is no easy method
to discern between the two diametrically opposed circumstances, we decide to combine
them all so that the remaining data would better fit the prediction algorithm. Moreover, we
tested our models using 5-fold cross-validation on the dataset.
The MJ Health Research Foundation gave permission for and provided all of the
datasets that were utilised. The MJ Health Research Foundation and the Far Eastern
Memorial Hospital’s Research Ethics Review Committee assessed and approved the study’s
strategy with regard to ethical concerns about the use of database data (FEMH-IRB-107126-
E, Protocol Version 7, 18 June 2020). (Approval No.: MJHRF-2016005A).
The dataset was then preprocessed using the normalising technique known as min–max
normalisation. One of the most popular methods for normalising data is the min–max
method. For each feature, the lowest value is converted to a 0, the highest value is converted
to a 1, and all other values are converted to a decimal between 0 and 1.

4. Proposed Work
Chronic kidney disease (CKD) often causes symptoms such as illness and constipation,
leading to a decrease in quality of life and increased risk of death. The inflammatory process
of CKD can impact the development of illness, cachexia, and kidney osteodystrophy, but
also increases the risk of stroke in CKD patients. Ghrelin, a form of oestrogen produced
in the stomach, has been found to have potential benefits in regulating food intake and
meal appreciation, making it a potential therapy for anorexic CKD patients. Ghrelin has
been shown to have anti-inflammatory properties and to stimulate food cravings. This
Appl. Sci. 2023, 13, 3621 6 of 14

evaluation discusses the metabolic changes in ghrelin and its potential implications for
CKD. The benefits, drawbacks, and unanswered questions about using ghrelin in CKD
healthcare is also discussed.
The provision of CKD care is a major challenge in modern times, particularly in
developed countries where people in remote locations want access to high-quality medical
care. Artificial intelligence has greatly benefited the healthcare industry, just as it has
transformed other aspects of life. However, the conventional telemedicine setup faces
certain challenges, such as the need for a local healthcare centre with a dedicated team, the
need for hospital equipment to process patient reports, treating patients within 48 h, access
to medical expertise within a healthcare centre, the cost of local healthcare centres, and the
requirement for a reliable Wi-Fi connection.
The smart CKD process is managed and monitored using fuzzy logic. There are two
main issues: when the model’s capacity is insufficient, more than two designs are merged
to resolve the issue. To provide an efficient solution to the crisis, a hybrid system was
created by combining multiple methods. In some forms of hybrid fuzzy neural network, a
fuzzy inference system is combined with an artificial neural network, resulting in a fuzzy
neural network (FNN).
This method involves a “fuzzy neuron”, and the fuzzy neuron method has been
separated into two classifications, as described in the following:
• The development of a fuzzy neuron model.
• Creation of a single model and algorithm of the model for incorporating neural systems
through fuzziness.
The neural system discovers the f [n, n + 1] operation, which is a partition of the
self-assurance earned through fuzzy inference. This should gain f (n + 1) utilizing the
period denoted by k and the framework condition k + 1. A stochastic modification module
enhances the authorization with f (k) the fuzzy role and also the expected possibility
regarding decisions, but also produces a finished product.

m0 (k) = d(m(k), g[k, k + 1]) (1)

To evaluate the fuzzy guideline, the fuzzy rule unit m0 (k) is organized and evaluated
with Equation (1). The data device is a standard predecessor that gains a unit d(m(k)). The
behaviour control is communicated by unit ( g[k, k + 1]). The procedure is finished with a
defused combination.
With input nodes, the signs and weight training are actual values. The data does not
affect these signs. The yield is nearly identical to the data. The signal ni may work with a
large number of materials si to build such items.

g = si ni , i = 1, 2. (2)

Here the data input is taken as g, which is gathered for the purpose of implementing
such data as represented in Equation (2).

FL = g1 + g2 = s1 k1 + s2 k2 , (3)

Cachexia is a disease characterized by muscle loss, anorexia, increased energy expen-

diture, and the presence of chronic disease (CKD). It is a strong predictor of mortality in
CKD patients, which is 100- to 200-fold higher than in the general population. Cachexia is
one of the most inflammatory conditions, distinct from malnutrition, which is a deficiency
of nutrients.
To determine the FL’s fuzzy logic production (refer to Equation (3)), the neuron
−1
employs its work transfer f (y), which can be a sigmoid function result, f (y) = (1 + e−y ) ,
which is represented in Equation (4).

y = f ( FL) = f (s1 k1 + s2 k2 ) (4)

Appl. Sci. 2023, 13, 3621 7 of 14

An ordinary neural net is a basic network that employs Sigmoid function f , redun-
dancy, and other inclusions.
The decision support system used in AI-based electronic health records is based on a
set of fuzzy rules. These rules are derived from both factual and fuzzy data. The following
are examples of fuzzy rules.
• If the blood pressure is high, the temperature is high, and the pulse rate is low,
judgment is good.
• If your blood pressure is high and your pulse rate is low, your judgment is likely to be
impaired.
• If the temperature is normal, the pulse is rapid, and the blood pressure is moderate,
then the judgment is low.
• If the temperature is low and the heart rate is high, then determine whether the blood
pressure is low.
• If the temperature and pulse rate are both normal, then the judgment is good if the
blood pressure is low.
Because it performs tasks once, the mode command technology uses both the point of
g
entry and the available spectrum for data transfer, but the web access transmits si as given
in Equation (5). !
g | gi,n |2 Yi,n g−n
si = αi R log log 1 + (5)
σ2

where i represents the percentage of access of internet bandwidth utilized by new terminal
update tasks, gi,n represents the relation recession scaling factor between access point and
terminal, and Yi,n represents terminal products and services, g−n represents node facility
distance, b represents loss, but σ2 also represents interaction noise level.
Accordingly, the efficiency of the gi data link data transfer is elaborated as in Equation (6).
!
k | gn,i |2 Xn g−b
di = β i B log log 1 + (6)
σ2

in which β i signifies the fraction of power transmission frequency bandwidth occupied by

the terminal able to receive work-related jobs, gn, i signifies the link economic downturn
relation between the entry point and terminal, and Xn signifies the foundation network’s
transmitting speed.
Muscles waste away due to cachexia, while fats are also underutilized. Patients with
chronic kidney disease (CKD) often experience anorexia, which is defined as a loss of
appetite. The disease in CKD patients can also be linked to reducing the sense of taste
and smell for food, early satiation, changes in neurohormonal filtration, instability in
acetylate cyclase, increased cognitive tryptophan, and increased levels of inflammatory
cytokines. Anorexia not only decreases verbal energy, but also protein intake, which is
a major contributor to cachexia. Increased resting energy consumption has been linked
to higher mortality rates and cardiovascular mortality in CKD patients, and it is also tied
to the prevalence of cachexia in these patients. Currently, there is no effective treatment
for cachexia in CKD. Nutritional and health strategies, such as caloric diets with anabolic
steroids, have largely proven ineffective. This highlights the urgency for the development
of new drug treatments for this potentially fatal condition in CKD patients.
Job ni is, however, evaluated here on gateways if it is not offloaded to edge networks.
Equation (7) shows the time delay in completing various jobs geographically.
gi
Xin = (7)
gik
where gik shows the capacity of the terminal gi to process information and organize tasks
regionally. As a consequence, the overall duration delay captured by gi research scholars
on a local scale is illustrated in Equation (8).
Appl. Sci. 2023, 13, 3621 8 of 14

gim = ∑ (1 − αi ) gim (8)

m∈ g

In this case, if various activities such as a t-norm or the ni –co-norm are used for
connecting the reach information to such a neuron, the result is termed a hybrid artificial
neuron and is shown in Equation (9).
mi
gin = (9)
gin
Discoveries about the pathophysiology of cachexia in CKD have led to innovative
therapeutic approaches. Cachexia in CKD is caused by an increase in the frequency of
inflammatory responses, which affects the central nervous system (CNS) and creates a
relationship between the release and function of several key neuropeptides, affecting
metabolic activity. Leptin and the melanocortin centre in the hypothalamus have already
been proposed as targets for cytokine activity, and they remain crucial regulators of appetite
and energy metabolism.
These changes result in a fuzzy neural design that relies on fuzzy mathematical tasks.
The bandwidth delay duration is proportional to the amount of information received and
the network throughput for data transfer, as stated in Equation (10).
gi
gin = (10)
bik
A set of fuzzy rules is defined for the AI-based CKD process delivery system. These
rules are based on fuzzy data, and the server’s computing time is proportional to the size
of the data and the server’s computing capability, as expressed in Equation (11).

f fi
bi = (11)
Xi
The temperature controller is an integrated circuit that measures the body temperature
in degrees Celsius. The voltage level corresponding to the temperature is displayed. The
make and model of the temperature sensor is LM35. The design of this body temperature
controller is believed to perform better than a linear temperature controller. As a result, the
duration spent on un-loading the assigned task si to the network edge is transmitted as in
Equation (12).
f
sin = sic + sih + si (12)
The following emergency requirements are monitored: respiratory arrest, heart condi-
tion, vagal convulsion, and pressure detector. As a result, the time frame related to the task
of unloading si to the edge device is conveyed as in Equation (13).
n
Xin = ∑ (αi din ) (13)
i =1

The pulse rate seems to be the primary indicator of critical medical behaviour and
health fitness. Within the patient outcomes and management field, the PRS is the most
commonly managed and investigated sensor.
n
ming = ∑ ( gin + din ) (14)
i =1

As in the hypothalamus, two different identity documents of neurons regulate food

intake. Each neuronal subset produces neurotrophic factor Y (NPY), which enhances food
intake, whereas another neuronal subset continues to produce melanocortin substances,
which restrict food intake.
Equation (14) is used to evaluate pulse rate and complicated diseases such as heart
attack. When a subject f 1 places its finger on the data panel, the sensor activates. The result
Appl. Sci. 2023, 13, 3621 9 of 14

is identified on the input panel. The sensor provides a 5-volt direct power source as in
Equation (15).
s.t. f 1 : ∑ mi ≤ my (15)
hi ∈ d
The smart CKD process of client management and monitoring framework is required.
The framework suggested f 2 in Equation (16) is a framework profiting from a fuzzy logic
system that is simple to use and enforced for creating decisions.

f2 : ∑ αi ≤ 1 (16)
gi ∈ d

Circulating levels of leptin and insulin suppress appetite by inhibiting the production
of neuropeptide Y (NPY) and increasing the production of monoaminergic protein, while
also suppressing the manufacturing of agouti-related peptide (AgRP) in the hypothalamus.
Inflammatory mediators cause anorexia through their central actions. Cytokines reduce
gastrointestinal activity, as metabolic changes affect the hormonal system and modulate the
neuropeptide identity in the hypothalamus, both of which can impact eating behaviour.
The f 3 in Equation (17) suggests a novel method for organization, as it makes use of
both detectable information and a fuzzy decision-making process.

f3 : ∑ βi ≤ 1 (17)
gi ∈ d

Desirability is explained in terms of lag time, seeing as f 4 in Equation (18) aims to

reduce the lag time of such a power sector, where less time delay was correlated with
greater strength.
f 4 : min ≥ 0, ∀i ∈ d (18)
The strength and endurance values are calculated as in Equation (19).

1
di = (19)
gi
Effective and accommodating treatment options for CKD patients are highly needed.
Ghrelin is more effective than most other orexigenic hormones, as it increases food intake in
both small mammals and humans. Recent findings support the potential use of ghrelin and
its analogues as food craving stimulants and anabolic treatments for cachexia associated
with uraemia and other diseases. Ghrelin may impair energy metabolism, but it can also
exacerbate cachexia through IGF-dependent and insulin-like growth factor processes.

5. Experimental Result
By changing various components of the proposed better FDNN model, we carried out
our tests as part of an ablation research. By changing various components, it is possible to
create a more dependable design with improved classification accuracy. Throughout the
ablation experiment, modifications were made to the FDNN, activation function, kernel
initializer, and optimizer.
The suggested approach was tested using the Origin pro simulation software. CKD
often causes illness and constipation, both of which are linked to reduced quality of
life and increased mortality risk. The inflammatory CKD process may contribute to the
development of illness, cachexia, and kidney osteodystrophy, but it is the increased risk
of stroke in people with CKD that is concerning. Ghrelin is a hormone produced in the
stomach and is thought to act as an oestrogen. Its effects are mediated by the growth
hormone secretagogue receptor (GHSR). Ghrelin’s potential to increase food consumption
and meal enjoyment make it a promising therapy for anorexic CKD patients. Ghrelin
has been shown to have anti-inflammatory properties and stimulate food cravings. The
recognition and prediction of kidney disease in the context of image processing in an
afferent and efferent vessel used the HFNN algorithm to identify the convoluted tubule
Appl. Sci. 2023, 13, 3621 10 of 14

of the Bowman capsule for the mean and standard deviation, PSNR, and accuracy in the
glomerular filtration rate of CKD (Table 2). Performance of the recognition and prediction
of kidney disease is shown in Figure 3.

Table 2. Renal Function Tests Analysis for Kidney Disease Identify using Hybrid Fuzzy Neural
Network Algorithm.

Glomerular Proximal Convoluted Bowman

Parameters Afferent Vessel Efferent Vessel
Filtration Rate Tubule Capsule
Mean 0.5643 0.4534 0.7654 0.7864 0.8658
Standard Deviation 0.2445 0.7563 0.6453 0.8796 0.6756
PSNR 0.6756 0.5564 0.7564 0.9675 0.7564
Accuracy 87.45 84.24 90.46 95.55 96.67

Figure 3
Figure 3. Performance of recognition and prediction of kidney disease. (a) Mean (b) Standard
Deviation (c) PSNR (d) Accuracy.

In today’s world, providing care for patients with chronic kidney disease (CKD) is a
major challenge, especially in countries with limited access to quality medical care. The
advancement of artificial intelligence (AI) has greatly impacted healthcare, just as it has
transformed other fields. However, the conventional telemedicine approach has certain
limitations, such as the need for a local health centre with dedicated personnel and hospital
equipment, the time it takes for patients to receive treatment and medical information from
experts, the cost of local clinics, and the requirement for stable Wi-Fi connectivity. The
percentage of wireless internet channel capacity used by the port to inform users of new
activities is represented by I. The connection downtime transformation function between
the entry point and the stations is represented by gin , while yin represents the terminal’s
products or services. The node’s building location is represented by g−n , b represents the
news team’s loss, and σ2 represents the quality of communication sound used within the
Appl. Sci. 2023, 13, 3621 11 of 14

network, as shown in Figure 4. Table 3 presents the analysis of CKD severity in five stages
using the hybrid fuzzy neural network algorithm to evaluate the overall accuracy of the
training and testing for CKD severity.

Figure 4
Figure 4. Analysis of CKD severity using a hybrid fuzzy neural network algorithm.

Table 3. Performance result analysis for the CKD severity using hybrid fuzzy neural network algorithm.

CKD Severity Training/Testing Using Hybrid

Value Overall Accuracy (%)
Parameter Fuzzy Neural Networks (%)
Stage 1 7 (8.4) 92.12 97.34
Stage 2 7 (8.4) 94.34 96.45
Stage 3 26 (31.3) 93.56 97.35
Stage 4 34 (41) 91.45 95.46
Stage 5 9 (10.8) 96.45 98.67
Stage 1: Normal renal function despite kidney damage. Stage 2: Mild kidney function loss. Stage 3: Moderate to
severe renal dysfunction and mild to moderate kidney dysfunction. Stage 4: Serious kidney function loss. Stage 5:
Failure of the kidney.

The fuzzy logic for smart CKD process management and monitoring is characterized
by the command line gi information processing ability to complete tasks locally. As an
outcome, the whole latency3 is recognized by gi researchers only at the local scale, as
in Equation (8), represented in Figure 5. There are several issues to consider: when a
single model is insufficient to solve a problem, more than two designs are collaborated
to overcome the problem. When two or more models are joined to offer an effective
answer to a challenging issue, a hybrid system is developed. In a hybrid model of fuzzy
neural networks, fuzzy logic systems analyze CDK in comorbidities using the HFNN
algorithm to evaluate hypertension values based on training and testing (Table 4) in
diabetes cerebrovascular disease to evaluate accuracy.

Table 4. Performance analysis for CDK in co-morbidities using hybrid fuzzy neural network algorithm.

Training/Testing Using Hybrid

Co-Morbidities Value Accuracy (%)
Fuzzy Neural Networks (%)
Hypertension 78 (94) 89.34 92.34
Diabetes 59 (71.1) 85.53 90.21
Dyslipidemia 73 (88) 82.23 88.34
Ischemic heart disease 24 (28.9) 94.56 96.42
Cerebrovascular disease 5 (6) 93.13 95.34
Appl. Sci. 2023, 13, 3621 12 of 14

Figure 5
Figure 5. Analysis for CDK in Co-morbidities using hybrid fuzzy neural network algorithm.

If different operations, such as a t-norm or ni -conorm, are used to provide information

to a neuron, it is referred to as a hybrid artificial neuron, as shown in Equation (9). Figure 6
shows that ghrelin stimulates pre-adipocyte differentiation while inhibiting adipogenesis,
suggesting that it works with adipocytes to promote adipogenesis. In conclusion, small-
scale clinical studies in CKD patients have provided valuable evidence to support the
short-term orexigenic effects of subcutaneous ghrelin administration. However, the clinical
utility of ghrelin in CKD will be determined by long-term improvements in appetite, muscle
mass, and function, as well as poor outcomes compared to our proposed method. Our
proposed method, the HFNN, is compared to the existing system for CDK diseases (refer
to Table 5). The training and testing stages 1 to 5 are carried out for the left kidney (98.34%)
and right kidney (97.46%) and then the overall accuracy is evaluated (99.23%). The analysis
of the existing TRM method shows a kidney stage 1 to 5 training and testing accuracy of
95.76% and an overall accuracy of 97.46%. The results indicate that the HFNN method
provides the best performance 4 compared to the existing method.

Figure 6
Figure 6. Analysis for CKD using hybrid on fuzzy neural network comparison with existing method.
Appl. Sci. 2023, 13, 3621 13 of 14

Table 5. Comparison Result Analysis.

Algorithm Kidney Stages CKD Training/Testing Accuracy

Stage 1 to 5 (Left) 98.34 97.46 99.23
Hybrid on Fuzzy Deep Neural Network
Stage 1 to 5 (right) 98.78 97.86 99.34
Stage 1 to 5 (Left) 91.34 95.65 97.46
Traditional Radioimmunoassay Method
Stage 1 to 5 (right) 91.84 95.76 97.56

6. Conclusions
Chronic kidney disease (CKD) is becoming more widespread across various age
groups due to poor diet, lack of sleep, and other factors. CKD starts with the slow decline
of kidney function and can lead to total kidney failure. This can result in various treatments
for patients, including dialysis and transplantation. As the kidneys are internal organs, it is
difficult to diagnose the disease in its early stages. Therefore, it is important for individuals
to have regular check-ups. This research focuses on early-stage prediction of CKD by
developing a hybrid fuzzy deep neural network model, which is compared and evaluated
against the current radioimmunoassay method. The results show that the proposed model
outperforms the existing method in more accurate disease identification. Due to the small
sample size of the dataset used in the research, it has been decided that future work will be
carried out with bigger datasets or by comparing the outcomes of this dataset with those of
another dataset. Also, in an effort to reduce the prevalence of CKD, an effort has been made
to determine whether an individual with this syndrome is more likely to have chronic risk
factors such diabetes, hypertension, or a family history of kidney failure.

Author Contributions: Conceptualization, K.K., M.P., M.M., S.V., M.V.L.N.R. and J.V.N.R.; Methodol-
ogy, K.K., M.P., M.M. and S.V.; Validation, K.K., M.P., S.V., M.V.L.N.R. and J.V.N.R.; Formal analysis,
K.K., M.P., S.V., M.M. and M.V.L.N.R.; Investigation, K.K., M.P., M.V.L.N.R. and J.V.N.R.; Resources,
K.K., M.P., S.V., M.V.L.N.R. and J.V.N.R.; Writing—original draft, K.K., M.P., M.M., S.V., M.V.L.N.R.
and J.V.N.R.; Writing—review and editing, K.K., M.M. and M.P. All authors have read and agreed to
the published version of the manuscript.
Funding: This work was supported by the project SP2023/074 Application of Machine and Process
Control Advanced Methods supported by the Ministry of Education, Youth and Sports, Czech Republic.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data presented in this study are available through email upon
request to the corresponding author.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Wang, X.; Li, S. Image saliency prediction by learning deep probability model. Signal Process. Image Commun. 2019, 78, 471–476.
[CrossRef]
2. Rajan, S.; Chenniappan, P.; Devaraj, S.; Madian, N. Novel deep learning model for facial expression recognition based on
maximum boosted CNN and LSTM. IET Image Process. 2020, 14, 1373–1381. [CrossRef]
3. Jwaid, W.M. Image Processing Technology and Deep Learning Application: In Relation to the Context of Laser Positioning. J.
Phys. Conf. Ser. 2021, 1879, 032130. [CrossRef]
4. He, J.; Lin, J.; Duan, M. Application of Machine Learning to Predict Acute Kidney Disease in Patients with Sepsis Associated
Acute Kidney Injury. Front. Med. 2021, 8, 792974. [CrossRef]
5. Goshua, A. Deep-learning model may accurately predict autism diagnosis. Spectrum 2021. [CrossRef]
6. Gupta, K.D.; Sharma, D.K.; Ahmed, S.; Gupta, H.; Gupta, D.; Hsu, C.-H. A Novel Lightweight Deep Learning-Based Histopatho-
logical Image Classification Model for IoMT. Neural Process. Lett. 2021, 55, 205–228. [CrossRef] [PubMed]
7. Lawal, O.M.; Zhao, H. YOLOFig detection model development using deep learning. IET Image Process. 2021, 15, 3071–3079.
[CrossRef]
8. Kim, Y.-K.; Kim, Y. DiPLIP: Distributed Parallel Processing Platform for Stream Image Processing Based on Deep Learning Model
Inference. Electronics 2020, 10, 1664. [CrossRef]
Appl. Sci. 2023, 13, 3621 14 of 14

9. Xu, N. The Application of Deep Learning in Image Processing is Studied Based on the Reel Neural Network Model. J. Phys. Conf.
Ser. 2021, 3, 032096. [CrossRef]
10. Karacan, H.; Sevri, M. A Novel Data Augmentation Technique and Deep Learning Model for Web Application Security. IEEE
Access 2021, 9, 150781–150797. [CrossRef]
11. Priyadarshini, J.; Premalatha, M.; Čep, R.; Jayasudha, M.; Kalita, K. Analyzing Physics-Inspired Metaheuristic Algorithms in
Feature Selection with K-Nearest-Neighbor. Appl. Sci. 2023, 13, 906. [CrossRef]
12. Ganesh, N.; Shankar, R.; Čep, R.; Chakraborty, S.; Kalita, K. Efficient Feature Selection using Weighted Superposition Attraction
Optimization Algorithm. Appl. Sci. 2023, 13, 3223. [CrossRef]
13. Schwartz, E.; O’Nell, K.; Alreja, A.; Ghuman, A.; Anzellotti, S. Deep networks trained to recognize facial expressions predict
ventral face-selective ECoG responses as well as networks trained to recognize identity. J. Vis. 2021, 21, 2221. [CrossRef]
14. Maddala, V.K.S.; Jayarajan, K.; Braveen, M.; Walia, R.; Krishna, P.; Ponnusamy, S.; Kaliyaperumal, K. Multisensor Data and
Cross-Validation Technique for Merging Temporal Images for the Agricultural Performance Monitoring System. J. Food Qual.
2022, 2022, 9575423. [CrossRef]
15. Kuo, C.-C.; Chang, C.-M.; Liu, K.-T.; Lin, W.-K.; Chiang, H.-Y.; Chung, C.-W.; Ho, M.-R.; Sun, P.-R.; Yang, R.-L.; Chen, K.-T.
Automation of the kidney function prediction and classification through ultrasound-based kidney imaging using deep learning.
NPJ Digit. Med. 2019, 2, 29. [CrossRef] [PubMed]
16. Sugiyarti, E.; Jasmi, K.A.; Basiron, B.; Huda, M.; Shankar, K.; Maseleno, A. Decision Support System of Scholarship Grantee
Selection using Data Mining. Int. J. Pure Appl. Math. 2018, 119, 2239–2249.
17. Aljaaf, A.J.; Al-Jumeily, D.; Haglan, H.M.; Alloghani, M.; Baker, T.; Hussain, A.J.; Mustafina, J. Early Prediction of Chronic Kidney
Disease Using Machine Learning Supported by Predictive Analytics. In Proceedings of the 2018 IEEE Congress on Evolutionary
Computation (CEC), Rio de Janeiro, Brazil, 8–13 July 2018; pp. 1–9.
18. Gorzaáczany, M.B.; RudziĔski, F. Interpretable and accurate medical data classification—A multi-objective genetic-fuzzy opti-
mization approach. Expert Syst. Appl. 2017, 71, 26–39. [CrossRef]
19. Kunwar, V.; Chandel, K.; Sabitha, A.S.; Bansal, A. Chronic Kidney Disease analysis using data mining classification techniques.
In Proceedings of the 2016 6th International Conference—Cloud System and Big Data Engineering (Confluence), Noida, India,
14–15 January 2016; pp. 300–305.
20. Hussein, S.E.; Hassan, O.A.; Granat, M.H. Assessment of the potential iridology for diagnosing kidney disease using wavelet
analysis and neural networks. Biomed. Signal Process. Control 2013, 8, 534–541. [CrossRef]
21. Lu, Y.; Yi, S.; Zeng, N.; Liu, Y.; Zhang, Y. Identification of rice diseases using deep convolutional neural networks. Neurocomputing
2017, 267, 378–384. [CrossRef]
22. Keçeli, A.S.; Kaya, A.; Keçeli, S.U. Classification of radiolarian images with hand-crafted and deep features. Comput. Geosci. 2017,
109, 67–74. [CrossRef]
23. Zawbaa, H.M.; Emary, E.; Parv, B. Feature selection based on antlion optimization algorithm. In Proceedings of the 2015 Third
World Conference on Complex Systems (WCCS), Marrakech, Morocco, 23–25 November 2015; pp. 1–7.
24. Janardhanan, P.; Heena, L.; Sabika, F. Effectiveness of Support Vector Machines in Medical Data mining. J. Commun. Softw. Syst.
2015, 11, 25–30. [CrossRef]
25. Rubini, L.J.; Eswaran, P. Generating comparative analysis of early stage prediction of Chronic Kidney Disease. Int. Open Access J.
Mod. Eng. Res. 2015, 5, 49–55.
26. Yang, S.; Guo, J.-Z.; Jin, J.-W. An improved Id3 algorithm for medical data classification. Comput. Electr. Eng. 2018, 65, 474–487.
[CrossRef]
27. Shankar, K.; Elhoseny, M.; Lakshmanaprabu, S.K.; Ilayaraja, M.; Vidhyavathi, R.M.; Elsoud, M.A.; Alkhambashi, M. Optimal
feature level fusion based ANFIS classifier for brain MRI image classification. Concurr. Comput. Pract. Exp. 2018, 32, e4887.
28. AlMuhaideb, S.; Menai, M.E.B. An Individualized Preprocessing for Medical Data Classification. Procedia Comput. Sci. 2016, 82,
35–42. [CrossRef]
29. Ponnusamy, S. Effective web content mining using noise removal from web pages. Wirel. Pers. Commun. 2015, 84, 99–121.
30. Zhang, H.; Hung, C.L.; Chu WC, C.; Chiu, P.F.; Tang, C.Y. Chronic kidney disease survival prediction with artificial neural
networks. In Proceedings of the 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Madrid, Spain,
3–6 December 2018; pp. 1351–1356.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.