2023 International Conference on Computer Communication and Informatics (ICCCI), Jan.

23-25, 2023,
Coimbatore, INDIA

A Deep Learning Approach Based on CT Images

for an Automatic Detection of Polycystic Kidney
Disease
2023 International Conference on Computer Communication and Informatics (ICCCI) | 979-8-3503-4821-7/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICCCI56745.2023.10128336

S.K.Aruna N.Deepa Devi.T,

Research Scholar, Department of Associate Professor, Department of Professor, Department of Computer
Computer Science and Engineering, Computer Science and Engineering, Science and Engineering,
Saveetha School of Engineering, Saveetha School of Engineering, Saveetha School of Engineering,
Saveetha Institute of Medical and Saveetha Institute of Medical and Saveetha Institute of Medical and
Technical Sciences, Technical Sciences, Technical Sciences,
Chennai Chennai Chennai
arunask9001.sse@saveetha.com deepa23narayanan@gmail.com devi.janu@gmail.com

Abstract— Deep Learning is a subdivision of machine replicates an “intellectual” feature that is in relationship with
learning; deep learning is mainly separated into multiple the other human brain. Statistical techniques, Artificial
neural network layers. These layers allow it to learn large Intelligence techniques and classical symbolic AI is one
amounts of the data and try to mimic the human brain. By among the approaches. The rise of AI is also connected to
adding more hidden layers it can help to improvise the almost limitless storage and an overflow of data from all
efficiency. Deep Learning techniques are utilized to identify kinds like videos, images, texts, audios, transactions and
diabetes mellitus, vascular dementia early diagnosis and geographical information.
mammography nodule ultrasonographic recognition.
Pathology and Radiology will be invested in the future for new
advancement in deep learning. Kidney irregularities is one of B. Machine Learning:
the most common diseases among people of all ages. A global Machine learning is a subsection of computer
shortage of urologists developed a deep learning model to programming that assists computers in learning without the
diagnose kidney problems with this study. The kidney area is aid of computer vision. Machine learning whichever
surrounded by a thick wall. It is possible to have multiple cysts progressed from the survey of pattern recognition and
with high moisture concentrations that cause renal diseases. computational learning theory in artificial intelligences. The
The units from zero -twenty indicate different types of renal development of algorithms can predict and learn from data
diseases. Tissues that can be grouped together form a solid on data. Machine learning uses a broad range of processing
lamp known as tumor. Tumors can cause direct damage to activities. These specific algorithms can implement and
glands, spinal cells and other organs. Urolithiasis is another create a high efficiency is either challenging or impossible.
name for renal stone disease where a large amount of solid
For example, include email sorting, sensing of network
materials enter into the digestive tract. In a research paper,
CNN is trained to learn how the cysts diagnosis should be done.
intruders or malicious users actually trying to data breach,
Tree Convolution neural deep learning techniques VGG19 are recognition of optical characters [2].
used to detect kidney abnormalities. To train and evaluate a
dataset of 4000 abdomen computed tomography with classifier
and segmentation is created. The increased categorization and
accuracy rate obtained to display the efficacy of kidney
disorders is 98%.

Keywords— Convolutional Neural Network(CNN),

Computed Tomography, Deep Learning, Cysts Kidney
Diseases

I. INTRODUCTION
Kidney irregularities are common for all age group
peoples. The manual examination is very difficult to clarify
the irregularities it costs high and time absorbing. medical
research focused on the precisely and quickly clarify the Fig. 1. Types of Machine Learning
kidney disorders. To better recognize the kidney disorder Supervised learning algorithm: This algorithm uses
class that can generalize the kidney disorders. experienced data and its outcome is predictable. It has two
types of techniques one is classification and another is
A. Artificial Intelligence regression technique. Classification technique mostly used to
Artificial intelligence is knowledge displayed by predict standard application. Regression is used to predict
machines succeeding to a target [1]. When a system data limit and time for a specific task.

979-8-3503-4821-7/23/$31.00 ©2023 IEEE

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on March 04,2024 at 19:13:44 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI), Jan. 23-25, 2023,
Coimbatore, INDIA
Unsupervised Learning Algorithm: It has two types one is arithmetical feature in CNN is called Convolution. It is a
clustering and other is association. Clustering method is used kind of linear procedure in which two activities are
to group similar objects into clusters, this cluster consists of a multiplied to produce the shape. Fig2 demonstrated the CNN
group of similar objects and less similar objects are in architecture. CNN is composed of many layers which
another group. Cluster analysis finds common things from include convolutional layers used to reduce the image size
these clusters like data objects and categorizes them. from vertically and horizontally. Second layer is a pooling
Association method is used to find the connections between layer that is used to get maximum value from the result.
variables from a huge database. Pooling layer separated by the max pooling and average
pooling [5]. The third layer is fully connected and the layer is
C. Deep Learning: used to flatten the input image. Fully connected layer is
Deep learning in image analysis is necessary for the process usually found at the end of the CNN architecture.
of separating an image into multiple segments. Image
segmentation is an act of separating an image into relevant
and irrelevant parts based on human perception. Image
segmentation is a primary goal that helps to divide an image
into many pieces for statistical evaluation, having allowed us
to extract only the required fraction of it. The image’s
division will be determined by its attributes. A few examples
include color, pixel intensity and texture [3]. K-means
clustering, fuzzy clustering and hierarchical clustering are
three basic clustering approaches. Image segmentation is the
process of labeling every pixel in an image and it has
numerous uses. Image compression and scene understanding
are two examples. Many picture segmentation methods have
been created. However, with the introduction of deep
learning in computer vision an exhibit of deep learning
models for picture segmentation have emerged. Deep
learning designs have developed a new creation of image
classification modeling techniques with significantly
improved performance often outperformed which is known
as benchmarks [4]. As a result, there has been a paradigm Fig 2. Architecture of Convolutional Neural Network
change inside the ground. Deep learning is basically known
as a state-of- the-art technique that can be quickly used to E. Kidney Diseases
analyze the medical image. It can detect the object fast and Kidney cysts are formed in the form of round pouches in
partitioning which helps to classify them, it can be involved the fluid that form on or in the kidneys. Kidney cyst can be
in the pathogenesis skeletal structure will help the examiner only associated with a condition that can impair the blood
by general clinical examination. Physicians can easily detect and kidney function. However simple cysts are the most
the disease using deep learning applications and they can frequent type of kidney cysts are not cancerous and rarely
make fast decisions regarding the diseases. cause complications. One cyst in the kidney is normal. But
Table 1 - Deep Learning Architecture’s Year by Year multiple cysts can form complications in one or both the
Distribution kidneys. Simple cysts are not as same as polycystic kidney
diseases. Simple cysts differ from the complex cysts.
Year Deep Learning Framework
Complex cysts must be monitored for changes that might be
indicated for the cancer. CT machine scans and x-rays beams
1960-1962 Back propagation Model Started are used to create a cross sectional image of the desired
1970 CNN
anatomy. It can provide three-dimensional knowledge about
the system. CT images can be well suited for the research
1979 Neocognitron kidney examination. It can provide the three-dimensional
1989 Back propagation Demonstrated at Bell
image as well as flexible images [2]. Renal irregularity can
Lab lead to renal failure. If they are not examined at an early
1990-1995 Recurrent Neural Network stage, it will result in critical kidney stones, cysts and tumors.
Furthermore, most previous studies used traditional
1995-2000 LSTM, CNN
algorithms for machine learning to categorize only one breed
2000-2010 Vanishing Gradient Problem of disease, it can be a stone or tumor or cyst. Ultrasound
(US) images were used in some studies. Tumors can cause
2010-2017 Deep stacked network, Gated recurrent
direct damage to glands, spinal cells and other organs.
unit
2017-2020 Alex Net
II. RELATED WORK
In this study, a summary of early work of renal diseases
is provided from various quadrants. Gathered and analyzed a
D. Convolutional Neural Network (CNN): maximum of 12,500 whole stomach and urogram Computed
CNN is a real form of multidimensional network tomography medical images consisting of cysts or tumors or
created to extract patterns from the image patches. An normal data set as well as stone identify source for this study.

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on March 04,2024 at 19:13:44 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI), Jan. 23-25, 2023,
Coimbatore, INDIA
The images were subjected to information extraction which the identification of the patient of high precision, lowering
helped to know that then visuals from across all categories the pathologist volume of the work and supplying people
shared the same form of average color features. This research with an instrument of the data that can automatically assist
develops six levels, 3 of which were based on current public the shape of the kidney, lowering the possibility of wrong
different versions of the images power transformer CCT, diagnosis [16].
Swim transformers, EANet of three other predicated on
widely used power transformers. Deep learning techniques Combined renal quality is a useful statistic in diagnosis
are adjusted inside a final output. A comparative made by and treatment of renal diseases. It creates a complete
each concept demonstrated, VGG16 conducted a Swim automated method for segmentation kidneys from T2
transistor that exceeded every concept in the way of weighted nuclear magnetic resonance scan to compute the
effectiveness along with a 95% consistency. Comparisons of total kidney volume and renal diseases of the patient. The
precision recall show the Swim transformer outperforms computer controlled clear understanding separation of the
other models. Furthermore, when compared with all other left renal and right renal from nuclear magnetic resonance
models Swim transformers took minimal hours to learn when scan using a two-dimensional neural network. The sample
using the one and the same amount by epochs. Research measurement of thirty HC patients and thirty CKD patients
discovered by the Blackbox of deep learning models and of the data [15].
discovered by the VGG16 model outperforms the Resnet50 A CAD system is used to sense more than two class renal
together with Inceptionv3 models by better displaying the irregularities from the x-ray images. The CAD system
desired abnormalities in the anatomy. Researchers think that model used pretrained features to extract the images using
Swim transformers build the concept and the VGG16 also ResNet101 and the purpose of the classifier is used to
build the concept which can be of terrific use for detecting support vector machines. X-ray images will get overripe by
kidney cysts or tumors or stones as well as minimizing noise image quality and stability. The proposed model with
patients suffering from pain [11]. the residual learning network to slow down the speckle
Sonography imaging is one among the most admired noise. The system should be able to retain a better result,
methods for irregularities. The vulnerable model kidney when compared to the existing system [14].
includes a variety of chronic renal diseases such as gallstone, The researchers always aim to recognize the diseases and
fibroids kidney and hydronephrosis diseases. The CDK find an easy way to find the nature of disease of those who
causes the onset of a variety of severe diseases, greatest are affected by the autosomal dominant polycystic kidney
notably cardiovascular diseases, pulmonary attacks, diseases for the clinical treatment [13].
congestive heart failure and so on. As just an outcome, early
classification of CDKs in healthcare situations is extremely Patients who have cysts with different categories to
covered. It may save countless lives. Researchers are identify by the individual patients. The researchers have
currently focusing on developing fully automated disease created a 3D image instance semantic segmentation
detection methods that do not require human contact. Deep algorithm for the kidney by MR scans. We recreated both the
learning algorithms are growing important in a variety of classification of objects as well as the task of identification in
smart healthcare not only because of their rapid and precise MR scans. It recognizes how to resolve the particular image
methods. But in its minimal interference of the data. Two in problem regulation. Renal irregularities are common
approaches for classification CDKs in sonography renal through individuals of every age group. It is necessary to do
images are advanced. The first stage of the standard way manual investigation and to know about irregularities of cost
utilizes the GAONN as a classification such as the second and pulse consumption. The accurate classification renal
stage of the pooling layers neural network model. Such as irregularities appear as a computer assist in the medical
Alex net to classify the diseases in terms of function renal activity [12].
images to demonstrate the CNN outclass of data in the A computerized planning to classify the various renal
optimized neural network [17]. irregularities of advanced abdominal sonography using
Kidney tumor is a disease that is effectively affected to CNN. Our advanced model provided 3 stages. The first stage
the groups of people and nowadays the tumor is commonly is used to classify the sonography images, the second stage
affected to both the genders. The classification of KT has used to classify the prototype absorbed by the renal
satisfaction in reducing rates of the death in lives, irregularities of the images and the third stage of multi-
constructing protective methods and trying to overcome the feature fusion neutral network. The equality of the abdominal
tumor in comparison to a tiresome and moment conventional normal and irregularities of renal synonyms of images. The 3
prognosis. Deep learning automatic recognition algorithms aspects that one works at the same time to identify the renal
could save specific diagnostic moments, enhance accuracy irregularities. The combined images of the dataset of the
rate, reduce the cost and reduce the radiologist workload. 4000 involved in the renal irregularities. The suggested
They classify the presence of CAT scanning. The research's feature selection helps to detect the method class in the MF-
aim is to produce 2D-CNN models for recognizing and NET architecture of median outperforms of ninety eight
categorizing KT, three variables are involved with KT percent and mean accuracy rate of ninety four percent. The
identifying the CNN-six layer, a fifty layer and hundred high designation and accuracy rate obtained identify the
layers. A new dataset from the hospital was managed to renal irregularities. The researchers say multiple-task 3D
gather which included 8400 pictures of one hundred twenty fully connected networks for ADPK classification
patients. The recognition models two dimensional CNN and demonstrate its effectiveness. Mean TKV mistake that is
recurrent neural networks achieved ninety-seven and ninety- acceptable for direct use in clinical applications following an
six, sixty percent of data. At the same time the data investigation. Researchers discovered, perhaps based on the
identification method uses ninety seven percent of the data. multiple 3D fully connected network method, classification

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on March 04,2024 at 19:13:44 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI), Jan. 23-25, 2023,
Coimbatore, INDIA
precision more than the conventional specific task 3D fully result of three models. Designers can observe that the
connected network strategy. While better contrast ADPK and Inception V3 algorithm performed with a lower accuracy
cases with both hepatic and renal cysts. Accurate values with our dataset of 62%,VGG16 has performed with the
achieved with the use of the multiple-task design suggest the accuracy of 91%. VGG19 gave accuracy of 98%
classifications can be enhanced without clearly adding data respectively. Fig3 and Fig4 represents the graph of accuracy
based on the time of target classification anatomy. rate and loss rate of kidney cysts diseases detection.
Researchers wish to analyze performance in the future by
adding additional activities such as spleen and colon A. Dataset Description
classification or even classification of unconnected The proposed system forms and analyses the "CT
anatomies such as the lung and heart. Researchers also KIDNEY Set of data in "Normal-Cyst-Tumour and Stone".
looked at how to simply resampling cross entropy. To The suggested auto-detection model for kidney disease
address these issues of class imbalance, loss works similarly diagnosis will also help in the development of a digital
to dice decline. It would be interesting to investigate its model of kidney function just at a physician as well as
ability to perform tasks with serious between and within tumour expansion. Table 2 is demonstrated about the dataset
class imbalance like aorta classification from CT images range and severity. Most of the studies are not aware of
[18]. analysis on a convert model on auto identification of renal
Stones, Tumours and Cysts.
III. PROPOSED WORK Table 2 - Dataset
Kidney irregularities are common for all age group
peoples. The manual examination is very difficult to clarify Range Types of Diseases Severity
the irregularities it costs high and time absorbing. Medical
research focused on the precisely and quickly clarify the 0-2281 Tumour 3
kidney disorders. To better recognize the kidney disorder
class that can generalize the kidney disorders. In this 2283-7359 Normal 1
proposed method CT scan image dataset where processed to
detect kidney cysts diseases by using VGG19. Fig 3 7360-8736 Stone 2
demonstrated to identify the kidney cysts, normal, tumour,
stone using VGG19 proposed work. Next step is pre-
8737-12445 Cysts 0
processing; it includes image resizing, augmentation of the
data, normalization and splitting the input image, converted
into matrix format to process the image. It moves to the
convolution process. It is used to reduce the size of an image The major contributions of this work are as follows: The "CT
to process the image by neural networks; it minimizes the KIDNEY dataset contains 12,500 images collected and
image size to reduce the image weight. Neural network analyzed using the whole abdomen and the urogram
carries the processed image to the pooling layer. The feature protocol. Tree Convolution neural deep learning techniques
map that is generated from the pooling layer converts into a VGG19 are used to detect kidney abnormalities. Fig 4
one dimensional feature vector. Dense layers connect all demonstrated the prediction of the normal kidney and
other layers to match the size of the feature map. SoftMax abnormal kidney. The model performances are presented
activation used to carry the image data from the last layer to using the confusion matrix, accuracy, sensitivity, specificity,
the output layer without any damages. Next process is the and F1 score.
classification that detects the diseases from the input image

Fig. 4. Prediction
In this analysis, the accuracy, recall, precision and f1 score
are evaluated with different models.
Fig. 3. Block diagram to diagnosis kidney cysts, normal, Table 3 summarizes the result of three models. Designers can
stone, tumor observe that the Inception V3 algorithm performed with a
lower accuracy with our dataset of 62%, VGG16 has
IV. RESULTS AND DISCUSSION performed with the accuracy of 91%. VGG19 gave accuracy
of 98% respectively. Fig5 and Fig6 represents the graph of
In this analysis, the accuracy,recall,precision and f1 score are accuracy rate and loss rate of kidney cysts diseases detection.
evaluated with different models . Table 4 summarizes the

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on March 04,2024 at 19:13:44 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI), Jan. 23-25, 2023,
Coimbatore, INDIA
Table 3 - Performances Metrics V. CONCLUSION AND FUTURE WORK

Ac Class Precision Reca F1 AUC

For this research work, collected and analysed 12,500 images
cu ll score of abdominal and grogram CT scan images with
ra stone,cysts,normal and tumour. The well-known deep
cy learning model VGG19 is used. Compared with other
Cysts 0.91 0.94 0.92 0.91 previously used models vgg16 and inceptionv3 it gives
VGG 91 better accuracy providing 98%. The VGG19 model is great
16 % Normal 0.91 0.91 0.93 0.92 to use to detect the kidney stone,cysts,normal and tumours
and it can be useful for reducing the pain of patients.
Stone 0.71 0.91 0.93 0.94

Tumour 0.92 0.93 0.92 0.91 REFERENCES

[1] P. Ongsulee, "Artificial intelligence, machine learning and deep
Cysts 0.64 0.74 0.92 0.72 learning," 2017 15th International Conference on ICT and Knowledge
Incept 61 Engineering (ICT KE), 2017, pp. 1-6, doi:
ion % Normal 0.54 0.83 0.73 0.67 10.1109/ICTKE.2017.8259629.
V3 [2] Carbonell, Jaime G., Ryszard S. Michalski, and Tom M. Mitchell.
Stone 0.54 0.57 0.63 0.53 "An overview of machine learning." Machine learning (1983): 3-23
[3] LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. "Deep learning."
nature 521.7553 (2015): 436-444.
Tumour 0.67 0.56 0.79 0.56
[4] Deng, Li, and Dong Yu. "Deep learning: methods and applications."
Foundations and trends® in signal processing 7.3 -- 4 (2014): 197-
Cysts 0.98 0.98 0.97 0.98 387.
VGG 98
19 % [5] O'Shea, Keiron, and Ryan Nash. "An introduction to convolutional
Normal 0.97 0.97 0.98 0.98 neural networks." arXiv preprint arXiv:1511.08458 (2015)
[6] Edvardsson, Vidar O., et al. "Temporal trends in the incidence of
Stone 0.98 0.98 0.98 0.98 kidney stone disease." Kidney international 83.1 (2013): 146-152.
[7] Islam, Md Nazmul, et al. "Vision transformer and explainable transfer
Tumour 0.98 0.98 0.98 0.98 learning models for auto detection of kidney cyst, stone and tumor
from CT-radiography." Scientific Reports 12.1 (2022): 1-14.
[8] Wu, Yu, and Zhang Yi. "Automated detection of kidney
abnormalities using multi-feature fusion convolutional neural
networks." Knowledge-Based Systems 200 (2020): 105873
[9] Gregory, Adriana V., et al. "Semantic Instance Segmentation of
Kidney Cysts in MR Images: A Fully Automated 3D Approach
Developed Through Active Learning." Journal of Digital Imaging
34.4 (2021): 773-787.
[10] Sudharson, S., and Priyanka Kokil. "Computer-aided diagnosis
system for the classification of multi-class kidney abnormalities in the
noisy ultrasound images." Computer Methods and Programs in
Biomedicine 205 (2021): 106071
[11] Daniel, Alexander J., et al. "Automated renal segmentation in healthy
and chronic kidney disease subjects using a convolutional neural
network." Magnetic Resonance in Medicine 86.2 (2021): 1125-1136.
[12] Alzu’bi, Dalia, et al. "Kidney Tumor Detection and Classification
Based on Deep Learning Approaches: A New Dataset in CT Scans."
Journal of Healthcare Engineering 2022 (2022).
[13] Priyanka, and Dharmender Kumar. "Kidney image classification
Fig. 5. Accuracy Rate for kidney cysts diseases using transfer learning with convolutional neural network."
International Journal of Computational Vision and Robotics 12.6
(2022): 595-613.
[14] Keshwani, Deepak, Yoshiro Kitamura, and Yuanzhong Li.
"Computation of total kidney volume from CT images in autosomal
dominant polycystic kidney disease using multi-task 3D convolutional
neural networks." International Workshop on Machine Learning in
Medical Imaging. Springer, Cham, 2018.

Fig. 6. Loss Rate for kidney cysts diseases

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on March 04,2024 at 19:13:44 UTC from IEEE Xplore. Restrictions apply.