Retinal Blood Vessel Segmentation using

Convolutional Neural Networks

Arun Kumar Yadav

, Arti Jain

, Jorge Luis Morato Lara

and Divakar Yadav

Department of Computer Science & Engineering, NIT Hamirpur, Himachal Pradesh, India

Department of Computer Science & Engineering, Jaypee Institute of Information Technology, Noida, Uttar Pradesh, India

Department of Computer Science, Universidad Carlos III de Madrid, Leganes, Madrid, Spain

Keywords: Blood Vessel Segmentation, Convolutional Neural Networks, CLAHE, Diabetic Patients, Retinal Images.

Abstract: Human beings often become victims to numerous diseases. Among these, diabetes stands out for its

impairment of quality of life and even potential mortality. The diabetes needs to be properly taken care of,

otherwise failure to detect its presence within proper time duration leads to a loss of life. According to the

World Health Organization, the worldwide number of diabetic patients were 463 million during 2019 and is

expected to cross 700 million by the 2045

. In the past, a lot of research has been carried out for retinal blood

vessel segmentation for identification of Diabetic Retinopathy using various machine learning and deep

learning models. In this research work, Convolutional Neural Network (CNN) and CLAHE are applied to

tackle the problem of retinal blood vessel segmentation. Experimental evaluation shows that the proposed

method outperforms with 0.9806 accuracy, quite competitive with respect to the state-of-art.

1 INTRODUCTION

Over the time, as age passes by, the human beings are

entangled in the clutches of some familiar or

unknown diseases. A major part of treatment for any

disease involves an identification and detection of

that disease. Although there are many techniques that

are responsible for the disease detection, however, in

most of the cases by the time the disease is detected,

it is already at an acute stage, and ultimately it

becomes un-curable, leading to the adverse

circumstances. On the other hand, there are diseases

like Diabetes where the mentioned tests are not so

effective to diagnose its presence at an earlier stage.

As is stated by (Fong et al., 2004) the Diabetes is

one of the prime reasons for blindness among the age

group of 20-74 years. This means that the human

retinas become blotted to such an extent, so that the

person’s vision goes down. During diabetes, the

blood sugar level increases in the body to a very high

level. Over a period of time can damage the retinal

blood vessels and make them swell up. In some cases,

https://orcid.org/0000-0001-9774-7917

https://orcid.org/0000-0002-3764-8834

https://orcid.org/0000-0002-7530-9753

https://orcid.org/0000-0001-6051-479X

the change is even visible to the naked eye. Also,

possibility arises that it can block the blood from

passing through the retinas, thus making the person

blind. There are Type-1 and Type-2 kinds of diabetic

patients. In the Type-1, initially destroys beta cells

from the human body which causes in the stoppage of

the insulin production. In the Type-2, the human body

still produces insulin but it is unable to use it

effectively. There are Type-1 (3.6%) patients, and

Type-2 (1.6%) patients that are even blind (Dabelea

et al., 2014) by blotting the retinal blood vessels. The

retinal layer contains many nerves or photo-receptors

that respond to light, thus enabling the person to see.

These photo-receptors transmit optic nerve sensors

and are converted into visual images. Most people do

not care about eye conditions until it becomes too bad

to the extent that the person becomes blind. Retinal

disease is a visual disorder which ultimately even

leads to blurriness in the eyes. Researchers (Bell et al.

2014) have stated that 280 million people in the world

live with the visual impairments, 34 million among

them are blind and 246 million with low vision. In

292

Yadav, A., Jain, A., Morato Lara, J. and Yadav, D.

Retinal Blood Vessel Segmentation using Convolutional Neural Networks.

DOI: 10.5220/0010719500003064

In Proceedings of the 13th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2021) - Volume 1: KDIR, pages 292-298

ISBN: 978-989-758-533-3; ISSN: 2184-3228

2019, the number of diabetic patients touches 463

million, and is expected to cross 700 million by the

2045

. The visual impairment can be prevented with

the early detection of the retinal detachment.

To work in this direction, here Convolutional

Neural Network (CNN) and CLAHE are applied to

dive into the retinal blood vessel segmentation. The

proposed method outperforms with 0.9806 accuracy

which is quite competitive with respect to the state-

of-art works. Thus, the research contributions are

stated here.

RC1: The optimized CLAHE + CNN model is

trained and tested on the publically available

DRIVE dataset and compared the result with

state-of art.

RC2: Performance evaluation is better than other

approaches over the DRIVE dataset.

Rest of the paper is organized as follows. Section 2

talks about related work, Section 3 discusses

proposed work, Section 4 illustrates results &

analysis, and Section 5 concludes the paper.

2 RELATED WORK

Researchers (Chaudhuri et al., 1989) have used two

dimensional filters to analyze the blood vessel

segmentation. A feature-removal operator based on

visual and real-estate properties is notified. The gray

cross section of artery is measured by Gaussian curve.

The signal acquisition determines the successive

sequence of blood vessels in the images. There are 12

varied templates that are created to search for ship-

parts in the available directions.

(Marın et al., 2010) have introduced approach for

finding blood vessels in the digital images of eyes.

Their method uses Neural Network (NN) and

incorporates 7-D vector with grey and secondary

features that are built into the pixel system

representation. It is tested on the publicly available

datasets- DRIVE and STARE. These datasets are

commonly used because they contain images of the

retina where the structure of blood vessels is

accurately marked by experts.

(Fraz et al., 2012) have learnt the blood vessel

segmentation problem using a hybrid method which

uses the concept of acquisition of vessel centerline for

the first order from the Gaussian. Also, observed

retrieval vessels while connecting the center lines

with a map of the position itself. They are able to get

97% accuracy. This paper reports the default way to

separate blood vessels in the images above. A unique

combination of ship acquisition techniques and slow-

flight aircraft are introduced to remove the artery tree

from the images. Mathematics of morphology is

already a good way to measure blood vessels in the

retina.

(Odstrcilik et al., 2013) have identified the

development of a two-way filter that is consistent

with Gaussian’s role as a character. They have

explored the path to STARE, DRIVE and HRF’s

decision-making process. They have designed five 2-

D filters according to standard container category

profiles and watched five shipwrecks from the

thinnest to the too big. Each image is processed to

match one in the five characters. Because the same

filters in each image, the process is slower. The

method effectiveness lies upon how accurate team of

ships is divided into the five parts.

(Sreejini and Govindan, 2015) have applied

improvised multi-scale matched filter for the blood

vessel segmentation using Particle Swarm

Optimization (PSO) (Jain et al., 2021). The same

filtering concept is widely used in the area of retinal

detachment. Multi-scale standard filters have higher

performance than the single-scale filters. The method

uses advanced audio compression features of

international filters.

(Singh and Srivastava, 2016) have utilized second

hand derivative of Gaussian as a filter for retinal

blood vessels segmentation. The method

corresponding to the measured filter are simple and

effective. However, the corresponding filter test

detects both vessels and non-vessel terminals which

gives false ships i.e., non-ship acquisitions. To

overcome the problem of finding non-ship edges, an

extension of the matching filter based on the Second

Gaussian discovery (SDOG-MF) is considered useful

for the separation of small and narrow retinal blood

vessels.

(Yao et al., 2016) have discussed CNN based

algorithm. Each pixel and its image neighbors are

tested by the CNN. The effects of the first

classification of fundus images are refined by two

phases of binarization and morphological

performance respectively. The algorithm is tested

over the DRIVE database. The data sensitivity is

0.7731, which is very close to that of the text

annotation.

(Sun et al., 2017) have illustrated four CNN

architectures (AlexNet, GoogLeNet, VGG-16, and

ResNet-50) from ImageNet image classification task

to Retinal fundus images quality classification. The

top two networks are picked out and then jointly fine-

tune them. The accuracy for different methods are

found to be as AlexNet (96.53%), GoogLeNet

(97.04%), VGG-16 (96.87%), ResNet-50 (96.20%),

Retinal Blood Vessel Segmentation using Convolutional Neural Networks

293

Joint CNN with GoogLeNet (97.00%), and Joint

CNN with VGG-16 (97.12%) repectively.

(Jebaseeli et al., 2019) have pre-processed retinal

blood vessels dataset through CLAHE- Contrast

Limited Adaptive Histogram Equalization, feature

vectors through TPCNN- Tandem Pulse Coupled

Neural Network, classification and extraction through

DLBSVM- Deep Learning Based Support Vector

Machine. It gives improved segmentation results of

sensitivity, specificity and accuracy as 0.7445,

0.9940, and 0.9897 accuracy respectively. However,

certain issues are quite unclear such as how deep

learning learns, whether neural net relays upon

certain images with the datasets, are they non-

transferable to all retinal photography. These are

limitations of accuracy for the stated technique.

(Wang et al., 2019) have illustrated novel

separation framework and a strong cascade separation

of the retinal vessel. Unlike other non-linear

partitions that require a pre-defined non-linear kernel

or repeated training, a separate cascade editing

framework is trained through the process of a single

pass transfer. Therefore, degree of non-compliance

with the separate line is not defined in advance, but is

determined by the complexity of data structures.

(Saroj et al., 2020) have worked upon a matched

filter approach with kernel as a Fréchet probability

distribution function. It uses principal component

analysis for the color conversion and CLAHE during

pre-processing. While during post-processing, it uses

entropy based optimal thresholding, and filtering by

the length. The specificity, sensitivity and accuracy

for the STARE- 0.9724, 0.7278, 0.9509 and for the

DRIVE- 0.9761, 0.7307, 0.9544 respectively.

(Escorcia-Gutierrez et al., 2021) have worked

upon Portfolio Theory of Markowitz for diabetic

retinopathy via optic disc. It produces an innovative

color fusion model which is applied over DRIVE,

Messidor, HRF, and in–house dataset (Hospital

Universitari Sant Joan de Reus, Spain). It gives an

accuracy and overlap as 0.9 and 0.80 respectively

with minimal execution time of 0.05seconds.

(Gegundez-Arias et al., 2021) have presented a

robust CNN based on UNet vessel

segmentation method in fundus images. It combines

residual blocks and batch normalization in the up-

down scaling. From actual images, patches are

extracted and trained with loss function while looking

at every pixel distance towards vascular tree, while

output produces binarized probability map of pixels.

The method is experimented over DRIVE, STARE

and CHASE_Db1.

(Pal et al., 2021) have proposed the Twin network

retinal scan system which extracts feature maps of

both query and database samples from the deep CNN.

The approach exploits deep features without the

resource, space and computation exhaustive network

training phase. The different variations of retrieval

performances are evaluated- AMD-Normal, DME-

Normal, AMD-DME, AMD-DME-Normal. The

system retrieves similar scans from a dataset of

abnormal and normal retinal scans with precision

(0.7571).

(Rajagopalan et al., 2021) have worked upon the

CNN model for the detection of retinal disorders. The

model classifies three types of retinal disorders-

Choroidal neovascularization (CNV), Drusen

macular degeneration (DMD) and Diabetic macular

edema (DME). It provides an accuracy (0.9701),

sensitivity (0.9343), and specificity (0.9807)

respectively.

3 PROPOSED WORK

The architecture of the proposed Retinal Blood

Vessel Segmentation method is described in Figure 1.

Figure 1: Basic architecture of proposed work.

The DRIVE dataset is fed into the system which

undergoes Pre-processing, namely- Image

Conversion, Image Enhancement (CLAHE) and

feature extraction (morphological). The image

conversion, converts the dataset images from colour

to grey format, applies the equalizer CLAHE for

converting the image to contrast form i.e.

KDIR 2021 - 13th International Conference on Knowledge Discovery and Information Retrieval

294

Enhancement. The feature extraction phase

undergoes morphological feature, which is followed

by Segmentation using Convolutional Neural

Networks.

3.1 Dataset Discussion

The proposed Retinal Blood Vessel Segmentation

model is executed over the chosen DRIVE dataset

(Staal et al., 2004). The DRIVE dataset contains data

from the diabetic program of Netherlands which is

collected from 400 people within age of 25-90 years.

This wide age group is considered to avoid model

overfitting and is a potential age range for the diabetic

patients. This dataset is also considered by several

researchers (Yao et al., 2016; Albargathe et al., 2021;

Escorcia-Gutierrez et al., 2021) and comprises of 20

training and 20 test images respectively.

3.2 Pre-processing

Once the DRIVE dataset is extracted, the system

undergoes pre-processing phase as follows.

3.2.1 Colour to Grey Conversion

The grey scaling process converts an image from

different colour spaces into the shades of grey, as is

seen in Figure 2.

(a) Colored Image (b) Grey Scale Image

Figure 2: Color to grey scale conversion.

The coloured spaces are such as RGB, CMYK,

and HSV etc. that are converted into the shades of

grey- i.e., in between complete black and complete

white. The python programming based OpenCV

library is used to perform the grey scale conversion.

The primary reason of doing this conversion is to

provide accurate results and faster processing, as

there are only two values- 0 and 1 in the resultant

colour histogram.

3.2.2 Applying Clahe

CLAHE is an adjustable extension of Histogram

Equalization which is followed by the threshold. It

helps in preserving the local contrast characteristics

of an image dynamically.

In the initial steps, stress is on the local contrast

rather than global contrast of the image. The global

histogram balance doesn’t stress on local contrast

enhancements, and subsequently have minor contrast

differences which is extremely common in NPDR

imagery and is totally missed if the quantity of pixels

falling in a specific dark shade is too little. To take

care of this issue, the proposed calculation is

characterized to work adaptively on the image that is

to be improved, not like normal standard histogram

adjustment. It improves the contrast upgrade on local

image data in a divide and overcome way,

subsequently effectively handles the global distortion

of the image.

All in all, the fundamental thought of the

calculation is to separate the image into various little,

non-covering context oriented areas which are called

Tiles (Figure 3).

(a) Original Image (b) Enhanced Image

Figure 3: Image enhancement using CLAHE.

3.2.3 Feature Extraction

In the feature extraction phase, the morphological

features of the images of an eye are fed into the

machine learning model to obtain the relevant results.

The encoder-decoder approach is used to convert an

Retinal Blood Vessel Segmentation using Convolutional Neural Networks

295

image data into embedding’s which is fed into the

Convolutional Neural Networks.

3.3 Segmentation

Identifying the sets of pixels which can go together is

the problem of image segmentation. Pixel-level

categorization is another name for this method. To put

it in another way, it comprises dividing images into

many segments or objects. One of the most important

applications in the computer vision area is image

segmentation (Moccia et al., 2018). It is used in a

variety of fields, including medicine and intelligent

transportation.

Image segmentation has advanced dramatically

with the deep neural networks (Siddiqui et al., 2020).

Among the deep learning concepts, CNN which is

also known as CovNet is useful for image detection

and recognition (Noh et al., 2019). The CNN inputs

image data, passes it through different layers-

convolutional, pooling, flattened and fully connected

layers and classifies data into appropriate category.

3.3.1 Convolutional Layer

It takes input as an image, extracts features and

preserves relationship among image pixels. It stores

the feature of an image as different sizes of matrix

then reduces the size of the matrix by taking filters of

varied sizes. The obtained reduced matrix is called as

Feature Map.

3.3.2 Non-Linear Activation Function

It uses non-linear activation function i.e., ReLU-

Rectified Linear Unit, mathematical function which

is given in equation (1).

(1)

3.3.3 Pooling Layer

It reduces the dimension of features map while

retaining important features of the image. The

pooling layers contains several variations like max

pooling, average pooling, and min pooling. In this

work, max pooling is used which contains maximum

values of the feature maps.

3.3.4 Flattening Layer

It is used to flatten the pooled matrix to its

corresponding vector which is serving as an input for

fully connected layer.

3.3.5 Fully Connected Layer

It contains several layers with many nodes. It

contains an input layer (first layer), output layer (last

layer), and hidden layers (other layers). In the fully

connection, each node is connected to every node of

its next layer. In the last layer, sigmoid function

categorizes the image to its most accurate category

while within rest of the layers, ReLU function is

applied.

4 RESULTS & ANALYSIS

The results of the proposed method are computed

mainly in terms of the model accuracy.

4.1 Model Accuracy

On training over the DRIVE dataset, using encoder-

decoder of CNN, the model gives an accuracy of

0.9806.

4.2 Training and Validation

The accuracy curve and loss curve are illustrated here.

4.2.1 Accuracy Curve

The training and validation accuracy curve of the

method is described in Figure 4.

Figure 4: Training and validation accuracy curve.

4.2.2 Loss Curve

The training and validation loss curve of the

proposed method is described in Figure 5.









outputnegativeo

outputpositive

xxf

),0max()(

KDIR 2021 - 13th International Conference on Knowledge Discovery and Information Retrieval

296

Figure 5: Training and validation loss curve.

4.3 State-of-Art Comparison

The proposed work is compared with respect to the

state-of-art by the other researchers (Table 1).

Table 1: Comparative analysis over DRIVE dataset.

S.No. References Method Accuracy

1 Marın et al.,

2010

Neural

Networks

0.9452

Fraz et al.,

2012

Hybrid

Approach

0.9430

3 Sreejini and

Govindan,

2015

Multi-scale

Matched Filter

0.9633

4 Singh and

Srivastava,

2016

SDOG-MF 0.9645

Yao et al.,

2016

CNN 0.9360

Soomro et al.,

2018

Independent

Component

Analysis

0.9530

Soomro et al.,

2019

FCM + CNN 0.9590

Wang et al.,

2019

Cascade

Classification

0.9541

Jebaseeli et

al., 2019

Tandem Pulse

CNN

0.9897

10 Saroj et al.,

2020

Frechet PDF

0.9544

11 Albargathe et

al., 2021

H-Minima

0.9672

12 Escorcia-

Gutierrez et

al., 2021

Markowitz 0.9556

13 Gegundez-

Arias et al.,

2021

Modified UNet

0.9547

14 Proposed

Work

CLAHE +

CNN

0.9806

5 CONCLUSIONS

In this work, CLAHE + CNN is applied for the retinal

blood vessel segmentation of images over the DRIVE

dataset. The method undergoes pre-processing- grey

scale conversion and CLAHE, feature extraction

using morphological feature, segmentation, training

and prediction using CNN. The results are evaluated

in terms of the model accuracy as 0.9806 which is

quite competitive with respect to the state-of-art work

over the DRIVE. Because of the ease-to-use and good

performance, the proposed method accelerates the

diagnosis of Diabetic Retinopathy. In future, model

accuracy can be enhanced further using more deep

learning strategies.

REFERENCES

Albargathe, S. M. B. K., Kamberli, E., Kandemirli, F., and

Rahebi, J., (2021). Blood vessel segmentation and

extraction using h-minima method based on image

processing techniques. Multimedia Tools and

Applications, 80(2): 2565-2582.

Bell, J. A., Kivimaki, M., and Hamer, M. (2014).

Metabolically healthy obesity and risk of incident type

2 diabetes: A meta‐analysis of prospective cohort

studies. Obesity Reviews, 15(6): 504-515.

Chaudhuri, S., Chatterjee, S., Katz, N., Nelson, M., and

Goldbaum, M. (1989). Detection of blood vessels in

retinal images using two-dimensional matched filters.

IEEE Transactions on Medical Imaging, 8(3): 263-269.

doi: 10.1109/42.34715.

Dabelea, D., Mayer-Davis, E. J., Saydah, S., Imperatore,

G., Linder, B., Divers, B. J., Bell, R., Badaru, A.,

Talton, J. W., Crume, T., and Liese, A. D. (2014).

Prevalence of type 1 and type 2 diabetes among

children and adolescents from 2001 to 2009.

Jama, 311(17): 1778-1786.

Escorcia-Gutierrez, J., Torrents-Barrena, J., Gamarra, M.,

Romero-Aroca, P., Valls, A., and Puig, D. (2021). A

color fusion model based on markowitz portfolio

optimization for optic disc segmentation in retinal

images. Expert Systems with Applications, 174, 14 697.

Fong, D. S., Aiello, L., Gardner, T. W., King, G. L.,

Blankenship, G., Cavallerano, J. D., Ferris, F. L., and

Klein, R. (2004). Retinopathy in diabetes. Diabetes

Care, vol. 27, no. suppl 1, s84–s87.

Fraz, M. M., Barman, S. A., Remagnino, P., Hoppe, A.,

Basit, A., Uyyanonvara, B., Rudnicka, A. R., and

Owen, C. G. (2012). An approach to localize the retinal

blood vessels using bit planes and centerline detection.

Computer Methods and Programs in Biomedicine,

108(2): 600–616.

Gegundez-Arias, M. E., Marin-Santos, D., Perez-Borrero,

I., and Vasallo-Vazquez, M. J. (2021). A new deep

learning method for blood vessel segmentation in

Retinal Blood Vessel Segmentation using Convolutional Neural Networks

297

retinal images based on convolutional kernels and

modified u-net model. Computer Methods and

Programs in Biomedicine, 205, 106 081.

Jain, A., Yadav, D., and Arora, A. (2021). Particle swarm

optimization for Punjabi text summarization.

International Journal of Operations Research and

Information Systems (IJORIS), 12(3): 1-17. IGI Global.

Jebaseeli, T. J., Durai, C. A. D., and Peter, J. D. (2019).

Segmentation of retinal blood vessels from

ophthalmologic diabetic retinopathy images.

Computers & Electrical Engineering, 73: 245-258.

Marın, D., Aquino, A., Gegu´ndez-Arias, M. E., and Bravo,

J. M. (2010). A new supervised method for blood vessel

segmentation in retinal images by using gray-level and

moment invariants-based features. IEEE Transactions

on Medical Imaging, 30(1): 146–158.

Moccia, S., Momi, E. De., Hadji, S. El., and Leonardo, S.

(2018). Mattos Blood vessel segmentation algorithms-

Review of methods, datasets and evaluation metrics.

Computer Methods and Programs in Biomedicine,

158: 71-91.

Noh, K. J., Park, S. J., and Lee, S. (2019). Scale-space

approximated convolutional neural networks for retinal

vessel segmentation. Computer Methods and Programs

in Biomedicine, 178: 237-246.

Odstrcilik, J., Kolar, R., Budai, A., Hornegger, J., Jan, J.

Gazarek, J., Kubena, T., Cernosek, P., Svoboda, O., and

Angelopoulou, E. (2013). Retinal vessel segmentation

by improved matched filtering: Evaluation on a new

high-resolution fundus image database. IET Image

Processing, 7(4): 373–383.

Rajagopalan, N., Venkateswaran, N., Josephraj, A. N., and

Srithaladevi, E. (2021). Diagnosis of retinal disorders

from optical coherence tomography images using

CNN. PLOS One, 16(7), e0254180.

Saroj, S. K., Kumar, R., and Singh, N. P. (2020). Frechet

pdf based matched filter approach for retinal blood

vessels segmentation. Computer Methods and

Programs in Biomedicine, 194, 105 490.

Siddiqui, F., Gupta, S., Dubey, S., Murtuza, S., and Jain, A.

(2020). Classification and diagnosis of invasive ductal

carcinoma using deep learning. In Proceedings of the

10th International Conference on Cloud Computing,

Data Science & Engineering (CONFLUENCE 2020),

242-247. IEEE.

Singh, N. P. and Srivastava, R. (2016). Segmentation of

retinal blood vessels by using a matched filter based on

second derivative of Gaussian. International

Journal of Biomedical Engineering and Technology,

21(3): 229-246.

Soomro, T. A., Khan, T. M., Khan, M. A., Gao, J., Paul, M.,

and Zheng, L. (2018). Impact of ICA-based image

enhancement technique on retinal blood vessels

segmentation. IEEE Access, 6, 3524-3538.

Soomro, T. A., Afifi, A. J., Shah, A. A., Soomro, S., Baloch,

G. A., Zheng, L., Yin, M., and Gao, J. (2019). Impact

of image enhancement technique on CNN model for

retinal blood vessels segmentation. IEEE Access, 7,

158183-158197.

Sreejini, K., and Govindan, V. (2015). Improved multiscale

matched filter for retina vessel segmentation using PSO

algorithm. Egyptian Informatics Journal, 16(3): 253-

260.

Staal, J., Abràmoff, M. D., Niemeijer, M., Viergever, M.

A., and Ginneken, B. Van. (2004). Ridge-based vessel

segmentation in color images of the retina. IEEE

Transactions on Medical Imaging, 23(4): 501-509.

Sun, J., Wan, C., Cheng, J., Yu, F., and Liu, J. (2017).

Retinal image quality classification using fine-tuned

CNN. In Fetal, Infant and Ophthalmic Medical Image

Analysis, pp. 126-133. Springer.

Wang, X., Jiang, X., and Ren, J. (2019). Blood vessel

segmentation from fundus image by a cascade

classification framework. Pattern Recognition,

88: 331-341.

Yao, Z., Zhang, Z., and Xu, L. Q. (2016). Convolutional

neural network for retinal blood vessel segmentation. In

Proceedings of the 9th International Symposium on

Computational Intelligence and Design (ISCID), 1:

406-409. IEEE.

END NOTES

Facts & Figures: Accessed on Feb 2021.

https://www.idf.org/aboutdiabetes/what-is-diabetes/facts-

figures.html

KDIR 2021 - 13th International Conference on Knowledge Discovery and Information Retrieval

298