International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 4, Jul-Aug 2020

RESEARCH ARTICLE OPEN ACCESS

A Review on Diabetic Retinopathy detection using Deep

Learning Methods
[Link]
Department of ECE, Excel College of Engineering and Technology
Namakkal, Tamilnadu, India

ABSTRACT
Diabetic retinopathy (DR) is a common type of diabetes in which people damage the blood vessels in the retina. The disease
occurs in malformations, beginning with microerosion, in the non-proliferative phase until the tumor appears in the reproductive
phase. Retinal specialists are urgently trying to get DR so that the disease can be treated before significant permanent vision loss
occurs. The severity of the DR specifies the processing effort required. Effective diabetes treatment in the mild (early) phase
prevents vision loss rather than invasive laser surgery. Artificial Intelligence (AI) can be used to develop efficient, high-precision
systems that health care providers can use to diagnose and diagnose DR without unrestricted access to resources for specialist
clinics. In specific, deep learning simplifies early diagnosis and increases specificity and sensitivity. Such schemes make
decisions based at least on boat characteristics and pave the way for personalized therapy. Therefore, this overview delivers a
detailed report of the current knowledge that is used in all phases of DR diagnostics. The first step is to introduce the technologies
and resources available in this area into the disease. A structure is then explained that identifies and classifies the various
commands. Finally, we conclude that DR systems and deep learning systems are revolutionary to prevent vision loss.
Keywords: Image Mining,Diabetic Retinopathy,Deep Learning, and Lesion Detection

I. INTRODUCTION especially convolutional neural networks (CNN),[11] define

actions and rules that correspond to taxonomic accuracy
DR is a difficulty of diabetes that can prime to with minimal manual components.
sightlessness if left untreated.[1] The limited resources
available to diabetics and the need for regular check-ups can Treatments such as laser surgery or intravenous
be very stressful for patients.[2] DR is categorized as lack of inoculations[12] of vascular endothelial growing factor
proliferation (NPDR), which includes microanalysis (MA), deliver an effective way to prevent visualization loss from
exudates, or an enlargement (PDR) of weakened blood DR or diabetic macular edema (DME) if the disease is
vessel neovascularization.[3] More accurate systems can noticed early.[13] However, due to limited human-technical
help determine DR early, so patients can seek treatment and resources, many are unable to perform timely
evade blindness.[4] Manual diagnoses struggle to achieve inspections[14]. Conventional imaging techniques rely on
80% UK recommended sensitivity, and developed areas expensive or non-transferable technologies, thus
particularly suffer from unbalanced patient-specialist limiting[15] non-local approaches to DR diagnosis.[16] The
ratios[5]. Studies show that by the age of 20,[6] almost all problem. In such a scheme, digital images can be converted
patients with type I diabetes and 60% of patients with type to cloud storage for later viewing. However, this increase in
II diabetes experience DR. Automated processing methods accessibility reduces image quality and[17] Fied of view
have appeared to solve problems in the DR classification,[7] (FOV). Along with the limitations of specialist staff, this
allowing screening to distinguish between those who need increases the need for computrized diagnostic systems
more instructions and those who are ordered as low-risk (CAD).
groups. [8]Machine Learning Algorithms (ML).[9] Pattern In the 1990s, a new clinical standard, the first
recognition is eased by training sets that can optimize modern advances in the diagnosis and classification of DR,
solutions to algorithms in a multi-dimensional space.[10] as well as the results of using artificial neural networks
Systems vary in manual performance. DL processes, (ANNs) in the detection of DR, appeared in the ETDRS

ISSN: 2347-8578 [Link] Page 14

 International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 4, Jul-Aug 2020

study group on the ETDRS scale (1991). [18]ANN seems to This was due to an increased interest in CAD systems that
have lost popularity in this area in the early 2000s, and compensate for low FOV with knowledge such as stitching
instead, processing techniques have focused more on image and image editing. Our survey differs from previous
handling techniques and many statistical classifiers[19]. operations in that we include new DL pipelines, panoramic
This is due to the belief that existing NNs are insufficient to images, and ML recognize all DR classification activities.
detect specific DR characteristics such as DR lesions.[20] This conversation facilitates the clinical implementation of
During this period, many methods deviated from image state of the art systems. [29]Our interdisciplinary material
level diagnostics. This may be due to the fear that HP[21] will provide research teams with a source to study and
will learn about the risk factors for SR or trivial symptoms, understand these systems.[30] Ultimately, the DL
and not about the underlying disease. Advanced learning framework offers special offerings for customizable and
systems have become important in the field of artificial patient-specific therapeutic and diagnostic treatments.
vision.[22] In particular, Google's work on classifying and
classifying DRs has shifted the focus of this space to an in-
depth study of algorithms.[23] Recent work demonstrates II. LITERATURE RESOURCES
DR's ability to make predictions in more complex and OVERVIEW
therapeutically important directions. For example,[24]
recent work on nature has produced excellent results leading 2.1 Datasets Description
to longitudinal forecasting and AR observation. Despite
these advances, there has been slow development in the Fundus Image Registration (FIRE) – There are 129
adoption and adoption of these systems in widespread use. fundus images and a total of 134 image pairs. These data
The United States Federal Narcotics Agency (USFDA) form three subgroups based on visual characteristics[31].
approved the first artificial intelligence detection system for The files include pairs of images and related key facts, a
DR. Despite this, the recognition system was of limited use. ROI mask (binary color of interest) and a ROI mask.
For instance, other eye diseases or diseases with severe DR Structured Analysis of the Retina (STARE) – the
are not comprised in accepted use cases. corresponding diagnosis can be accessed separately in the
TXT file. The second zip file encloses explanations of all
Investigators should take into account advances in DR images as text files. This includes advanced deformities and
visualization methodology when developing appropriate local characteristics[32]. The vascular class includes
artificial intelligence systems for identifying and classifying vascular cards that are labeled with 40 hands. 80 images
DR. Traditionally, the seven standard ETDRS fields are the recognize the optical nerve and show the basic truth. The
gold standard for DR detection. The standard stereoscopic entire data set offers a variety of illnesses and injuries.
images,[25] including the fiber optic disc, macula, temporal However, the definition system is somewhat difficult. In
macula, and four vascular port. Although this method is comparison, the division of the blood vessels provides a
related to DR detection, it only includes 75 approaches, and more precise identification of the arteries.
these digital tools were created before imaging can be Digital Retinal Images for Vessel Extraction –It has 40
performed. In fact, a study of ultra-wide field photography images of which 33 are normal and 7 are sick. Experts have
showed that at least 0.11 eyes with DR were at least twice as labeled pixels in all 40 images for boats. [33]The 20-image
heavy as those obtained using the ETDRS [Link] test set has a single gold standard division, while the 20-
may be due to the significance of the peripheral retina in image test set has a gold standard section and a separate
noticing retinopathy symptoms such as micro vascular section.
abnormalities, neovascularization, vascular leakage and lack Retinal Vessel Image set for Estimation of Widths –
of olfactory areas.[26] Thus, there is ample evidence that comprising 16 films with 193 vehicle categories. The
diagnostics should encourage a broader understanding of the dataset has four subdivisions: high resolution, vascular
retina, especially in taxonomic operations. Currently, DR is disease, intermediate light reflex, and impact points. These
determined mainly by retinal specialists based on prefixes reflectdiverse types of diseases and differentiation
stereographic color photography, fluorescence angiography, problems.
and optical coherence tomography (OCT).[27] However, High-Resolution Fundus Image Database – It arranges the
new areas, such as fundus auto fluorescence and OCT fundus image according to the patient’s condition;
angiography (OCTA), are becoming interesting areas. These [34]There are 15 healthy glaucoma, 15 DR and 15
compatible technologies and CAD systems show great glaucoma. Binary split cards work as a golden truth
promise for identifying disease complexity. At the same category with these images.
time, the growing demand for DR shielding in low-access
areas has increased interest in using low-quality imaging
equipment[28].

ISSN: 2347-8578 [Link] Page 15

 International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 4, Jul-Aug 2020

III. OVERVIEW OF PROCESSING

APPROACHES 3.1.3. Random Forest (RF)

Image processing techniques eliminate the The KNN feature classifies objects into invisible images
limitations of DR shielding due to image quality, the based on the distance from the "K" training patterns in
imbalance between the patient and the specialist, and the space.[40] In the traditional KNN classifier, there may be a
discrepancies between them. DR assessment is done by range imbalance due to the same weight in the marked
determining the image structure,[35] but it is difficult to samples. The classification may be ambiguous during the
distinguish features such as micro probability, bleeding and interclass link, made by K.E. It can be converted to “pure”
exudate from the background of the retina. Steps can KNN by reducing or reducing the distance of each neighbor
include normalizing focus and lighting, improving contrast, to the assigned class.[41] Obscure-KNN eliminates
and reducing noise and processing. The need for consistency ambiguity by transmission input vectors to different classes.
in the wide-ranging review process and in the order of It makes decisions based on the most demanding
quality of this process; [36]In the preprocessing phase, they responsibilities
can be helpful for human interpretation or for automated .
algorithms. In addition, DM tracking improves the accuracy 3.1.4. Neural Networks (NNs)
of image processing. For example, longitudinal planning of
retinal scans enables more accurate tracking of local lesion NN learns a mathematical weight that determines
changes and other related symptoms. [37]The segmentation the potential of the input data to be suitable for a particular
of the optical disc (OD) is based on the intensity and output class of output[42]. To achieve this, NN identifies
geometry of the OD and the nearby blood vessels. Typical key features for determining class output (DR intensity) in
image processing strategies include morphological input samples of a recognized class.[43] Some handwriting
operations, pattern matching, huff wrapping, and texture is needed to set the parameters. In addition, NNs can be
cropping. Conventional methods often rely on image used to improve performance and pass them to the
processing to remove the candidate wound before classifier.[44] RNA can remove imaging features to identify
candidates are classified using machine-based or rule-based disease candidates by labeling them with rules or clinical
learning systems. classifications.[45] Alternatively, ANN can predict the
names of injuries or diseases based on input characteristics.
3.1. Traditional ML
3.2 Deep Learning (DL)
1.1.1. SVM
The DLL selects a superordinate task from four
SVMs use pattern recognition to define class separation entries in the output label and uses the training data to
rules using object expression profiles. For example, a string complete the associated tasks.[46] The most successful DL
may distinct DR classes by the presence / amount of methods were created with the popular Deep Conveying
damage[38]. When there are many objects, classes are Nerve Networks (CNN). CNN uses convection filters to
separated from the upper limit of measurement based on the convert input data and allows you to distribute the weight in
maximum distance. However, this technique is imperfect in the room.[47] DL algorithms are less specific to the user
the following points: it accepts that training and test data than conventional ML systems, but it is not clear which
comes from the same distributions, requires user-defined features influence their classification.
limits, and may usenumerous input variables.
IV. OVERVIEW OF TASKS
3.1.2. K-Nearest Neighbors (KNN)
4.1. Optic Disk (OD) Segmentation
KNN classifies objects into invisible images based on the
distance from the "K" training patterns in the feature Removing the OD recovers detection of minor
space.[39] Conventional KNN classifiers can experience a injuries due to the highest intensity values in this region as a
class imbalance due to the same weight among the labeled whole.[48] The OD edge may interfere with the detection of
samples. The classification can be ambiguous during the blood vessels.[49] Alternatively, the localization of the OD
connections between the classes, which can be resolved in can facilitate the separation of the blood vessels since the
"clear" KNN by plummeting K or lessening the sum of the main retinal hair is freed from the OD.[50] Ultimately, the
distances of each neighbor with the assigned class. Fuzzy- OD can be found to determine the distance from the bubble.
KNN eliminates uncertainty by transfer an input vector to Maculapathy or macular retinopathy is very harmful due to
different classes. It chooses on the basis of the strongest the high visual importance of this area.[51] Therefore,
adhesion. accurate macular localization may be helpful in assessing

ISSN: 2347-8578 [Link] Page 16

 International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 4, Jul-Aug 2020

severity, especially if it is hidden by exudates. In particular, suggests that wound research is more complex than system-
the expansion of DR deep learning methods is associated level DR results.[61] At the same time, differences in
with a decrease in the use of preprocessing techniques such patients in the system that diagnoses DR based on the
as OD splitting and extraction.[52] Theoretically, this presence of lesions can lead to incorrect diagnoses.[62] It
implies that fewer craft systems can distinguish combines global imaging information with wound mapping
characteristics that can distinguish OD from other for accurate and comprehensive diagnosis and treatment as
characteristics; However, due to the nature of the “black an excellent system.
box” of more in-depth training systems, it is difficult to
distinguish whether the algorithm studies the characteristics V. CONCLUSIONS
and whether it weighs correctly (that is, it strikes the
OD).[53] This can irritate patients with different OD. For DR is a high-risk diabetic complication that leads to
example, a model studied in patients with a dark OD may permanent blindness without proper treatment. Adequate AI
not have a pale OD due to trauma. Another,eliminating the scheme have the potential to facilitate quick screening and
OD prevents the algorithm from simultaneously learning related care of people suffering from this disease. Mild DR
about other significant eye illnesses[54]. Therefore, OD Symptoms, especially those transmitted by MA, greatly
segmentation is significant for detecting DR delay and benefit early detection systems. Fast and inexpensive
intensity classification, although its extraction depends on computer algorithms also provide great potential for
the planned function. applications in resource-restricted areas that would benefit
from faster screening. However, more complex and diverse
4.2. Blood Vessel (BV) Segmentation imaging and classification systems emerge that are more
likely to be identified by humans. May increase the
BV screening is required for DR screening to sensitivity and specificity of the classification. Overall, in-
minimize false positives in dark lesion detection and to track depth education is the most important promise to investigate
micro vascular changes in PDR detection.[55] In the first DR and assess its severity at the same time. Future
case, false positive results can occur due to the same pixel introduction of this algorithm will help doctors to quickly
intensity between the BV and this type of injury and the differentiate between different eye diseases with
proximity of most BV lesions. The PDR is then branded by simultaneous or similar symptoms and diagnose the stage of
angiogenesis or by the growth of a novel retinal BV without the disease so that the most relevant and useful treatments
structural honesty. These ships cannot continue to function can be carried out quickly and efficiently.
normally and are prone to explosion or permanent damage.
REFERENCES
4.3. Lesion Detection
[1]. Chi-Hua Chen, “A Cell Probe-based Method for
Wounds must be examined to identify and classify Vehicle Speed Estimation,” IEICE Transactions on
DR. These algorithms help create complete wound Fundamentals of Electronics, Communications and
maps.[56] Therefore, they increase the reliability of the Computer Sciences, vol. E103-A, no. 1, pp. 265-267,
system by allowing doctors to track which areas of the January 2020.
image are affecting the DR. In addition, dissimilar studies [2]. R. Meera, [Link] “A Review On Automatic
also focus on dissimilar lesions depending on the purpose. Detection of Brain Tumor Using Computer Aided
For example, most algorithms focus on micro-ergonomics Diagnosis System Through MRI” The Energy Green,
research to help ophthalmologists diagnose preventable Intelligent in Computing & Communication
vision loss. [57]Other methods track damage such as Technologies in Journal of Energy Web and
exudates and bleeding to aid in DR amplification. Recording Information Technologies, Vol5,Issue20,2018.
images allows the DR to track the spread of injuries and [3]. Shankar, K., Lakshmanaprabu, S. K., Gupta, D.,
visit sites based on progress indicators[58]. Lesson tracking Khanna, A., & de Albuquerque, V. H. C. (2020).
offers significant advantages over end-to-end DR Adaptive optimal multi key based encryption for
assessment. An algorithm that constantly learns from digital image security. Concurrency and
images[59] without custom input functions such as "in a Computation: Practice and Experience, 32(4), e5122.
black box" means that the user does not know what the [4]. G. Keerthana , [Link] “A Survey on Security
system is doing. Therefore, an apparently accurate system Issues and Challenges in Mobile Ad-hoc Network”
can learn incorrect information during training and fail if an The Energy Green, Intelligent in Computing &
invisible image is provided.[60] In addition, injury-specific Communication Technologies in Journal of Energy
doctors provide more accurate information about the Web and Information
procedures patients may need. In fact, accuracy works for Technologies,Vol5,Issue20,2018.
DR despite low lesion identification above 0.80. This

ISSN: 2347-8578 [Link] Page 17

 International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 4, Jul-Aug 2020

[5]. Chi-Hua Chen, Fangying Song, Feng-Jang Hwang, Based on Vehicular Streaming
Ling Wu, “A Probability Density Function Generator Communications,” Journal of Internet Technology,
Based on Neural Networks,” Physica A: Statistical vol. 20, no. 7, pp. 2083-2097, December 2019.
Mechanics and its Applications, vol. 541, Article [16]. [Link], [Link], [Link],
ID 123344, March 2020. [Link], [Link] and [Link]
[6]. [Link] , [Link], [Link] “A “Modeling of Sheet Carrier Density, DC and
Computational Approach of Highly Secure Hash Transconductance of Novel InxAl1-XN/GaN-Based
Algorithm For Color Image Steganography Using HEMT Structures” Advanced Materials Research
Edge Detection And Honey Encryption Algorithm” Vol. 1105 (2015) pp 99-104.
International Journal of Engineering & Technology, 7 [17]. Shankar, K., Zhang, Y., Liu, Y., Wu, L., & Chen, C.
PP. 239-242, 2018. H. (2020). Hyperparameter Tuning Deep Learning
[7]. Mingyang Pan, Yisai Liu, Jiayi Cao, Yu Li, Chao for Diabetic Retinopathy Fundus Image
Li, Chi-Hua Chen, “Visual Recognition Based on Classification. IEEE Access.
Deep Learning for Navigation Mark [18]. [Link],[Link],“ A Novel Whispered
Classification,” IEEE Access, vol. 8, pp. 32767- Speaker Identification System Based on Extreme
32775, February 2020. Learning Machine”, International Journal of Speech
[8]. [Link],[Link] “A Multi Objective Technology, Springer,(2018) ,21 (1), pp.157–165.
Tabu Particle Swarm Optimization for Effective [19]. Shankar, K., & Elhoseny, M. (2019). Trust Based
Cluster Head Selection in WSN” Cluster Computing, Cluster Head Election of Secure Message
Vol. 22,Issue5,12275–12282,2019. Transmission in MANET Using Multi Secure
[9]. Shankar, K., Lakshmanaprabu, S. K., Khanna, A., Protocol with TDES. Journal of Universal Computer
Tanwar, S., Rodrigues, J. J., & Roy, N. R. (2019). Science, 25(10), 1221-1239.
Alzheimer detection using Group Grey Wolf [20]. [Link], Kiranmai Bellam, “Improved
Optimization based features with convolutional Distributed Frameworks to Incorporate Big Data
classifier. Computers & Electrical Engineering, 77, through Deep Learning”, Journal of Advanced
230-243. Research in Dynamical & Control Systems, Vol. 12,
[10]. [Link], [Link], [Link] “The Survey to 03-Special Issue, [Link]-338
Implement Recent Reversible Watermarking [21]. Manickam, P., Shankar, K., Perumal, E., Ilayaraja,
Techniques In Medical Images And Other M., & Kumar, K. S. (2019). Secure data transmission
Applications” Journal of Advanced Research in through reliable vehicles in VANET using optimal
Dynamical & Control Systems, Vol.10-Special Issue lightweight cryptography. In Cybersecurity and
03, May 2018. secure information systems (pp. 193-204). Springer,
Cham.
[11]. Ling Wu, Qishan Zhang, Chi-Hua Chen, Kun Guo, [22]. [Link], Mohamed Elhoseny,
Deqin Wang, “Deep Learning Techniques for [Link],Mahmoud [Link] , [Link] ,
Community Detection in Social Networks,” IEEE “Deep Learning Model for real- time image
Access, vol. 8, pp. 96016-96026, May 2020. compression in Internet of Underwater
[12]. [Link], [Link] “A Review on Pixel Things(IoUT)”, Journal of Real-time Image
Performance in CMOS Image Sensors” Journal of Processing ,2019
Advanced Research in Dynamical & Control [23]. Shankar, K. (2017). Prediction of most risk factors in
Systems, 05-Special Issue, July 2017. hepatitis disease using apriori algorithm. Research
[13]. Chin-Ling Chen, Tsai-Tung Yang, Yong-Yuan Journal of Pharmaceutical Biological and Chemical
Deng, Chi-Hua Chen, “A Secure IoT Medical Sciences, 8(5), 477-484.
Information Sharing and Emergency Notification [24]. [Link],[Link],[Link],[Link]
System Based on Non-repudiation Kumar , “A Glove based approach to recognize
Mechanism,” Transactions on Emerging Indian Sign Languages”, International Journal of
Telecommunications Technologies, Accepted Recent Technology and Engineering (IJRTE)
Manuscript. Volume-7, Issue-6, March 2019, pp.1419-1425.
[14]. [Link], [Link],[Link] [25]. Chi-Hua Chen, “An Arrival Time Prediction Method
“Characterization of Flicker noise in Dual Material for Bus System,” IEEE Internet of Things Journal,
Gate Silicon Nanowire Transistors” Journal of vol. 5, no. 5, pp. 4231-4232, October 2018.
Nanoelectronics and Optoelectronics, 12, 72–75 [26]. [Link] ,Dr P Kiran
(2017) (Impact Factor 0.369) Kumar, Mr K Subash Bhagahavn , “Conceptual
[15]. Hsu-Yang Kung, Chi-Hua Chen, Mei-Hsien Lin, Tai- Semantic Model for Web Document Clustering
Yang Wu, “Design of Seamless Handoff Control Using Term Frequency”, EAI Endorsed Transactions

ISSN: 2347-8578 [Link] Page 18

 International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 4, Jul-Aug 2020

on Energy Web and Information Technologies, [37]. Uthayakumar, J., Elhoseny, M., & Shankar, K.
Volume 5, Issue 20,2018,pp.1-4. (2020). Highly Reliable and Low-Complexity Image
[27]. Elhoseny, M., Shankar, K., &Uthayakumar, J. Compression Scheme Using Neighborhood
(2019). Intelligent diagnostic prediction and Correlation Sequence Algorithm in WSN. IEEE
classification system for chronic kidney Transactions on Reliability.
disease. Scientific reports, 9(1), 1-14. [38]. G.S.S.S.S.V. Krishna Mohan & Yarravarapu
[28]. N. Krishnaraj, P. Ezhilarasu, X Z Gao ,”Hybrid Soft Srinivasa Rao: “Optimal Order of the Differentiator
Computing Approach for Prediction of Cancer in Selection in Noise Removal of ECG Signals”,
Colon Using Microarray Gene Data” , Current Signal International Journal of Recent Technology and
Transduction Therapy Vol.11 (2),pp71-75,June 2016. Engineering (IJRTE),Volume-7, Issue-6, 260-267,
[29]. Elhoseny, M., Bian, G. B., Lakshmanaprabu, S. K., March 2019.
Shankar, K., Singh, A. K., & Wu, W. (2019). [39]. Lydia, E. L., Raj, J. S., PandiSelvam, R., Elhoseny,
Effective features to classify ovarian cancer data in M., & Shankar, K. (2019). Application of discrete
internet of medical things. Computer Networks, 159, transforms with selective coefficients for blind image
147-156. watermarking. Transactions on Emerging
[30]. N. Krishnaraj, P. Ezhilarasu, [Link] , Manoj Telecommunications Technologies, e3771.
Prabhakar.J, ,”Enhancing Security in Mobile Devices
through Multimodal biometrics” , Middle-East [40]. [Link] Mohan and Yarravarapu Srinivasa
Journal of Scientific Research 23 (8) ,pp. 1598- Rao, “An efficient design of fractional order
1603,Jun 2016 differentiator using hybrid Shuffled frog leaping
[31]. Elhoseny, M., & Shankar, K. (2020). Energy efficient algorithm for handling noisy electrocardiograms”,
optimal routing for communication in VANETs via International Journal of Computers and
clustering model. In Emerging Technologies for Applications,Feb 2019.
Connected Internet of Vehicles and Intelligent [41]. Sivaram, A. M., Lydia, E. L., Pustokhina, I. V.,
Transportation System Networks (pp. 1-14). Pustokhin, D. A., Elhoseny, M., Joshi, G. P., &
Springer, Cham. Shankar, K. (2020). An Optimal Least Square
[32]. Krishnaraj,N.,Ezhilarasu,p., Dharun, V.S.,” Smart Support Vector Machine Based Earnings Prediction
Phone Application For Automatic Public of Blockchain Financial Products. IEEE Access.
Transportation Though Providing Intelligent Bus [42]. [Link] Mohan and [Link] Lakshmi
Status Information To The Users” International Narayana, “Design Of A Fractional Order PID For A
Journal of Applied Engineering Research (IJAER), Three Tank System”, International Journal of
Vol 59, pp.163-167, Jun -2015. Applied Engineering Research, Volume 10, Number
[33]. Krishnaraj, N., Elhoseny, M., Lydia, E. L., Shankar, 2 (2015) pp. 3133-3148, Research India Publications,
K., &ALDabbas, O. (2020). An efficient radix April 2015.
trie‐based semantic visual indexing model for [43]. Elhoseny, M., Selim, M. M., & Shankar, K. (2020).
large‐scale image retrieval in cloud Optimal Deep Learning based Convolution Neural
environment. Software: Practice and Experience. Network for digital forensics Face Sketch Synthesis
[34]. [Link] Mohan and Komanapalli in internet of things (IoT). International Journal of
Venkata Lakshmi Narayana, “Auto Tuning Smith- Machine Learning and Cybernetics, 1-12.
Predictive Control of Delayed Processes Based on [44]. [Link], [Link] Kumar, P. Kameswara Rao,
Model Reference Adaptive Controller”, Jour of Adv [Link] Mohan “Tuning Of Fractional
Research in Dynamical & Control Systems,Vol. 12, Order PID Controller For Interacting Systems By
04-Special Issue, p.p.1224-1230, 2020. Different Methods”, i--manager’s Journal on
[35]. Mohanty, S. N., Lydia, E. L., Elhoseny, M., Al Instrumentation & amp; Control Engineering Vol.2
Otaibi, M. M. G., & Shankar, K. (2020). Deep No.2 May July 2014.
learning with LSTM based distributed data mining [45]. Chi-Hua Chen, Feng-Jang Hwang, Hsu-Yang Kung,
model for energy efficient wireless sensor “Travel Time Prediction System Based on Data
networks. Physical Communication, 101097. Clustering for Waste Collection Vehicles,” IEICE
[36]. [Link] Mohan and Yarravarapu Srinivasa Transactions on Information and Systems, vol. E102-
Rao, “An efficient design of finite impulse response D, no. 7, pp.1374-1383, July 2019.
— Fractional-order differentiator using shuffled frog [46]. A,Venkata Naga Vamsi, [Link]
leaping algorithm heuristic”, International Journal of Mohan, [Link], “Simplified Thermocouple
Wavelets, Multiresolution and Information Interface For Hot Only Or Cold Only Measurement
Processing, World Scientific Publishing Company, With Linearization Circuit”, (IJERA) International
Vol. 17, No. 2 March 2019. Journal of Engineering Research and Applications,

ISSN: 2347-8578 [Link] Page 19

 International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 4, Jul-Aug 2020

Vol. 2, Issue5, September- October 2012, pp.1663- [56]. “A Cost-Aware Method for Tasks Allocation on the
1667. Internet of Things by Grouping the Submitted
[47]. Lakshmanaprabu, S. K., Shankar, K., Ilayaraja, M., Tasks”R Dhanapal, T Akila, SS Hussain, D
Nasir, A. W., Vijayakumar, V., &Chilamkurti, N. Mavaluru - Journal of Internet Technology, Volume
(2019). Random forest for big data classification in 20 (2019) No.7,Pages 2055-2062.
the internet of things using optimal [57]. Raj, R. J. S., Shobana, S. J., Pustokhina, I. V.,
features. International journal of machine learning Pustokhin, D. A., Gupta, D., & Shankar, K. (2020).
and cybernetics, 10(10), 2609-2618. Optimal Feature Selection-Based Medical Image
[48]. [Link], [Link] Mohan, Classification Using Deep Learning Model in
[Link], [Link] Rao, “An Implementation Internet of Medical Things. IEEE Access, 8, 58006-
of different non-linear PID controllers on a single 58017.
tank level control using Matlab”, (IJCA) International [58]. “Real Time Health Care Monitoring System for
Journal of Computer Applications (0975 – 8887) Driver Community Using Adhoc Sensor Network”
Volume 54– No.1, September 2012. Dhanapal, R.; Visalakshi, [Link] of Medical
[49]. Mohanty, S. N., Ramya, K. C., Rani, S. S., Gupta, D., Imaging and Health Informatics, Volume 6, Number
Shankar, K., Lakshmanaprabu, S. K., & Khanna, A. 3, June 2016, pp. 811-815(5)
(2020). An efficient Lightweight integrated [59]. Pustokhina, I. V., Pustokhin, D. A., Rodrigues, J. J.,
Blockchain (ELIB) model for IoT security and Gupta, D., Khanna, A., Shankar, K., & Joshi, G. P.
privacy. Future Generation Computer Systems, 102, (2020). Automatic Vehicle License Plate Recognition
1027-1037. using Optimal K-Means with Convolutional Neural
[50]. Dhanapal, R & Visalakshi, P 2016, Real Time Health Network for Intelligent Transportation
Care Monitoring System for Driver Community Systems. IEEE Access.
Using Adhoc Sensor Network‟, Journal of Medical [60]. “A Sector Based Energy Efficient Adaptive Routing
Imaging and Health Informatics, ISSN 2156-7018, Protocol for Large Scale MANET” R Dhanapal, P
vol. 6, no. 3, pp. 811-815. Visalakshi - Research Journal of Applied Sciences,
[51]. Kathiresan, S., Sait, A. R. W., Gupta, D., Engineering and Technology,volume 9(7): pages
Lakshmanaprabu, S. K., Khanna, A., & Pandey, H. 478-484,2015.
M. (2020). Automated detection and classification of [61]. Dhanapal, R & Visalakshi, P 2016, “Optimizing
fundus diabetic retinopathy images using synergic Trust Based Secure Routing for Unified Efficient
deep learning model. Pattern Recognition Letters. Resource Sharing for Large Scale MANET-TSRRS‟,
[52]. “Distributed Security Model for Remote Healthcare Asian Journal of Information Technology, ISSN
(DSM-RH) Services in Internet of Things :1682-3915, vol. 15, no. 19, pp. 3756-3762.
Environment” Cyril Mathew, R. Dhanapal, P. [62]. Dhanapal, R & Visalakshi, P 2015, “Efficient
Visalakshi, K. G. Parthiban, S. Karthik, Journal of Clustering Protocol on Ant-Bee agent for Large Scale
Medical Imaging and Health Informatics, Volume Manet‟, International Journal of Applied Engineering
10, Number 1, January 2020, pp. 185-193(9). Research, ISSN 0973-4562, vol. 10, no. 52, pp. 349-
361.
[53]. Sankhwar, S., Gupta, D., Ramya, K. C., Rani, S. S.,
Shankar, K., &Lakshmanaprabu, S. K. (2020).
Improved grey wolf optimization-based feature
subset selection with fuzzy neural classifier for
financial crisis prediction. Soft Computing, 24(1),
101-110.
[54]. “Hybrid Dragonfly Optimization-Based Artificial
Neural Network for the Recognition of Epilepsy” R.
Dhanapal K. G. Parthiban, S. Vijayachitra,
International Journal of Computational Intelligence
Systems,Volume 12, Issue 2, 2019, Pages 1261 -
1269.
[55]. Pustokhina, I. V., Pustokhin, D. A., Gupta, D.,
Khanna, A., Shankar, K., & Nguyen, G. N. (2020).
An Effective Training Scheme for Deep Neural
Network in Edge Computing Enabled Internet of
Medical Things (IoMT) Systems. IEEE Access, 8,
107112-107123.

ISSN: 2347-8578 [Link] Page 20