Prediction of Neurological Disorders using Deep Learning: A Review

Akhilesh Kumar Tripathi, Arvind Kumar Tiwari

CSED, KNIT, Sultanpur, India

Keywords: Deep learning, Machine learning, Multilayer perceptron, Auto encoders, Convolution neural networks,

Deep belief networks, Neurological disorder.

Abstract: Artificial intelligence (AI) is a field of computer science that is efficiently as well as effectively used to

analyze composite health data and extract key association in datasets. Deep learning methods are field of

machine learning method that has received important consideration in methodical society. It varies from

straightforward machine learning techniques by desirable quality to study the most favorable illustration from

untreated data. Given its capability to find abstract along with intricate patterns, deep learning has been

functional in field of neuroimaging analysis of neurological diseases featured by delicate as well as disperse

changes. This paper presents a key aspect of deep learning along with review various past work that have

been used to move toward a different machine learning algorithms to forecast the neurological disorders.

1 INTRODUCTION

Structural, metabolic, or electrical deviation of the

cortex, spinal cord, and nerves include neurological

conditions. The prevalence of chronic neurological

disorders has risen dramatically with the growing

population and aging, independent of reducing

mortality due to stroke and other communicable

neurological diseases. Diseases similar to the

marginal and central nervous systems are

neurological diseases. The disorder's popular signs

include muscle fatigue, paralysis, convulsions, pain,

short-term memory loss, and clouded mental states

(WHO, 2012).

Neurological diseases are among the most

prevalent nervous system disorders that involve

people of all genders, genders, and ethnic

backgrounds. In some instances, brain disorders can

have no detectable origin. Different syndromes for

neurological disorders are characterized based on a

particular symptom combination. We can analyze

medical data by machine learning methods, and the

issue can be diagnosed optimally. For examine the

efficiency of Machine Learning Methods,

neurological data sets are obtained from the

Neuroclinic Centre. Some of them were listed as

critical for diagnosing the problem, along with all

the attributes ordered. According to the author, the

findings show that the preferred ML techniques

provided more comprehensive outcome, and there is

only a small gap between their performances

(Ahammad, N.,2014).

Diagnosis of neurological conditions consists of

many phases since conditions such as epilepsy,

Alzheimer's, and schizophrenia are the most realistic

and unable to make the biochemical pathway

method simple. The diagnosis includes symptom

examination, medical history evaluation, and

physical testing. In neurological disorders, the EEG

test report is usually reported to forecast the usual or

pathological condition. To diagnose neurological

conditions, there are numerous psychological

measures, including behavioral neuropsychiatric

observation, audio estimation, and rational

coefficients (Helix, 2021).

Neurological disease diagnosis is a rising

problem and one of recent medicine's most daunting

challenges. According to a new study by the World

Health Organization, neurological diseases, such as

epilepsy, Alzheimer's, and headache stroke, concern

one billion people globally. Present diagnostic

technologies (MRI, EEG) provide vast amounts of

neurological disease identification and treatment

results. In general, to classify and explain the

anomalies, the study of these broad medical details

is conducted manually by specialists. According to

the author, it is hard for a person to gather, control,

examine, and understand such great quantities of

data by image inspection. Therefore, the experts

have attempted computerized diagnostic programs,

dubbed computer-aided diagnosis, to automatically

Tripathi, A. and Tiwari, A.

Prediction of Neurological Disorders using Deep Learning: A Review.

DOI: 10.5220/0010564100003161

In Proceedings of the 3rd International Conference on Advanced Computing and Software Engineering (ICACSE 2021), pages 135-139

ISBN: 978-989-758-544-9

135

diagnose neurological disorders automatically using

massive medical data(Siuly, S., 2016).

Data mining mechanism are being implemented

in biomedical sciences analysis for prediction and

deep understandings of diseases’ categorization are

done. The utilize of classification schemes in

medical analysis is increasingly growing. This recent

development in technology has required large

volumes of data to be recorded. To help the analysis

of such data for scientific choice making and

analysis, machine learning approaches have been

suggested. Promising prediction accuracy has been

obtained by several of these techniques. However,

on numerous pathologically untested data sets,

methods were assessed, making it impossible to

compare them. Other considerations have a

distinctive effect on prediction precision, such as

pre-processing, the quantity of critical attributes for

element collection, and class imbalance (Tejeswinee,

K., 2017).

2 OVERVIEW OF DEEP

LEARNING

Work on deep learning is currently on the rise. Deep

learning are based on artificial neural networks and

inspired by the brain's organization and operation.

Deep learning depends on supervised learning or

branded data for learning. Each layer of artificial

intelligence goes through the same process. It

comprises an algorithm that can be used to produce a

mathematical model as output. The methodology

utilizes covered neural networks proficient of

studying complex constructs and achieving high

degrees of pensiveness. Deep learning can be either

local or global, and it can be propagated across the

network in two different ways. In the feed-forward

process, information moves from the input layer to

the output layer in one direction. The recurrent

neural network approach helps the data from

previous input influence the current output by the

network. To facilitate information to remain

throughout the brain, neurons are linked in parallel

to one another. This enables the models to promote

the study of sequential data, such as expression and

vocabular (Vieira, S., 2017).

2.1 Data-driven Neural Network

(CNN)

A Convolution neural network performs the biased

process on an input image, which can then

distinguish one picture from another. CNN applies

the convolution operation in place of straight matrix

multiplication, and CNN is primarily used in areas

of unstructured data (e.g., image and video). 2D

CNN uses 2D kernels for segmentation prediction,

and it is a simple tool. A 2D CNN can affect aspects

of its spatial dimensions only. Since 2D CNNs only

see single inputs, they are incapable of extracting

meaning from neighboring inputs (Noor, M. B. T.,

2020).

2.2 Deep Belief Network or DBN

(DBN)

A recurrent neural network, also known as "deep"

neural network requires a diagram to store both the

going to and undirected edges. This network consists

of several layers of secret units and each layer is

connected with another. These networks comprise a

stack of restricted Boltzmann machines that connect

and previous layers. Both the nodes of the network

are not interconnected with each other proximal.

DBN can readily recognize, categorize, and describe

photos, videos, and motion data. Their

implementations are EEG, EEG. A tool to calculate

the electrical activity of the brain (Pinaya, W. H.,

2016).

2.3 Autoencoder (AE)

The unsupervised autoencoder traces out its input to

its output in an original way. To define the code, an

internal layer is used. An auto Encoder comprises of

two key parts; the encoder converts the input to the

text, and the decoder converts the code into the

origination of the original information. The three

variants of the artificial neural network are called

sparse, denoising, and contractive. The sparse

Attentional Element model includes more hidden

elements than input units. The secret units should be

triggered only a few times during the training period,

allowing the model to learn the input data's

statistical behavior. In comparison, denoising AE is

learned to recreate the original input and takes

skewed information. Contractive AE has the

potential to append explicit normalization to its

objective function that requires the model to

conform to minor inequality conditions for its

performance (Payan, A., 2015).

2.4 Multilayer Perceptron

The MLP consists of layers, where the number of

levels rises from the bottom to the top. The first

ICACSE 2021 - International Conference on Advanced Computing and Software Engineering

136

layer is where data is entered into the model; from

the input side. Neural data can be visualized as a

basic vector with each value corresponding to one

centric position. At the bottom of the process, the

output contains the likelihood of a given subject

becoming a member of one group or another. The

hidden layers and the percentages of layers are

meaningless. The number of hidden layers shows the

network's level of sophistication. In information

processing, each layer in the MLP includes a series

of interconnected artificial neurons or nodes, wholly

connected to all neurons in the previous layer. Each

relation is weighted such that the intensity and

direction of the input is expressed in the output.

Gradient descent-based algorithms can be used to

boost neural network function. Gradient descent is

an algorithm used to find the best solution for the

error (difference) between the expected result and

the actual result. Back propagation method will

predict how often the weights in the layers below

need to be modified by the algorithm. Next, neural

networks use random weights to create the training

set. This forward propagation propagates the data

during each of the layers' nonlinear mathematical

transformations. The result is compared to the

predicted outcome. There is a gradient in the

weights, and the errors are propagated into the

output. This allows the gradient descent algorithm to

change the weights as appropriate. As long as there

is an error, the process proceeds iteratively (Vieira,

S., 2017).

3 LITERATURE REVIEW

Centered on Tejeswinee, K,ShomonaGracia Jacob,

and other commented on the new data mining

methods being used in the field of neuro-

degenerative data. In the current data mining

algorithms, 93 percent of individuals were correctly

categorized using a selection technique based on

similarities. They presented a Selective Descent

Approach that offers a more optimal subset that

offers better precision in prediction (Tejeswinee, K.,

2017).Centered on Eugene Lin, Po-HsiuKuo, and

others, the research aimed to create deep learning

models that differentiate responders from non-

responders in major depressive disorder and use

these models to make predictions about treatment

outcomes. Their analysis suggests that the MFNN

model with two hidden layers has the most

significant predictive capacity for evaluating the

dynamic interaction between antidepressant reaction

and biomarkers (Lin, E., 2018).

According to Suk H-I, Shen's ADNI dataset

proposedapprox 95.8, 85.01, and 75.80 percent

effective at AD, MCI, and MCI translation,

respectively (Suk, H. I., 2013). According to B.A.,

Jonsson, G., Bjornsdottir, and all others, they have a

novel deep learning method for estimating a person's

brain age from magnetic resonance imaging (MRI)

scans of the brain. Their technique was learned and

tested on two datasets: IXI and UK Biobank. Their

technique was trained on a healthy dataset and

changed (Jónsson, B. A, 2019). According to Sandra

Vieira, Walter H. L. Pinaya, and other Deep

Learning has been extended to brain scans of

neurological diseases marked by fragile and sluggish

changes. They gave an introduction to Deep

Learning and provide a summary and overview of

Deep Learning science. Studies suggested that Deep

Learning is helping with the effort to classify

neurological diseases (Vieira, S., 2017).

Filippone et al.offeredan examination of the

various techniques of neuroimaging used to the

discrimination of three neurological conditions. The

paper illustrated the capacity for disease

identification by non-probabilistic classifiers

dependent on multiple modalities (Filippone, M,

2012). Gautam& Sharma offered a deep-learning

viewpoint that can be used to diagnose various

neurological disorders, including stroke, autism,

migraine, cerebral palsy, Alzheimer's, Parkinson's,

epilepsy, and multiple sclerosis. Their research

shows what neuroimaging software could be used to

detect different human neurological disorders.

Several papers are linked to using various deep

learning approaches for diagnosing neurological

disorders (Gautam, R., 2020). Yuhui Du and Vince D.

Calhoun published a summary paper that addressed

the different brain connectivity tests available and

the various ways those measures are categorized.

They offered a survey of the existing approaches for

FC analysis, including static and dynamic methods

and methods that have been introduced. Their

research analyzed representative applications for

mood and neurological disorders and showed

impressive classifications with precise precision (Du,

Y., 2018).

Based on Dr.SudhirG.Akojwar,

Dr.PravinR.Kshirsagar, the work integrated a

singular state signal. The comprehensive Radial

premise work method was better at work and

involves a significant number of its spread factor.

Choosing the most effective research strategy was

challenging and requires a great deal of analysis to

be done. The probability of the particles spreading

was dependent on how fast the particles travel.

Prediction of Neurological Disorders using Deep Learning: A Review

137

Combining PSO with GRNN significantly increased

the precision and efficacy of GRNN for complicated

neurological problems (Filippone, M., 2012). Hisham

and Magdy offered a novel seizure forecast method

based on deep learning and extended to durable

scalp EEGs. They offered a test method that

guaranteed a product's consistency. They obtained

the highest degree of the correct response, along

with the shortest false alarm rate and the earliest

seizure prediction time, rendering their proposed

system the most qualified among the state of the art

methods (Daoud, H., 2019).

According to Kaur and Malhi, advanced

machine learning technology has been used to

estimate the motor Unified Parkinson's Disease

Rating Score for the collected automated speech

procedures. For comparative research, they used

evaluation parameters such as similarity, R-Square,

RMSE. For assessing the results, the implications

from various ensemble models have been

recalculated. The K-fold cross-validation procedure

quantifies the robustness of the ensemble through

the statistical validation. A model that works with an

accuracy of 99.5 percent is adequate to identify

Parkinson's disease (Kaur, H., 2020).

NusratZerinZenia, MananBinthTaj Noor, and others

contrasted the latest available deep learning

techniques concerning neurological disorders. The

author addresses numerous diseases, including

Alzheimer's, Parkinson's, and schizophrenia based

on magnetic resonance imaging results obtained

using multiple modal imaging methods, including

functional and structural MRI. They investigated

how different neural network architectures operate

through a range of tasks and modalities. The

Convolutional Neural Network has outperformed

other structures for recognizing cognitive

impairments (Noor, M. B. T., Zenia, 2020).

Mohamad-ParsaHosseini, Hamid Soltanian-

Zadeh discussed a computer-based brain-computer

interface device aimed at exploring brain function.

An initial method of decreasing dimensionality is

built to improve classification correctness and

decrease training time. After a deep learning

approach and a stacked autoencoder approach are

educated, unsupervised feature extraction, and

classification results could be achieved. According

to the author, cloud computing is a solution for

processing large electroencephalograms at a real-

time scale. Their findings on a clinical dataset show

how the proposed patient-specific BCI system is

possibly the superior tool for treating epilepsy

patients and that it is intended to be useful in the

real-life cure of epilepsy patients (Hosseini, M. P.,

2016). Al-AmynValliani, Aly. Daniel Ranti

explained the various reasons deep learning has been

used in multiple fields, particularly in the healthcare

sector. The paper addresses the key obstacles that

remain in incorporating deep learning tools in the

clinical environment and sets out a plan for tackling

those(Valliani, A. A. A., 2019). Authors have

explained how a hybrid Artificial Neural Network

algorithm could be employed to identify and forecast

various neurological disorders. When doing their

analysis, the percentage of correctness, sensitivity,

and mean squared error is determined. Using this

modern method, electroencephalographic (EEG)

signals can now be identified with at least 99%

precision (

Kshirsagar, P. R., 2018).

4 CONCLUSION

Machine Learning (ML) is a field of computer

science that is efficiently as well as effectively used

to analyze composite health data and extract key

association in datasets. Deep learning methods are

field of machine learning method that has received

important consideration in methodical society. It

varies from straightforward machine learning

techniques by desirable quality to study the most

favorable illustration from untreated data. Given its

capability to find abstract along with intricate

patterns, deep learning has been functional in field

of neuroimaging analysis of neurological diseases

featured by delicate as well as disperse changes This

paper presented a key aspect of deep learning along

with reviewed various past work that have been used

to move toward a different machine learning

algorithms to forecast the neurological disorders.

REFERENCES

Ahammad, N., Fathima, T., & Joseph, P. (2014).

Detection of epileptic seizure event and onset using

EEG. BioMed research international, 2014.

“Prediction of Neurological Disorders among Children

Using Machine Learning Techniques – Helix.”

http://helix.dnares.in/2019/02/28/prediction-of-

neurological-disorders-among-children-using-

machine-learning-techniques/ (accessed Feb. 13,

2021).

Siuly, S., & Zhang, Y. (2016). Medical big data:

neurological diseases diagnosis through medical data

analysis. Data Science and Engineering, 1(2), 54-64.

Tejeswinee, K., Shomona, G. J., & Athilakshmi, R.

(2017). Feature selection techniques for prediction of

neuro-degenerative disorders: A case-study with

ICACSE 2021 - International Conference on Advanced Computing and Software Engineering

138

Alzheimer’s and Parkinson’s disease. Procedia

Computer Science, 115, 188-194.

Vieira, S., Pinaya, W. H., & Mechelli, A. (2017). Using

deep learning to investigate the neuroimaging

correlates of psychiatric and neurological disorders:

Methods and applications. Neuroscience &

Biobehavioral Reviews, 74, 58-75.

Noor, M. B. T., Zenia, N. Z., Kaiser, M. S., Al Mamun, S.,

& Mahmud, M. (2020). Application of deep learning

in detecting neurological disorders from magnetic

resonance images: a survey on the detection of

Alzheimer’s disease, Parkinson’s disease and

schizophrenia. Brain informatics, 7(1), 1-21.

Pinaya, W. H., Gadelha, A., Doyle, O. M., Noto, C.,

Zugman, A., Cordeiro, Q., ... & Sato, J. R. (2016).

Using deep belief network modelling to characterize

differences in brain morphometry in schizophrenia.

Scientific reports, 6(1), 1-9.

Payan, A., & Montana, G. (2015). Predicting Alzheimer's

disease: a neuroimaging study with 3D convolutional

neural networks. arXiv preprint arXiv:1502.02506.

Aoe, J., Fukuma, R., Yanagisawa, T., Harada, T., Tanaka,

M., Kobayashi, M., ... & Kishima, H. (2019).

Automatic diagnosis of neurological diseases using

MEG signals with a deep neural network. Scientific

reports, 9(1), 1-9.

Lin, E., Kuo, P. H., Liu, Y. L., Yu, Y. W. Y., Yang, A. C.,

& Tsai, S. J. (2018). A deep learning approach for

predicting antidepressant response in major depression

using clinical and genetic biomarkers. Frontiers in

psychiatry, 9, 290.

Suk, H. I., & Shen, D. (2013, September). Deep learning-

based feature representation for AD/MCI

classification. In International Conference on Medical

Image Computing and Computer-Assisted

Intervention (pp. 583-590). Springer, Berlin,

Heidelberg.

Daoud, H., & Bayoumi, M. A. (2019). Efficient epileptic

seizure prediction based on deep learning. IEEE

transactions on biomedical circuits and systems, 13(5),

804-813.

Du, Y., Fu, Z., & Calhoun, V. D. (2018). Classification

and prediction of brain disorders using functional

connectivity: promising but challenging. Frontiers in

neuroscience, 12, 525.

Filippone, M., Marquand, A. F., Blain, C. R., Williams, S.

C., Mourão-Miranda, J., & Girolami, M. (2012).

Probabilistic prediction of neurological disorders with

a statistical assessment of neuroimaging data

modalities. The annals of applied statistics, 6(4), 1883.

Filippone, M., Marquand, A. F., Blain, C. R., Williams, S.

C., Mourão-Miranda, J., & Girolami, M. (2012).

Probabilistic prediction of neurological disorders with

a statistical assessment of neuroimaging data

modalities. The annals of applied statistics, 6(4), 1883.

Gautam, R., & Sharma, M. (2020). Prevalence and

diagnosis of neurological disorders using different

deep learning techniques: a meta-analysis. Journal of

medical systems, 44(2), 1-24.

Hosseini, M. P., Soltanian-Zadeh, H., Elisevich, K., &

Pompili, D. (2016, December). Cloud-based deep

learning of big EEG data for epileptic seizure

prediction. In 2016 IEEE global conference on signal

and information processing (GlobalSIP) (pp. 1151-

1155). IEEE.

Jónsson, B. A., Bjornsdottir, G., Thorgeirsson, T. E.,

Ellingsen, L. M., Walters, G. B., Gudbjartsson, D. F.,

... & Ulfarsson, M. O. (2019). Brain age prediction

using deep learning uncovers associated sequence

variants. Nature communications, 10(1), 1-10.

Kaur, H., Malhi, A. K., & Pannu, H. S. (2020). Machine

learning ensemble for neurological disorders. Neural

Computing and Applications, 1-18.

Kshirsagar, P. R., Akojwar, S. G., & Bajaj, N. D. (2018).

A hybridised neural network and optimisation

algorithms for prediction and classification of

neurological disorders. International Journal of

Biomedical Engineering and Technology, 28(4), 307-

321..

Valliani, A. A. A., Ranti, D., & Oermann, E. K. (2019).

Deep learning and neurology: a systematic review.

Neurology and therapy, 8(2), 351-365.

WHO | Neurological Disorders: Public Health Challenges,

WHO, 2012.

Prediction of Neurological Disorders using Deep Learning: A Review

139