CHAPTER 6

Application of machine learning for

early diagnosis of Parkinson’s disease
Sudip Paul and Amitava Datta
School of Computer Science and Software Engineering, The University of Western Australia, Perth, Australia

6.1 Introduction
Parkinson’s disease can be described as a degenerative disease of the brain cells that
usually starts where the spinal cord connects to the brain and then slowly spread even
higher and later across the brain to the cortex, thus from deep within the brain to the
outside. The syndrome prevalently occurs in individuals over 50 years [1]. The exact
reason why the disease begins remains unknown but in about 5% of cases, genetic
factors play a role [1]. The brain cells in the substantia nigra in the basal ganglia deep
in the brain are damaged and a dopamine transfer agent is lacking.
Dopamine is the transmitter in five or more areas of the brain. The transmission
makes someone feel good. In the brain’s basal ganglia, which is very deep in the brain,
dopamine is essential for smooth, unconscious motor action [2]. So with movements
that have become automated, for example, if someone has cycled for a long time and
no longer think how to step and stay upright, but does it for granted [1]. As such, the
more movements are automated, the more space in the brain is open or available to
learn new tasks. If the transmitter is dopamine in the basal ganglia, the nerve messages
cannot be easily transmitted, and it becomes difficult to make auto-moving
movements [3].
People with Parkinson’s disease develop a dopamine deficiency in the brain - not
only in the basal ganglia, but also eventually in all areas of the brain where dopamine
is important as a transfer agent and later it affects the action of other transmitters [2].
Parkinson’s disease is an advanced neurological disease the symptoms include
poor movement and non-motor symptoms [1]. Speech disorders and swallowing
Parkinson’s problems with speech and swallowing will appear at an advanced stage of
the disease and are based on Parkinson’s autoimmune decomposition [2]. These disor-
ders severely impair the quality of patients’ lives.
Serious depression is a part of the disease due to changes in brain cells and transmis-
sible because the levels of serotonin and adrenaline also decrease in the brain. The
depression can be treated with antidepressants [1]. About 10 20 years before

Smart Healthcare for Disease Diagnosis and Prevention r 2020 Elsevier Inc.
DOI: https://doi.org/10.1016/B978-0-12-817913-0.00006-7 All rights reserved. 33
34 Smart Healthcare for Disease Diagnosis and Prevention

the motor or movement symptoms appear, there are subtle sensory symptoms such as
change of smell and constipation.
The conventional methods for diagnosing this disease mainly rely on MR imaging
technique. However, proper diagnosis of the disease using this method requires years
of medical education [4]. Even then, diagnosis is often a tedious and time-consuming
process. In many areas, demand for experts far exceeds available supply [5]. This in
turn increases the pressure on doctors and not infrequently delays life-saving patient
diagnostics [6]. Machine learning - especially deep learning algorithms - have recently
made great strides in the automatic diagnosis of disease, making diagnostics cheaper
and more accessible [5].
Artificial Intelligence and deep learning have proved vital and it can be used to
diagnose Parkinson. There are not enough experts to investigate the early signs of
Parkinson’s disorder [6]. Brown (2018) warns that these few experts need to do
hundreds of MR scans every day. This is not only boring and difficult work, but can
cause errors due to human fatigue [7]. Artificial Intelligence has been trained and
taught to strengthen the work of the experts to allow them to more accurately and
effectively diagnose Parkinson’s [8]. Software engineers who specialize in deep learning
have created a machine that can mimic the thought processes of our own minds [7].
While machine learning has successfully beat people in the games, what is interesting
is the possibilities for health applications such as planning treatments [6]. Very soon,
facilities all over the world will be using machines to diagnose Parkinson’s diseases.

6.2 Literature review

Parkinson disease affects in the world 0.5% of the population and on average about
50 years, but in 5% of cases even under 40 [2]. It usually begins on the right or on the
left side of the basal ganglia, with symptoms such as tremor, slowness of movement,
and rigidity of passive limb mobilization [9]. Tremor is the most well known symp-
tom, but appears only in 50% of cases [10]. Then there are different types of tremor
[11]. As such, a person’s hands shakes while performing action like brings the cup to
the mouth [11]. This is the most common form of the trajectory tremor and falls
within the vast group of essential tremor [3]. This tremor cannot be controlled and
stopped by the patient for a while. If the tremor is missing, the slowness of the move-
ment is always present, with a hindrance of writing or in the execution of fine and
repeated movements [11].
Software engineer are currently developing apps that use artificial intelligence to
detect this disease. Some facilities have installed machine-learning algorithms that have
been implement in state of art medical equipment [12]. Such systems can learn to see
patterns similar to doctors. However, algorithms need thousands of concrete examples
to learn. These examples need to be neatly digitized - machines cannot read in
 Application of machine learning for early diagnosis of Parkinson’s disease 35

textbooks between the lines [13]. Therefore, machine learning is particularly helpful
where the diagnostic information examined by the physician has already been digitized
[12]. Since there are a large number of good data in these cases, algorithms in diagnos-
tics can achieve as good results as the experts [13]. The difference is that the algorithm
can deliver results in a fraction of a second, and it can be used cost effectively any-
where in the world [14]. Soon, everyone, no matter where, could access the same
quality from top experts in psychiatry diagnostics at a low price.

6.3 Convectional diagnosis are inefficient and costly

The diagnosis of Parkinson’s disease is in most cases done using clinical examination.
Neurologists who are specialized in general movement disorder can diagnose this dis-
ease with the history and examination findings in the clinic [15]. However, in order
to exclude some secondary conditions that resemble Parkinson’s disease but do not
have Parkinson’s disease and mimic their symptoms, MR imaging is performed [16].
There is no specific laboratory method for Parkinson’s disease [17]. There is no specific
imaging method [17]. Recently, however, some distinctive methods have been devel-
oped from other similar conditions of idiopathic Parkinson’s disease.
This kind of imaging method applied is in nuclear medicine departments [15].
However, this technique should not be performed for all patients with Parkinson’s
disease. Proper diagnosis is made with neurological examination information [11]. The
response to the drug is considered and in such cases, no daTscan is required.
However, some patients with tremor may be suspicious of essential tremor or
Parkinson’s disease [18]. In some cases, when doctors have a suspicion of vascular
congestion or a side effect of psychiatric medications, it can be mistaken for a
Parkinson’s disease [19]. In such cases, the condition might not respond well to the
treatment that is being applied [11].

6.4 Effectiveness of machine learning technique for Parkinson

detection
Neurological conditions such as Parkinson’s are as of now analyzed in a facility where
an authority can break down the dialect of the patient, however extra tests are per-
formed to check psychological capacity and general engine control [1]. The patient
may encounter the primary side effects of these conditions for a long time before
being analyzed, so any information source that can furnish the specialists with conve-
nient data could empower early determination and proper treatment [1]. The need for
practical tools that facilitate the diagnosis of Parkinson’s stems from the fact that many
of symptoms can be associated with other diseases that can confuse the symptoms [20].
As such, it would seem that the smartphone could be used as a new tool to improve
36 Smart Healthcare for Disease Diagnosis and Prevention

clinical care and develop new drugs. Smartphone data can be used to diagnose
Parkinson’s disease [21]. Scientists have proposed the use smartphone data and
machine learning to effectively diagnose Parkinson’s disease [1]. The idea is to use data
collected from a smartphone and analyse them for signs of the disease, as well as its
progression.
Hospitals can used a smartphone app to collect data. Participants will be requested
to enter demographic and medical information. After receiving the information,
several tests can be performed using the measurement transmitted in real time by the
patient’s phones [15]. The tests, which can been conducted several times in a day,
include walking, voice, touch and memory [21]. The walk test provides instructions,
such as walking predetermined steps forward and going around. The voice test asks
the user to utter given word repeatedly [15]. The interception test requires the user to
place the phone on a flat surface and alternatively touch two buttons on the screen for
20 s [2]. The memory test requires the user to remember the pattern of color that
show in a screen, and then touch colors in the same order. In comparison, multiple
walking tests were the most effective way to diagnose Parkinson [15]. Doctors could
also access long-term observational data without requiring the presence of patients,
according to the report.
Different tests can be utilized to test distinctive parts of reasoning capacity. It tends
to be hard to quantify some intellectual changes and connect them with turmoil.
Nonetheless, the discourse examination can give extra insights [20]. Deepening engi-
neers can writing software or a signal-processing algorithms that can be used to discov-
ered voice changes in Parkinson [15]. As such, the manifestations of the infection can
be identified by investigating discourse signals utilizing PC calculations [15]. This is
machine learning, programs gather a lot of information when somebody has the infec-
tion or not.
Programming engineers have train the database to figure out how we can isolate
the genuine manifestations of the illness from different variables [15]. There are differ-
ent reasons that reason the voice to change, regardless of whether one is smoking or
simply having a chilly. In any case, the calculations will have the capacity to differenti-
ate between these causes and Parkinson’s infection [15]. It is more detailed than
attempting to identify a specific tremor in his voice. The calculations likewise consider
different estimates that put the tremor in the correct setting, regardless of whether the
patient has chilly or different indications [15]. From the discourse design, the calcula-
tions ascertain a basic dysphonic proportion of Parkinson’s side effect seriousness on a
standard clinical scale utilized by doctors.
The Fig. 6.1 below compares the effectiveness of machine learning in detecting
Parkinson disease. Image ‘A’ represents the level of tremor when simulation process
and medication has not been performed. Image ‘B’ is the state of this condition when
the machine has been used to detect it but with medication. Image ‘C’ has no
 Application of machine learning for early diagnosis of Parkinson’s disease 37

Figure 6.1 Compares the effectiveness of machine learning in detecting Parkinson disease [22].

simulation but 150% medication has been prescribed. In image ‘deep study simulation’
has been applied followed by 150% medication (Effect of Deep Brain Stimulation on
Parkinsonian Tremor, 2018).
The CogniFit Cognitive Assessment Battery for Parkinson’s (CAB-PK) is a main
expert instrument that incorporates a battery of clinical preliminaries and approved
errands to rapidly recognize and assess the signs, qualities, and dysfunctions in the sub-
jective procedures influenced by the malady of Parkinson’s illness [23]. This creative
online Parkinson’s test is a logical resource that enables you to play out an entire
subjective investigation, comprehend the client’s intellectual qualities and shortcomings
evaluate Parkinson’s risk record, and better comprehend the regions influenced by the
sickness [11]. This test was intended for youthful grown-ups, more seasoned individuals,
and elderly individuals who have Parkinson’s risk factors [15]. Any expert or individual
client can without much of a stretch utilize this neuropsychological battery [24].

6.5 Discussion
Artificial Intelligence (AI) has also brought new diagnostic tools to medicine and the
most renowned of them is known as deep learning. The deep learning has many
applications in different industry segments. In medicine, we highlight some examples
in these references. Parkinson’s malady has no biomarkers and must be determined
through expensive clinical examination to have a nervous system specialist [15].
38 Smart Healthcare for Disease Diagnosis and Prevention

Machines can examining 30-s voice accounts with deep learning programming,
accomplishing 98.6% precision in identifying regardless of whether a member endures
of the malady [15]. Outside the health area, it is also worth noting another niche being
explored in the voice segment is the nature of the conversations between people [15].
Who is capturing this research is nothing less than the famous MIT that is seeking to
identify through AI technology if the tone of a given conversation is happy, sad or
neutral [15]. The challenges in diagnosing diseases through voice biomarkers remain.
There is some breakthroughs to go in for this diagnosis to become truly viable and
widespread.
CAB-PK utilizes protected calculations and man-made reasoning (AI) that permits
investigating a large number of factors and telling the client of any risk of Parkinson’s
with extremely palatable psychometric outcomes [11]. The subjective profile in the
neuropsychological report gives high unwavering quality, consistency and soundness
[11]. The test can approved utilizing rehashed tests and evaluation procedures.
Hospitals can use the projected cross-sectional survey, such as the Cronbach Alpha
coefficient, reaching values of about 0.8, Test-Retest test scores can go close to 1,
which indicates a high reliability and precision [25].

6.6 Importance of machine learning for early detection

Only when a doctor is sure of his diagnosis, one can treat patients adequately.
Conventional diagnostic methods are very expensive and require complicated labora-
tory equipment as well as expert knowledge. They make the process of diagnosing a
disease safe and cost effective [26]. It can also be used to determine the progress of a
disease - making it easier for physicians to choose the right treatment and monitor the
effect of the drug [26]. However, it is difficult to find suitable biomarkers for a partic-
ular disease. Another expensive, time-consuming process that involves testing tens of
thousands of potential molecular candidates.
Machine learning can automate much of the manual work and speed up the
process overall [27]. The algorithms classify molecules into suitable and
unsuitable candidates - which in turn helps researchers to focus on the analysis of the
best candidates [28]. In addition, machine learning assists in MR images analysis
process [19]. One of the major research areas of the leading technology companies in
the area of AI lies in automated image recognition [29]. There is also great potential
in the medical application of image recognition methods [28]. Machine learning ana-
lyzes MR images in search of Parkinson’s symptoms and classifies them according to
both good and bad. Machine learning -based systems can mitigate the threat of supply
shortages due to the shortage of skilled workers [28]. Already algorithms are approved
which realize Parkinson’s signs in a few seconds by analyzing MR images. Otherwise,
experts need an average of 30 minutes for this process.
 Application of machine learning for early diagnosis of Parkinson’s disease 39

6.7 Recommendation
Although AI technology is at progress worldwide it is still unclear how algorithms
developed for some patients will work for other people in different. As such, the
conclusion of Parkinson’s illness requires a multidisciplinary assessment and a thorough
differential finding to preclude a maladaptive or inefficient symptomatology, which
might be better clarified by an inclination issue, an alternate degenerative ailment or
different pathologies. Clinical history, physical and neurological examination, research
facility results, scales, neuroimaging tests, and neurophysiological examinations are the
best instruments for the discovery of Parkinson’s infection, yet it is not adequate to
comprehend the dimension of weakening caused by the disease. A clinical and neuro-
psychological assessment is needed to understand the severity of the disease. Therefore,
it is recommended using this complete Parkinson’s test as an additional tool for a
professional diagnosis. This evaluation cannot replace a clinical appointment.

6.8 Conclusion
Parkinson’s is a disease accompanied by tremors, stiffness of muscles, slow movements,
and stability and balance difficulties. As such, the analysis of the disease often based on
these manifestations. The issue is that these indications will in general happen when
the disease is as of now in a propelled procedure and accordingly analysis is likewise
made when about 60% of the nerve cells in the brain region that controls the motor
exercises are as of now crushed and thusly the rehabilitation and therapeutic capacities
are low. Using machine learning to detect Parkinson’s not only allows us to better
diagnose a series of very serious pathologies for those who suffer them, but also offers
a great advance in automating processes that can be tedious and inefficient when done
by humans. Once the initial purpose of demonstrating the effectiveness of its method
has been achieved, the scientific team can begin to develop a simple system that
allows both neurologists and non-specialists to use this technology in hospitals and
health centers.
The after effect of this work is an amazing classification engineering that joins
directed and unsupervised figuring out how to consequently remove the most perti-
nent qualities of an arrangement of pictures. The proposed technique has been assessed
utilizing a colossal database from the Parkinson’s disease Neuroimaging Initiative
(ADNI). The consequences of this work, which has likewise included patients with
other intellectual shortages that can build up Parkinson’s inside two years, demonstrate
the capability of AI procedures to uncover the examples associated with the disease.
The precision rates acquired for the conclusion permit an incredible advance forward
in the learning of the neurodegenerative procedure engaged with the improvement of
the disease, as well as filling in as a beginning stage for the advancement of more
40 Smart Healthcare for Disease Diagnosis and Prevention

dynamic therapeutic medicines. Then again, the strategies created can serve or be a
beginning stage for enhancing exactness in the finding of different dementias such as
Parkinson’s disease. Moreover, the strategies created are being utilized to enhance the
finding and research of the beginning of different difficulties such as dyslexia.

Acknowledgment
We would like to acknowledge the Ministry of Science & Technology, Department of
Biotechnology, Government of India for providing the financial support as Biotechnology
Overseas Associateship for North- Eastern region. Also my sincerely gratitude to my colleagues and
staffs of the School of Computer Science and Software Engineering, The University of Western Australia,
Perth for essential software and technical and also provide us to explore more in this domain.

References
[1] E. Abdulhay, N. Arunkumar, K. Narasimhan, E. Vellaiappan, V. Venkatraman, Gait and tremor
investigation using machine learning techniques for the diagnosis of Parkinson disease, Future
Gener. Comput. Syst. 83 (2018) 366 373.
[2] D. Avci, A. Dogantekin, An expert diagnosis system for Parkinson disease based on genetic
algorithm-wavelet kernel-extreme learning machine, Parkinson’s Dis. 2016 (2016).
[3] T.V. Sriram, M.V. Rao, G.S. Narayana, D. Kaladhar, Diagnosis of Parkinson disease using machine
learning and data mining systems from voice dataset, Proceedings of the 3rd International
Conference on Frontiers of Intelligent Computing: Theory and Applications (FICTA) 2014,
Springer, 2015, pp. 151 157.
[4] A. Burik, AI is Being Put to Work to Treat Parkinson’s Disease in the UK. (2018). Available:
,https://labiotech.eu/medical/benevolent-ai-parkinsons-disease/..
[5] Z. Cai, J. Gu, C. Wen, D. Zhao, C. Huang, H. Huang, et al., An intelligent Parkinson’s disease
diagnostic system based on a chaotic bacterial foraging optimization enhanced fuzzy KNN approach,
Comput. Math. Methods Med. 2018 (2018).
[6] F. Faghri, S.H. Hashemi, H. Leonard, S.W. Scholz, R.H. Campbell, M.A. Nalls, et al., Predicting
onset, progression, and clinical subtypes of Parkinson disease using machine learning, Biorxiv (2018)
338913.
[7] K.N.R. Challa, V.S. Pagolu, G. Panda, B. Majhi, An improved approach for prediction of
Parkinson’s disease using machine learning techniques, Signal Processing, Communication, Power
and Embedded System (SCOPES), 2016 International Conference on, IEEE, Paralakhemundi,
India, 2016, pp. 1446 1451.
[8] C. Gao, H. Sun, T. Wang, M. Tang, N.I. Bohnen, M.L. Müller, et al., Model-based and model-
free machine learning techniques for diagnostic prediction and classification of clinical outcomes in
Parkinson’s disease, Sci. Rep. 8 (2018) 7129.
[9] R. Cellan-Jones, Parkinson’s Diagnosis Set to be Sped Up by Tencent’s AI. (2018). Available:
,https://www.bbc.com/news/technology-45760649..
[10] S. Emrani, A. McGuirk, W. Xiao, Prognosis and diagnosis of Parkinson’s disease using multi-task
learning, in: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge
Discovery and Data Mining 2017 1457 1466.
[11] J. Schaff, Using Machine Learning to Predict Parkinson’s Disease. (2018). Available: ,https://
www.elderresearch.com/blog/machine-learning-for-predicting-parkinsons..
[12] A. Keown, Can AI Detect Parkinson’s Disease from a Video? Tencent and Medopad Team up to
Find Out. (2018). Available: ,https://www.biospace.com/article/can-ai-detect-parkinson-s-dis-
ease-from-a-video-tencent-and-medopad-team-up-to-find-out/..
 Application of machine learning for early diagnosis of Parkinson’s disease 41

[13] A. Gaskell, Using AI to Provide Early Diagnoses of Parkinson’s. (2017). Available: ,https://dzone.
com/articles/using-ai-to-provide-early-diagnoses-of-parkinsons..
[14] C. Russey, Study Explores Use of Wearable Technology and Machine Learning to Monitor
Parkinson’s Disease. (2018). Available: ,https://www.wearable-technologies.com/2018/11/study-
explores-use-of-wearable-technology-and-machine-learning-to-monitor-parkinsons-disease/..
[15] M. Nilashi, O. Ibrahim, H. Ahmadi, L. Shahmoradi, M. Farahmand, A hybrid intelligent system for
the prediction of Parkinson’s disease progression using machine learning techniques, Biocybernet.
Biomed. Eng. 38 (2018) 1 15.
[16] S. Joshi, D. Shenoy, P. Rrashmi, K. Venugopal, L. Patnaik, Classification of Alzheimer’s disease and
Parkinson’s disease by using machine learning and neural network methods, Machine Learning and
Computing (ICMLC), 2010 Second International Conference on, IEEE, Bangalore, India, 2010,
pp. 218 222.
[17] K.-C. Kwak, A design of fuzzy classifier using fuzzy clustering and voice analysis for diagnosis of
Parkinson disease, J. KIISE, 11 (2013) 71 76.
[18] N. Quinn, Accuracy of clinical diagnosis in early Parkinson disease, Arch. Neurol. 58 (2001)
316 317.
[19] J.M. Lopes, Certain People at Risk of Parkinson’s Detected by Machine Learning. (2018). Available:
,https://parkinsonsnewstoday.com/2018/07/25/people-reduced-smell-taste-at-risk-parkinsons-
detected-machine-learning/..
[20] Early Detection of Neurological Disorders Using Machine Learning systems: IGI Global, 2018.
[21] C.R. Pereira, S.A.T. Weber, C. Hook, G.H. Rosa, J.P. Papa, Deep learning-aided parkinson’s dis-
ease diagnosis from handwritten dynamics, 2016 29Th SIBGRAPI Conference on Graphics,
Patterns and Images (SIBGRAPI), IEEE, Sao Paulo, 2016, pp. 340 346.
[22] Effect of Deep Brain Stimulation on Parkinsonian Tremor, https://physionet.org/physiobank/data-
base/tremordb/g2.gif, Ed., ed: physionet.org, 2016.
[23] A.S. Chen-Plotkin, Parkinson disease: blood transcriptomics for Parkinson disease, Nat. Rev.
Neurol. 14 (2018) 5.
[24] R. Prashanth, S.D. Roy, P.K. Mandal, S. Ghosh, High-accuracy detection of early Parkinson’s
disease through multimodal features and machine learning, Int. J. Med. Inf. 90 (2016) 13 21.
[25] J.M.F. Vicente, J.R.Á. Sánchez, F. de la Paz López, F.J.T. Moreo, Natural and Artificial
Computation in Engineering and Medical Applications, Springer, 2013.
[26] Global Trends in Computing and Communication Systems, Berlin, Heidelberg, 2012.
[27] D. Brown, Companies hope to use AI for early detection of Parkinson’s. (2018). Available:
,https://www.aiin.healthcare/topics/artificial-intelligence/companies-use-ai-early-detection-
parkinsons..
[28] N.M. Tahir, H.H. Manap, Parkinson disease gait classification based on machine learning approach,
J. Appl. Sci. 12 (2012) 180 185.
[29] A. Bough, Could artificial intelligence be used to detect Parkinson’s disease? (2018). Available:
,https://www.neuro-central.com/2018/10/08/could-artificial-intelligence-be-used-to-detect-
parkinsons-disease/..