Article Preview
TopIntroduction
Parkinson’s Disease is the second most common neurological disorder after Alzheimer’s (Post et al., 2007; Pringsheim et al., 2014) Globally more than 7 to 10 million people are estimated to be suffering from Parkinson’s Disease (PD). Although the disease in itself is not fatal, but it affects the quality of life of the suffering person drastically, even leading to a comparatively shorter expectancy of life than their healthy counterparts.
The degradation in quality of life is judged by the difficulties faced by the people in carrying out day to day activities, for example holding a pen straight in hand for some time, walking without trembling, etc. Major symptoms that appear in PD are: an involuntary shaking or tremor in the body, slowed down movement of limbs called bradykinesia, difficulties in sitting and standing, loss of balance, stiffness of muscles, drooping face, speech impairment, difficulty in writing and drawing, eventual loss of control over finger movement, unstable posture, etc (Jankovic, 2008; Marusiak et al., 2010; Moore et al., 2007).
Even though a lot of research has been conducted on PD since decades, the primary cause behind the disorder in many cases is not known. However, in 1961, a strong link between the levels of neurotransmitter dopamine in the brain and PD was deciphered. Death of neurons and lack of their regeneration in the basal ganglia portion of the brain leads to a decrease in the dopamine level which is diagnosed in a vast number of PD cases. Hence the treatment of dopaminergic therapies is used to slow down the advancement of the disease. Complete recovery is never guaranteed though.
Deep Learning for PD Prediction
In the present age, when Artificial Intelligence is changing the status quo in every aspect of our lives, its subdomain of Machine Learning including Deep Learning is making strides in the healthcare sector as well. Methods like Support Vector Machines, Naive Bayes, Random Forests, Artificial Neural Networks, etc. are being excessively used in the prediction of diseases such as various cancers, diabetes, neurological disorders from medical data (Chen et al., 2019; Fathi et al., 2020; Narayan & Dwivedy, 2021). A lot of research is also being done to predict PD from various sorts of data collected from the patients, and good results achieved, generally from Deep Neural Networks like Convolutional Neural Networks. Over the years the tool that has been trusted the most for assessing PD is the Unified Parkinson’s Disease Rating Scale (UPDRS) endorsed by Movement Disorder Society (MDS) (Disease, 2003; Evers et al., 2019). The scale consists of four parts that in combination assess the development and severity of all the 65 major characteristics of the disorder, on a total score of 0 to 260 i.e. 0 to 4 on each characteristic. The upcoming literature review section is filled with many examples of classification of subjects as suffering or non-suffering from PD, by using ML or DL techniques. A substantial amount of data is available in repositories over the internet, which is generally being used to create models that can classify the subjects into suffering or normal. Our study however, takes a different approach and uses voice data of PD patients to predict the UPDRS values which can be used further for prediction of disease severity.