Machine Learning for the Diagnosis of Parkinson's Disease: A Review of Literature

Abstract

Diagnosis of Parkinson's disease (PD) is commonly based on medical observations and assessment of clinical signs, including the characterization of a variety of motor symptoms. However, traditional diagnostic approaches may suffer from subjectivity as they rely on the evaluation of movements that are sometimes subtle to human eyes and therefore difficult to classify, leading to possible misclassification. In the meantime, early non-motor symptoms of PD may be mild and can be caused by many other conditions. Therefore, these symptoms are often overlooked, making diagnosis of PD at an early stage challenging. To address these difficulties and to refine the diagnosis and assessment procedures of PD, machine learning methods have been implemented for the classification of PD and healthy controls or patients with similar clinical presentations (e.g., movement disorders or other Parkinsonian syndromes). To provide a comprehensive overview of data modalities and machine learning methods that have been used in the diagnosis and differential diagnosis of PD, in this study, we conducted a literature review of studies published until February 14, 2020, using the PubMed and IEEE Xplore databases. A total of 209 studies were included, extracted for relevant information and presented in this review, with an investigation of their aims, sources of data, types of data, machine learning methods and associated outcomes. These studies demonstrate a high potential for adaptation of machine learning methods and novel biomarkers in clinical decision making, leading to increasingly systematic, informed diagnosis of PD.

Keywords: Parkinson's disease; deep learning; diagnosis; differential diagnosis; machine learning.

Figure 1

PRISMA Flow Diagram of Literature Search and Selection Process showing the number of studies identified, screened, extracted, and included in the review.

Figure 2

Sample size of the included studies. (A) Cumulative relative frequency graph depicting the frequency of the sample sizes studied. (B) Histogram depicting the frequency of a sample size of 0–50, 50–100, 100–200, 200–500, 500–100, and over 1,000 for studies using locally recruited human participants and studies using previously published open databases. Green, studies using locally recruited human participants; gray, studies using data sourced from public databases. (C) Model performance as measured by accuracy in relation to sample size, shown in means (SD).

Figure 3

Data modality (A) and number of subjects (B,C) of included studies, summarized by objectives (i.e., methodology or clinical application). Orange, studies with a focus on the development of a novel technical approach to be used in the diagnosis of Parkinson's disease (i.e., methodology); blue, studies that investigate the use of published machine learning models or novel data modalities (i.e., clinical application). (A) Proportion of data modalities in included studies displayed as percentages. (B) Sample size in all included studies. (C) Sample size in studies that collected data from recruited human participants. Data shown are means (SD).

Figure 4

Data type, machine learning models applied, and accuracy. (A) Accuracy achieved in individual studies and average accuracy for each data type. Error bar: standard deviation. (B) Distribution of machine learning models applied per data type. MRI, magnetic resonance imaging; SPECT, single-photon emission computed tomography; PET, positron emission tomography; CSF, cerebrospinal fluid; SVM, support vector machine; NN, neural network; EL, ensemble learning; k-NN, nearest neighbor; regr, regression; DT, decision tree; NB, naïve Bayes; DA, discriminant analysis; other: data/models that do not belong to any of the given categories.