Enhanced Machine Learning Method for Predicting Parkinson's Disease Based on Non-Motor Symptoms

Background:
Early detection of Parkinson’s disease (PD) is imperative for timely intervention. Olfactory dysfunction, a prominent non-motor symptom, alongside biomarkers derived from Cerebrospinal Fluid (CSF) analysis and dopamine transporter imaging, holds promise for early PD prediction. The burgeoning utilization of machine learning (ML) methodologies for prognosticating various pathological conditions has sparked interest in developing an enhanced ML model for PD prognosis, specifically targeting olfactory impairment symptoms.

Methods:
This study employed a systematic approach consisting of four stages: Data acquisition, Feature extraction, ML classifier development, and Results analysis. Initial data procurement involved accessing the Parkinson’s Progression Markers Initiative (PPMI) database, from which relevant non-motor features were extracted. Furthermore, features from the University of Pennsylvania Smell Identification Test, along with CSF markers such as Aβ1-42, α-synuclein, phosphorylated tau protein (P-tau181), total tau protein (Ttau), ratios of T-tau/Aβ1-42, P-tau181/Aβ1-42, and P-tau181/T tau, alongside striatal binding ratio (SBR) data, were incorporated. Subsequently, a comparative analysis of ML models was conducted based on their accuracy in predicting PD.

Results:
Automated diagnostic models leveraging ML techniques, including boosted logistic regression, classification trees, Bayes Net, and multilayer perceptron, were developed utilizing the significant features identified. The dataset was partitioned into training (80%) and testing (20%) subsets to assess model performance. Evaluation metrics such as accuracy and Area under the ROC Curve (AUC) were computed, with boosted logistic regression demonstrating the highest performance, achieving an accuracy of 98.29% and an AUC of 99.2%, surpassing existing models.

Conclusions:
Given the indirect nature of PD diagnosis and the substantial misdiagnosis rates attributed to the absence of definitive tests, the integration of ML models, particularly boosted logistic regression, presents a promising approach for enhancing diagnostic accuracy. The utility of ML algorithms in aiding clinical decision-making for PD diagnosis and emphasizes the potential for assisting healthcare professionals in more accurate disease prognosis and management.