Development of an Artificial Intelligence Framework for Accurate Diagnosis of Synucleinopathies using Clinical Features
Wajih Raza1, Juan Martinez Lemus3, Renji Hu2, Timothy Ellmore5, Charles Green4, Claudio Soto3, Chiamaka Onuigbo3, Emily Tharp3, Robert Ritter III3, Xin Fu1, Mya Schiess3
1Department of Electrical and Computer Engineering, 2Department of Information Science Technology, University of Houston, 3Department of Neurology, 4Center for Clinical Research & Evidence-Based Medicine, The University of Texas Health Science Center at Houston, 5Department of Psychology, The City College of the City University of New York
Objective:
To evaluate the utility of clinical and functional scales from large multicenter datasets in differentiating manifest synucleinopathies using AI modeling.
Background:
Synucleinopathies such as Parkinson’s disease (PD), dementia with Lewy bodies (DLB), and multiple system atrophy (MSA) have overlapping features that hinder early diagnosis. Accurate early classification is essential for timely intervention and trial readiness. Harmonized datasets like Parkinson’s Progression Markers Initiative (PPMI) and Parkinson’s Disease Biomarkers Program (PDBP) enable predictive modeling with standardized clinical scales.
Design/Methods:
Data from PPMI (PD) and PDBP (MSA and DLB) were analyzed. An XG-boost based classification framework was developed, incorporating multiple dimensions of synucleinopathies, including Schwab and England Activities of Daily Living (ADL), Movement Disorder Society–Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) Parts I–III, the Epworth Sleepiness Scale, Montreal Cognitive Assessment (MoCA), and University of Pennsylvania Smell Identification Test (UPSIT). Models were trained and cross-validated. For per-class feature analysis, a One-vs-Rest strategy was applied, computing feature importance based on gain, which reflects the improvement in accuracy provided by each feature. Scores were normalized to sum to 1, with higher values indicating greater information gain. Model performance was evaluated using sensitivity, specificity, accuracy, and area under the curve (AUC).
Results:
A total of 2,009 participants were included: controls (n=721), PD (n=688), DLB (n=374), and MSA (n=68). The XG-boost based model achieved high performance, with overall accuracy of 93.3%, macro-average AUC of 0.981, and macro-average F1 score of 0.851. Class-specific sensitivity and specificity were: control (0.935, 0.973), DLB (0.902, 0.969), MSA (0.550, 0.988), PD (0.980, 0.970). Feature importance analysis highlighted Schwab and England ADL, MDS-UPDRS Part III motor scores, and MDS-UPDRS Parts II and I as the strongest predictors.
Conclusions:
Clinical and functional measures demonstrate strong discriminative power for differentiating synucleinopathies. These findings support the use of AI approaches to enhance early diagnosis and improve patient stratification in synucleinopathy research.
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.