Deep Learning Algorithm of 12-Lead Electrocardiogram for Parkinson Disease Screening

Area of Science:

• Neurology
• Cardiology
• Artificial Intelligence

Background:

• Idiopathic Parkinson's disease (IPD) diagnosis lacks adequate early screening methods, despite increasing prevalence.
• Cardiac autonomic dysfunction is an early indicator of IPD, detectable via electrocardiogram (ECG).

Purpose of the Study:

• To develop a deep learning algorithm utilizing ECG data for efficient early screening of IPD.
• To assess the algorithm's performance in distinguishing IPD from non-IPD and drug-induced Parkinsonism (DPD).

Main Methods:

• A 16-layer deep convolutional neural network (CNN) was developed using Bayesian optimization on ECG data.
• Data included 751 IPD patients, 751 matched non-IPD controls, and 297 DPD patients.
• Model robustness was validated using 5-fold cross-validation.

Main Results:

• The CNN model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.924 for IPD detection.
• The model showed lower performance for detecting DPD (AUROC 0.473).
• Model sensitivity correlated with established Parkinson's disease rating scales.

Conclusions:

• A CNN-based deep learning model using ECG data demonstrates significant potential for identifying IPD patients.
• Standardized 12-lead ECG testing presents a clinically feasible candidate for future early IPD screening.