Parkinson's disease detection based on multi-pattern analysis and multi-scale convolutional neural networks.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Early diagnosis of Parkinson's disease (PD) remains a significant clinical challenge.
• Current diagnostic methods lack consensus on brain characterization, necessitating improved detection techniques.
• There is a critical need for efficient and robust methods for detecting PD, including patients on and off medication.

Purpose of the Study:

• To explore Parkinson's disease (PD) features using brain activity and functional connectivity.
• To develop and validate a novel multi-scale convolutional neural network (MCNN) model for PD detection.
• To assess the efficacy of combining brain activation and functional connectivity for improved PD diagnosis.

Main Methods:

• Multi-pattern analysis of brain functional activity in PD patients and healthy controls (HCs).
• Analysis of power spectral density (PSD) and phase-locked value (PLV) in multiple frequency bands from resting-state electroencephalography (EEG) datasets.
• Development and application of a multi-scale convolutional neural network (MCNN) for automated PD detection.

Main Results:

• Significant differences in PSD and PLV were observed between HCs and PD patients (off and on medication), particularly in β and γ bands.
• Combined analysis of PSD (brain activation) and PLV (functional connectivity) enhanced PD detection performance.
• The proposed MCNN model achieved cross-validation accuracy, sensitivity, specificity, and AUC exceeding 99%.

Conclusions:

• PSD and PLV in specific EEG frequency bands are effective biomarkers for PD detection.
• Integrating brain activation and functional connectivity patterns improves the accuracy of PD diagnosis.
• The MCNN model demonstrates significant potential for automatic and highly accurate Parkinson's disease diagnosis using spontaneous EEG activity.