Deep Learning Recognition of Paroxysmal Kinesigenic Dyskinesia Based on EEG Functional Connectivity

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Diagnostics

Background:

• Paroxysmal kinesigenic dyskinesia (PKD) is a rare neurological disorder with diagnostic challenges, especially for secondary cases due to symptom overlap.
• Current diagnostic methods, including genetic screening, have limitations in accurately identifying all PKD cases.

Purpose of the Study:

• To develop and validate a novel deep learning-based method for recognizing PKD using resting-state electroencephalogram (EEG) functional connectivity.
• To investigate brain network properties and neural circuit differences in PKD patients compared to healthy controls.

Main Methods:

• Collected resting-state EEG data from 44 PKD patients and 44 healthy controls (HCs) using a 128-channel system.
• Computed functional connectivity matrices, transformed them into graph data, and applied graph theory analysis.
• Utilized a deep learning model (AT-1CBL) integrating 1D-CNN and Bi-LSTM with attentional mechanisms for classification.
• Performed source localization to explore neural circuit differences.

Main Results:

• The AT-1CBL model achieved 93.77% classification accuracy using Theta band phase lag index (PLI) features.
• Graph theoretic analysis revealed significant phase synchronization impairments in the Theta band of PKD patients' functional brain networks.
• Source localization indicated functional connectivity differences in sensorimotor and frontal-limbic regions in PKD patients.

Conclusions:

• Deep learning models utilizing EEG functional connectivity show promise for accurate and cost-effective PKD diagnosis.
• The findings suggest abnormalities in motor integration and brain network synchronization in PKD patients.
• Further research with larger datasets and multimodal integration is needed to enhance diagnostic model robustness.