Deep-learning-optimized microstate network analysis for early Parkinson's disease with mild cognitive impairment

Area of Science:

• Neuroscience
• Graph Theory
• Machine Learning

Background:

• Graph-theory-based topological impairment is a hallmark of mild cognitive impairment (MCI).
• Understanding topological features in early Parkinson's disease (ePD) with MCI faces challenges in identifying characteristic frequency bands and optimal windowing methods for brain network construction.

Purpose of the Study:

• To identify the characteristic frequency band for MCI in ePD.
• To develop a novel microstate window construction method for building functional brain networks.
• To explore graph-theory-based topological features and their clinical correlations in ePD-MCI.

Main Methods:

• A convolutional neural network and gradient-weighted class activation mapping were used to determine the characteristic frequency band (1-4 Hz) for ePD-MCI.
• A microstate window construction method based on electroencephalography microstate sequences was proposed to build functional brain networks.
• Graph-theory-based topological features (clustering coefficient, global efficiency, local efficiency) of the occipital lobe were analyzed.

Main Results:

• The 1-4 Hz frequency band was identified as characteristic for recognizing MCI in ePD.
• Significant decreases in clustering coefficient, global efficiency, and local efficiency were observed in the occipital lobe of ePD-MCI patients.
• These findings suggest impaired node interconnection and information transmission efficiency in the occipital lobe network.

Conclusions:

• The 1-4 Hz frequency band and specific topological impairments in the occipital lobe may serve as neural markers for ePD-MCI.
• Personalized topological impairments in brain networks could be a potential characteristic of early Parkinson's disease with MCI.