Clustered pattern projection for EEG dementia classification: evaluating the reliability of disorder patterns

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Electroencephalography (EEG) is used for classifying Alzheimer's disease (AD) and frontotemporal dementia (FTD) against cognitively normal (CN) controls, often relying on stable, disease-related patterns.
• However, classification-relevant EEG signals can be fragmented or weakly preserved, complicating subject-level reliability assessment.
• Investigating recurrent, disorder-like EEG patterns may offer new insights into subject-wise discriminative organization in dementia classification.

Purpose of the Study:

• To investigate if recurrent disorder-like EEG patterns can provide exploratory evidence of subject-wise discriminative organization in dementia classification.
• To assess the subject-level reliability of EEG representations in distinguishing between AD, FTD, and CN subjects.
• To explore the utility of margin-based analysis for evaluating the reliability of learned EEG representations.

Main Methods:

• Analysis of a publicly available resting-state EEG dataset comprising AD, FTD, and CN subjects.
• Application of Clustered Pattern Projection (CPP) to Dynamic Mode Decomposition (DMD)-based epoch descriptors for EEG representation.
• Linear support vector machine classification within a nested leave-one-subject-out cross-validation (LOSO-CV) framework, with margin-based analysis for subject-level reliability.

Main Results:

• Clustered Pattern Projection (CPP) demonstrated competitive subject-level performance, particularly in distinguishing FTD from CN subjects.
• Task-dependent subject-level margin patterns were observed under strict subject-wise validation.
• Margin analysis emerged as a potentially useful exploratory tool for evaluating the reliability of learned EEG representations.

Conclusions:

• EEG generalization in dementia classification should consider recurrent disorder-like patterns, not solely stable biomarkers.
• These patterns can contribute to computationally detectable decision structures in dementia classification.
• CPP offers a framework for examining such patterns under rigorous subject-wise validation.