EEG-Based Stroke Severity Classification Using Higher-Order Topological Features

86% accuracy was achieved in discriminating mild from moderate stroke using a framework integrating persistent homology features with a graph convolutional network. The dataset likely contains EEG-derived functional network data and topological features for classification. It was authored by Lu Zhang and last updated on 2026-04-17.

Use Cases

• Classifying stroke severity based on EEG-derived higher-order topological features.
• Analyzing post-stroke brain network alterations, particularly in the prefrontal cortex, based on cycle ratio analysis.
• Training graph convolutional networks for medical time-series classification using combined conventional and topological EEG representations.

Strengths

• Achieved a reported classification accuracy of 86% for a specific medical task.
• Employs a method (persistent homology) that reduces sensitivity to threshold selection in network analysis.
• The dataset is shared under a permissive CC-BY-4.0 license.

Limitations

• The primary data format is a 5.5 MB PDF, which may not contain the raw underlying data in a directly machine-readable form.
• Column-level documentation and sample data are unavailable, making field semantics and data structure unclear.
• Row count and specific data dimensions are unknown, limiting suitability assessment for model training.

Provenance

Source

figshare

Collection Method

The description suggests EEG data was processed using persistent homology to extract topological features from functional networks.

Freshness

Last updated 2026-04-17 10:45:48.