Pseudo-HFOs Elimination in iEEG Recordings Using a Robust Residual-Based Dictionary Learning Framework.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• High-frequency oscillations (HFOs) in intracranial EEG (iEEG) are vital biomarkers for identifying the seizure onset zone (SOZ) in epilepsy.
• Artifacts in iEEG data can mimic HFOs, leading to false positives and reduced diagnostic accuracy.

Purpose of the Study:

• To develop an automated method for accurately identifying and eliminating false-positive HFO events in iEEG recordings.
• To enhance the reliability of HFO analysis for clinical applications in epilepsy diagnosis.

Main Methods:

• An attention-based cascaded residual dictionary learning framework was combined with a random forest classifier.
• A second-stage refinement process evaluated event reconstruction quality using a dictionary learned from genuine HFOs to remove artifacts.

Main Results:

• The proposed method achieved 92.14% accuracy in distinguishing real HFOs from pseudo-HFOs, outperforming human expert visual assessment.
• Seizure onset zone (SOZ) localization accuracy improved by 20% in noisy iEEG data and 4% in clean iEEG data.
• The dictionary learning approach effectively captured HFO morphology and components without human supervision.

Conclusions:

• The developed algorithm effectively detects pseudo-HFOs in corrupted iEEG data, enhancing the clinical utility of HFOs as biomarkers.
• This automated method offers a robust solution for artifact-free HFO analysis, improving SOZ localization in epilepsy patients.