Spurious Spike Elimination using Sparse Signal Processing Improves Seizure Onset Zone Delineation in Brief...

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Interictal epileptiform spikes and high-frequency oscillations (HFOs) are key neuro biomarkers for identifying the seizure onset zone (SOZ) in drug-resistant epilepsy.
• Accurate SOZ identification is crucial for surgical planning and improving patient outcomes.

Purpose of the Study:

• To develop and evaluate a sparse signal processing based denoising model for epileptiform spikes.
• To enhance the accuracy of SOZ delineation using denoised spike data.

Main Methods:

• A sparse signal processing model using orthogonal matching pursuit (OMP) with a Gabor dictionary was developed to represent and denoise epileptiform spikes.
• A Random Forest (RF) classifier was used for event classification.
• The model was trained on expert-labeled data and validated on intracranial EEG (iEEG) from 29 subjects during intraoperative monitoring (IOM).

Main Results:

• The denoising method significantly improved SOZ delineation from 36% to 52% in IOM.
• Comparison with HFO analysis showed an average denoised SOZ ratio of 69% in IOM.
• The model effectively removed artifacts and spurious detections from the initial spike pool.

Conclusions:

• Sparse signal processing offers a robust method for denoising epileptiform spikes, enhancing SOZ identification accuracy.
• This approach can improve clinical decision-making by providing reliable initial assessments during brief intraoperative recordings.
• The developed model shows promise for advancing epilepsy diagnostics and treatment planning.