Long-Term, Patient-Specific Seizure Prediction using Absolute Mean Instantaneous Frequency Difference (AMIFD) and...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy seizure prediction is difficult due to signal variability and non-stationarity.
• Existing methods struggle with diverse seizure types and long-term prediction accuracy.
• Patient-specific approaches are needed for reliable seizure forecasting.

Purpose of the Study:

• To develop a patient-specific prediction pipeline for intracranial electroencephalography (iEEG) data.
• To introduce and validate the Absolute Mean Instantaneous Frequency Difference (AMIFD) biomarker for seizure prediction.
• To improve the accuracy and reliability of long-term epilepsy seizure prediction.

Main Methods:

• Utilized a patient-specific pipeline for iEEG analysis.
• Employed the Absolute Mean Instantaneous Frequency Difference (AMIFD) biomarker.
• Applied minimum uncertainty and sample elimination (MUSE) for feature ranking and seizure clustering.
• Developed an ensemble of random forest classifiers, one for each seizure cluster.
• Implemented k-of-N aggregation logic for alarm triggering.

Main Results:

• The AMIFD biomarker demonstrated a significantly larger effect size compared to standard electrophysiological features (adjusted p < 0.04).
• The patient-specific framework achieved a mean sensitivity of 92.08% on 10 subjects.
• The system reported a low false-positive rate of 1.32 per day.
• Performance surpassed minimum Redundancy Maximum Relevance (mRMR) and MUSE baselines.

Conclusions:

• The proposed patient-specific, multi-model strategy offers a viable pathway for personalized seizure prediction.
• This approach enhances clinical reliability and accuracy in epilepsy management.
• The AMIFD biomarker and MUSE technique are effective for distinguishing seizure phases and types.