A strategy combining intrinsic time-scale decomposition and a feedforward neural network for automatic seizure...

Area of Science:

• Neurology
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy is a global neurological disorder affecting all ages.
• Electroencephalography (EEG) is crucial for diagnosing epileptic seizures by detecting epileptiform discharges.
• Accurate and timely seizure detection is vital for effective clinical treatment.

Purpose of the Study:

• To develop an adaptive seizure detection model for EEG recordings.
• To propose a novel feature extraction method based on intrinsic time-scale decomposition (ITD).
• To enhance the accuracy and efficiency of epileptic seizure identification.

Main Methods:

• EEG signals were decomposed into proper rotation components (PRCs) using ITD.
• Statistical indices of instantaneous amplitude and frequency of PRCs were extracted.
• Feature vectors were generated by combining these indices and classified using a feedforward neural network (FNN).

Main Results:

• The proposed method effectively extracts discriminative features from EEG signals.
• Experimental results demonstrated comparable performance against state-of-the-art seizure detection techniques.
• The ITD-based feature extraction model is computationally simple and fast, involving only one parameter.

Conclusions:

• The developed system shows significant potential for real-time epileptic seizure detection.
• This method can serve as a valuable real-time diagnostic aid for physicians.
• The adaptive decomposition and feature extraction approach offers an efficient solution for EEG analysis in epilepsy.