Multiple epileptiform waves detection algorithm based on improved VMD and multidimensional feature fusion

Abstract

Background:

Spikes, ripples, and ripples on spikes (RonS) during non-rapid eye movement (NREM) sleep are all important biomarkers associated with epileptic seizures, and accurate detection of these epileptiform waves is vital for epilepsy analysis.

New Method:

An improved variational mode decomposition (VMD) decomposes frequency bands to isolate target epileptiform waves. Multidimensional handcrafted features are extracted from low and high frequency bands to characterize these waves, with recursive feature elimination (RFE) selecting key ones. Meanwhile, a dual-stream 1-dimension convolutional neural network (1D CNN) with an adaptive scale factor extracts deep features from VMD-decomposed bands, which are then fused with the handcrafted features.

Results:

Experimental results show that the proposed algorithm achieves an average precision of 91%, a recall of 90.36%, and an F1-score of 90.62% on the scalp electroencephalogram (EEG) data of 16 children with benign childhood epilepsy with centrotemporal spikes (BECTS) from the Children’s Hospital of Zhejiang University School of Medicine (CHZU).

Comparison with existing methods:

Previous studies have often focused on only one type of epileptiform discharge. This narrow focus limits the translation of these biomarkers into clinical practice and their comprehensive application. In the present study, three types of epileptiform discharges are focused on simultaneously.

Conclusion:

Our method achieves the optimal overall detection performance in the detection of multiple epileptiform waves. It can be concluded that the proposed technique is capable of serving as an effective tool for evaluating multiple epileptiform waves.