Automated Detection of Epileptic Spikes and Seizures Incorporating a Novel Spatial Clustering Prior

TL;DR

The paper tackles automated epileptic spike and seizure detection by introducing a spatial clustering prior that reorganizes multi-channel MEG/EEG data into a 3D input for a custom MEEG-ResNet3D. A spatial attention module enhances the model’s ability to capture intra- and inter-cluster relationships, yielding state-of-the-art performance on the Sanbo-CMR MEG spike and TUSZ EEG seizure tasks. Key contributions include the spatial-cluster based input encoding, a 17-layer MEEG-ResNet3D architecture, and thorough ablations plus qualitative analyses that demonstrate both performance gains and interpretability. The approach offers robust, interpretable spike localization and improves automation in clinical MEG/EEG analysis, with practical implications for seizure monitoring and preprocessing pipelines.

Abstract

A Magnetoencephalography (MEG) time-series recording consists of multi-channel signals collected by superconducting sensors, with each signal's intensity reflecting magnetic field changes over time at the sensor location. Automating epileptic MEG spike detection significantly reduces manual assessment time and effort, yielding substantial clinical benefits. Existing research addresses MEG spike detection by encoding neural network inputs with signals from all channel within a time segment, followed by classification. However, these methods overlook simultaneous spiking occurred from nearby sensors. We introduce a simple yet effective paradigm that first clusters MEG channels based on their sensor's spatial position. Next, a novel convolutional input module is designed to integrate the spatial clustering and temporal changes of the signals. This module is fed into a custom MEEG-ResNet3D developed by the authors, which learns to extract relevant features and classify the input as a spike clip or not. Our method achieves an F1 score of 94.73% on a large real-world MEG dataset Sanbo-CMR collected from two centers, outperforming state-of-the-art approaches by 1.85%. Moreover, it demonstrates efficacy and stability in the Electroencephalographic (EEG) seizure detection task, yielding an improved weighted F1 score of 1.4% compared to current state-of-the-art techniques evaluated on TUSZ, whch is the largest EEG seizure dataset.

Figures (6)

• Figure 1: Automatic epileptic MEG spike detection is challenging due to the high similarity between epileptic and non-epileptic spikes, and the regional spiking characteristics.
• Figure 2: Overview of the paradigm. The spatial and clustering priors are integrated into a self-designed 3D input module, which also incorporates signal changes within a time clip. Subsequently, a custom MEEG-ResNet3D is developed to extract relevant knowledge from the input and perform spike or seizure classification.
• Figure 3: Illustration of the padded spatial arrangements for the spatial-clustering map in the input.
• Figure 4: Performance variation of our model on MEG spike detection along the number of clusters: assessment via F1 Score and AUROC.
• Figure 5: Four activation maps from the test set of Sanbo-CMR, each corresponding to a spatial-cluster map in the 3D input.