Transformer-Based High-Frequency Oscillation Signal Detection on Magnetoencephalography From Epileptic Patients

Abstract

High-frequency oscillations (HFOs), observed within 80–500 Hz of magnetoencephalography (MEG) data, are putative biomarkers to localize epileptogenic zones that are critical for the success of surgical epilepsy treatment. It is crucial to accurately detect HFOs for improving the surgical outcome of patients with epilepsy. However, in clinical practices, detecting HFOs in MEG signals mainly depends on visual inspection by clinicians, which is very time-consuming, labor-intensive, subjective, and error-prone. To accurately and automatically detect HFOs, machine learning approaches have been developed and have demonstrated the promising results of automated HFO detection. More recently, the transformer-based model has attracted wide attention and achieved state-of-the-art performance on many machine learning tasks. In this paper, we are investigating the suitability of transformer-based models on the detection of HFOs. Specifically, we propose a transformer-based HFO detection framework for biomedical MEG one-dimensional signal data. For signal classification, we develop a transformer-based HFO (TransHFO) classification model. Then, we investigate the relationship between depth of deep learning models and classification performance. The experimental results show that the proposed framework outperforms the state-of-the-art HFO classifiers, increasing classification accuracy by 7%. Furthermore, we find that shallow TransHFO (< 10 layers) outperforms deep TransHFO models (≥10 layers) on most data augmented factors.

Discovering Neurophysiological Characteristics of Pathological High-Frequency Oscillations in Epilepsy with an Explainable Deep Generative Model

“…Instead of pre-selecting potential pathological HFO features, we implemented a data-driven, self-supervised strategy using VAE to define pathological HFOs. In the field of HFOs, prior studies have explored various types of neural networks, including convolutional neuronal networks, 39,40 long short-term memory, 41,42 , and transformer 43 , to classify HFOs into pathological and physiological based on human-annotated labels. This reliance limits scalability on large datasets and is constrained by the subjectivity inherent in expert labeling.…”

Magnetoencephalography-based approaches to epilepsy classification

“…Liu et al applied MEGNet, an improved CapsuleNet model, to classify MEG, and achieved the result of 95% accuracy, 94% recall, 94% F1-score and 94% accuracy ( Liu et al, 2020 ). Guo et al developed a Transformer-based model for classifying HFO (TransHFO), which combined the advantages of virtual sample generation and multi-head attention mechanism to achieve classification results with 96.15% accuracy, 100% precision, 92.86% sensitivity, and 100% specificity ( Guo et al, 2022 ).…”

Comprehensive review of Transformer‐based models in neuroscience, neurology, and psychiatry

“…Ma et al [165] improved epilepsy detection using a Transformer model called Transformers for seizure detection, which achieved an AUROC of 92.10%. Guo et al [166] introduced TransHFO for detecting high-frequency oscillations in MEG data, achieving an accuracy of 95.80% and an F1-score of 95.93%. For psychiatric disorders, Qayyum et al [19] proposed a multimodal approach to diagnose mild depression by integrating audio spectrograms with brain signals.…”

“…Nogales et al [161] -Private EEG PD 86.00% ACC BERT models Ke et al [162] MFCvT CHB-MIT [163] EEG Epilepsy 96.02% sensitivity Mutil-view gated inputs and Fourier transform 97.94% specificity Chen et al [164] -Kaggle EEG Epilepsy 82.00% sensitivity Integrating CNN and Transformer 74.60% AUROC Ma et al [165] TSD TUH EEG Epilepsy 92.10% AUROC Short-time Fourier transform Guo et al [166] TransHFO Private MEG HFO 95.80% ACC Presurgical diagnosis 95.93% F1 Qayyum et al [19] -MODMA [167] EEG Depression 97.20% precision End-to-end multimodal depression diagnosis 97.30% recall 97.30% F1 D'Costa et al [168] MS-BERT Private Text MS Testing: 88.00% Macro-F1 A pre-trained BERT specifically for MS Teferra et al [20] -Private Text Anxiety 64.00% AUROC Transformer improved anxiety prediction Huang et al [169] PABLO Private Text NAT Testing: 84.40% AUROC Patients prediction and risk factor identification Validation: 84.90% AUROC Deng et al [170] ST-Transformer ABIDE I [171] fMRI ASD 71.00% ACC Linear spatial-temporal multi-headed attention ABIDE II [171] 70.60% ACC Li et al [91] -ABIDE [148] fMRI ASD 74.18% ACC Functional connection for positional decoding Li et al [18] -Olejarczyk et al [172] EEG Paranoid SCZ Subject-independent: classify EEG data across different categories. They achieved accuracies of 91.30% and 84.26% on the BCICIV 2a [177] and BCICIV 2b [177] datasets, respectively, demonstrating the effective utilization of time and space features.…”

“…improved epilepsy detection using a Transformer model called Transformers for seizure detection, which achieved an AUROC of 92.10%. Guo et al [166] . introduced TransHFO for detecting high‐frequency oscillations in MEG data, achieving an accuracy of 95.80% and an F1‐score of 95.93%.…”