Computer-Aided Intracranial EEG Signal Identification Method Based on a Multi-Branch Deep Learning Fusion Model and Clinical Validation

Abstract

Surgical intervention or the control of drug-refractory epilepsy requires accurate analysis of invasive inspection intracranial EEG (iEEG) data. A multi-branch deep learning fusion model is proposed to identify epileptogenic signals from the epileptogenic area of the brain. The classical approach extracts multi-domain signal wave features to construct a time-series feature sequence and then abstracts it through the bi-directional long short-term memory attention machine (Bi-LSTM-AM) classifier. The deep learning approach uses raw time-series signals to build a one-dimensional convolutional neural network (1D-CNN) to achieve end-to-end deep feature extraction and signal detection. These two branches are integrated to obtain deep fusion features and results. Resampling is employed to split the imbalanced epileptogenic and non-epileptogenic samples into balanced subsets for clinical validation. The model is validated over two publicly available benchmark iEEG databases to verify its effectiveness on a private, large-scale, clinical stereo EEG database. The model achieves high sensitivity (97.78%), accuracy (97.60%), and specificity (97.42%) on the Bern–Barcelona database, surpassing the performance of existing state-of-the-art techniques. It is then demonstrated on a clinical dataset with an average intra-subject accuracy of 92.53% and cross-subject accuracy of 88.03%. The results suggest that the proposed method is a valuable and extremely robust approach to help researchers and clinicians develop an automated method to identify the source of iEEG signals.

Automatic Localization of Seizure Onset Zone Based on Multi-Epileptogenic Biomarkers Analysis of Single-Contact from Interictal SEEG

“…Specifically, our team applied three different pre-trained models and algorithms to detect spikes, both from traditional signal processing methods and deep learning models. Traditional signal processing methods include nonlinear feature extraction from the high-frequency bands [ 24 ], the specific features are presented in Table S1 in Supplementary Materials . For deep learning models, such as baseline model 1D-CNN [ 24 ] and the novel model SEEG-Net [ 25 ], we loaded the optimal parameters which had been pre-trained well to specific models.…”

“…Traditional signal processing methods include nonlinear feature extraction from the high-frequency bands [ 24 ], the specific features are presented in Table S1 in Supplementary Materials . For deep learning models, such as baseline model 1D-CNN [ 24 ] and the novel model SEEG-Net [ 25 ], we loaded the optimal parameters which had been pre-trained well to specific models. The model structure is presented in Supplementary Materials in Figures S1 and S2 .…”

“…Research in recent years has suggested multiple promising biomarkers for SOZ localization, such as spikes [ 17 ], high-frequency oscillations (HFOs) (ripples (Rs), fast ripples (FRs), ripples co-occurring with FRs(R&FRs)) [ 8 , 18 , 19 , 20 ], and digital features of interictal epileptiform discharges [ 21 , 22 , 23 ]. Among these, for spikes detection, we demonstrated that deep learning can detect subtle changes in SEEG [ 24 ], and a more adaptive and highly interpretable SEEG-Net was then designed [ 25 ]. In addition, we have detected HFOs accurately from the filtered band-pass part and time-frequency image part, showing strong generalization ability and consistency with the gold standard [ 26 , 27 ].…”

“…The methods mentioned above still have shortcomings, such as lacking adaptability, and low sensitivity. In other fields, attention-based methods attracted the interest of researchers due to their advantages [ 24 ]. Therefore, designing a suitable feature selection method and classifier for SOZ identification is also crucial in our research.…”

Stereo-Encephalographic Presurgical Evaluation of Temporal Lobe Epilepsy: An Evolving Science

“…A sensitivity of 97.78%, an accuracy of 97.60%, and a specificity of 97.42% were found in the Bern–Barcelona database. Deep learning may pivot new standardized studies for automated SEEG seizure detection systems ( 93 ).…”