DeepIED: An epileptic discharge detector for EEG-fMRI based on deep learning

Abstract

Presurgical evaluation that can precisely delineate the epileptogenic zone (EZ) is one important step for successful surgical resection treatment of refractory epilepsy patients. The noninvasive EEG-fMRI recording technique combined with general linear model (GLM) analysis is considered an important tool for estimating the EZ. However, the manual marking of interictal epileptic discharges (IEDs) needed in this analysis is challenging and time-consuming because the quality of the EEG recorded inside the scanner is greatly deteriorated compared to the usual EEG obtained outside the scanner. This is one of main impediments to the widespread use of EEG-fMRI in epilepsy. We propose a deep learning based semi-automatic IED detector that can find the candidate IEDs in the EEG recorded inside the scanner which resemble sample IEDs marked in the EEG recorded outside the scanner. The manual marking burden is greatly reduced as the expert need only edit candidate IEDs. The model is trained on data from 30 patients. Validation of IEDs detection accuracy on another 37 consecutive patients shows our method can improve the median sensitivity from 50.0% for the previously proposed template-based method to 84.2%, with false positive rate as 5 events/min. Reproducibility validation on 15 patients is applied to evaluate if our method can produce similar hemodynamic response maps compared with the manual marking ground truth results. We explore the concordance between the maximum hemodynamic response and the intracerebral EEG defined EZ and find that both methods produce similar percentage of concordance (76.9%, 10 out of 13 patients, electrode was absent in the maximum hemodynamic response in two patients). This tool will make EEG-fMRI analysis more practical for clinical usage.

Clinical benefit of presurgical EEG‐fMRI in difficult‐to‐localize focal epilepsy: A single‐institution retrospective review

“…Our recent study shows that automatic spike detection can provide comparable and, in some cases, even superior results compared to manual EEG markup in EEG‐fMRI analysis . Similarly, another recent study shows that deep‐learning–based semi‐automatic spike detector can produce EEG‐fMRI results comparable to those of traditional markup methods …”

Simultaneous EEG-fMRI: A novel approach to localize the Seizure Onset Zone

“…Moreover, previous EEG-fMRI studies reported detection of IED-related BOLD changes in 50-60% and 60-70% of the patients with spike-triggered or continuous EEG-fMRI, respectively [45]. The literature also brings into light the possibility of increasing the EEG-fMRI yield to 80-90% through using multiple HRFs peaking at 3, 5, 7, and 9 s to calculate the Page: 136 www.raftpubs.com convolutions [1,10]. It has been wellestablished that neuroelectrical activity and the corresponding hemodynamic response of the physiological and pathological brain function do not overlap precisely.…”

“…Temporal autocorrelations are corrected with an autoregressive model of order one [40], and low-frequency drifts are modeled with a thirdorder polynomial. The traditional spike-based model uses the time and duration of each event to build an IED-specific regressor and convolves it with spike related hemodynamic response function, whereas, the current study proposes to convolve the independent component time series with 4 HRF peaking at 3,5,7, and 9 seconds [1]. All components are included in the same general linear model (GLM).…”

“…Thus, the simultaneous recording and analysis of EEG and fMRI brings about a higher prospect of understanding the reality of neural behaviors. This method is now increasingly available after crucial technical challenges have been resolved in terms of developing suitable amplifiers and procedures for correcting the scanner-related artifacts in the EEG signal [1][2][3][4][5][6][7][8].…”

An Amplitudes-Perturbation Data Augmentation Method in Convolutional Neural Networks for EEG Decoding

“…The simplest data augmentation method is adding noise to the EEG signal in time domain directly. However, considering the characters of EEG such as low signal-to-noise ratio, non-stationarity and insufficient spatial resolution, this method may destroy EEG signal's feature in time domain [23]. So we propose the method that first transform the EEG recordings to the frequency domain via STFT [24], then add perturbations to amplitudes in frequency domain, finally reconstruct the time-series by inverse STFT.…”