A multimodal AI system for out-of-distribution generalization of seizure detection

Abstract

Epilepsy is one of the most common severe neurological disorders worldwide. The International League Against Epilepsy (ILAE) define epilepsy as a brain disorder that generates (1) two unprovoked seizures more than 24 hrs apart, or (2) one unprovoked seizure with at least 60% risk of recurrence over the next ten years. Complete remission has been defined as ten years seizure free with the last five years medication free. This requires a cost-effective ambulatory ultra-long term out-patient monitoring solution. The common practice of self-reporting is inaccurate. Applying artificial intelligence (AI) to scalp electroencephalogram (EEG) interpretation is becoming increasingly common, but other data modalities such as electrocardiograms (ECGs) are simpler to collect and often recorded simultaneously with EEG. Both recordings contain biomarkers in the detection of seizures. Here, we propose a state-of-the-art performing AI system that combines EEG and ECG for seizure detection, tested on clinical data with early evidence demonstrating generalization across hospitals. The model was trained and validated on the publicly available Temple University Hospital (TUH) dataset. To evaluate performance in a clinical setting, we conducted non-patient-specific inference-only tests on three out-of-distribution datasets, including EPILEPSIAE (30 patients) and the Royal Prince Alfred Hospital (RPAH) in Sydney, Australia (31 patients shortlisted by neurologists and 30 randomly selected). Across all datasets, our multimodal approach improves the area under the receiver operating characteristic curve (AUC-ROC) by an average margin of 6.71% and 14.42% for prior state-of-the-art approaches using EEG and ECG alone, respectively. Our model's state-of-the-art performance and robustness to out-of-distribution datasets can improve the accuracy and efficiency of epilepsy diagnoses.

Resting-State EEG Signal for Major Depressive Disorder Detection: A Systematic Validation on a Large and Diverse Dataset

“…Even if the EEG recording is in a resting state, this will have an effect on the participants’ mental states. These inevitable factors of EEG variability have been considered to be a challenge to real-world brain-computer interface (BCI) [ 29 , 30 ] and EEG-based applications (e.g., seizure detection [ 31 ]). In this view, most of the previous EEG studies [ 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 ] have lacked a validation on datasets with high diversity (or large EEG variability) because their data were all collected from a single site.…”

Data Augmentation Effects on Highly Imbalanced EEG Datasets for Automatic Detection of Photoparoxysmal Responses

“…Simple ML techniques such as the Artificial Neural Network or K-Nearest Neighbors algorithm were also applied in the automatic detection of ictal discharges and inter-ictal states [ 20 ], while a more complex Deep Learning (DL) method such as a channel-independent Long Short-Term Memory Network was proposed in [ 21 ]. Other studies made use of different clinical equipment, bio-markers, and biomedical measures for the same purpose, such as Electrocardiography (ECG) [ 22 , 23 , 24 ], ECG combined with EEG data [ 25 ], electromyography [ 26 , 27 ], or magnetoencephalography [ 28 ]. The existence of such recent studies on a subject that has been so widely studied over several decades clearly shows that the detection and prediction of epileptic seizures are far from being solved [ 29 ].…”

Virtual reality and machine learning in the automatic photoparoxysmal response detection

“…In this sense, [25] designed a PPR detection method by analysing the potential and oscillation of the response provoked by a flashing stimulation, but following a different stimulation pattern from the standard one. There are other recent studies that analyse the photosensitivity and epilepsy based on other generalized discharges or seizures than PPRs: in [18], a detection method based on the band amplitude fluctuation computed from a high-frequency and a low-frequency components of the EEG windows in each EEG channel is proposed; [30] applied the extreme gradient boost technique for the classification of seizures in two different ways (applying a standard partitioning of the data and applying a leave-one-out cross-validation scheme), while a channel-independent long short-term memory network is used in [5]; the information extracted from EEG and electrocardiogram (ECG) signals is used in [35] in a multi-modal neural network which analyse the data in three different ways (only EEG data with a convolutional LSTM network; only ECG data with a residual convolutional network; and a fused network which combines the outputs of the individual networks to perform the final classification); in [6], K-nearest neighbours and artificial neural networks are used for the detection of ictal discharges and inter-ictal states; [32] proposed an EEG single-channel analysis applying three types of visibility graphs (basic, horizontal and difference) to represent different EEG patterns.…”