Weak self-supervised learning for seizure forecasting: a feasibility study

Abstract

This paper proposes an artificial intelligence system that continuously improves over time at event prediction using initially unlabelled data by using self-supervised learning. Time-series data are inherently autocorrelated. By using a detection model to generate weak labels on the fly, which are concurrently used as targets to train a prediction model on a time-shifted input data stream, this autocorrelation can effectively be harnessed to reduce the burden of manual labelling. This is critical in medical patient monitoring, as it enables the development of personalized forecasting models without demanding the annotation of long sequences of physiological signal recordings. We perform a feasibility study on seizure prediction, which is identified as an ideal test case, as pre-ictal brainwaves are patient-specific, and tailoring models to individual patients is known to improve forecasting performance significantly. Our self-supervised approach is used to train individualized forecasting models for 10 patients, showing an average relative improvement in sensitivity by 14.30% and a reduction in false alarms by 19.61% in early seizure forecasting. This proof-of-concept on the feasibility of using a continuous stream of time-series neurophysiological data paves the way towards a low-power neuromorphic neuromodulation system.

Neuromorphic Deep Spiking Neural Networks for Seizure Detection

“…Accurate and objective seizure detection and recording is important for epilepsy disease management and diagnosis. If seizure detection is sufficiently accurate and real-time, it can be used to develop and test self-learning systems that are continuously learning from an incoming data and are able to adapt themselves to neural signatures of individuals and hence enable effective seizure prediction [29]. In order to deliver this vision for implantable or wearable devices to be used in continuous monitoring of epilepsy, in this study, we proposed a neuromorphic-compatible approach to modeling seizure detection with rigorous benchmarking across three epilepsy datasets, with different seizure and patient types.…”

“…However, three problems stand with the dSNN conversion approach: i) SNN conversion has not yet been optimized for sequential neural networks, and thus, temporal data is represented as an image. Forecasting models become prone to the future data leakage problem; ii) the SNN is an approximation of the DNN, and thus, the non-spiking network sets an upper-bound on performance [14], and iii) as the initial DNN is trained using error backpropagation, online learning for patient adaptation is no longer an option on resource-constrained hardware [29], [30].…”

The performance evaluation of the state-of-the-art EEG-based seizure prediction models

“…It has been for a while since researchers sought to promote the practical application of high-performance models for predicting seizures including individualized and adaptive models per patient (18,20,114,115). Along similar lines of considering the variability across patients, Yang et al (116) have recently proposed a Self-Supervised Learning ML system for predicting seizures, training a prediction model based on changes in EEG characteristics unique to each patient before a seizure resulting to a more robust performance of this adaptive model. In addition, some researchers have focused on other than EEG physiological changes in patients, such as hormonal levels, mood changes, circadian rhythms, as well as on different types of epilepsy, which could lead to more accurate seizure prediction (62).…”

Neuromorphic deep spiking neural networks for seizure detection

“…However, three problems stand with the deep SNN (dSNN) conversion approach: (a) SNN conversion has not yet been optimized for sequential neural networks, and thus, temporal data is represented as an image. Forecasting models become prone to the future data leakage problem; (b) the SNN is an approximation of the deep neural network (DNN), and thus, the non-spiking network sets an upper-bound on performance [14], and (c) as the initial DNN is trained using error backpropagation, online learning for patient adaptation is no longer an option on resource-constrained hardware [29,30].…”