An electronic neuromorphic system for real-time detection of High Frequency Oscillations (HFOs) in intracranial EEG

Abstract

The analysis of biomedical signals for clinical studies and therapeutic applications can benefit from compact and portable devices that can process these signals locally, in real-time, without the need for off-line processing. An example is the recording of intracranial EEG(iEEG) during epilepsy surgery with the detection of High Frequency Oscillations (HFOs, 80-500 Hz), which are a biomarker for the epileptogenic zone. Conventional approaches of HFO detection involve the offline analysis of prerecorded data, often on bulky computers. However, clinical applications during surgery or in long-term intracranial recordings demand a self-sufficient embedded device that is battery-powered to avoid interfering with other electronic equipment in the operation room. Mixed-signal and analog-digital neuromorphic circuits offer the possibility of building compact, embedded, and low-power neural network processing systems that can analyze data on-line and produce results with short latency in real-time. These characteristics are well suited for clinical applications that involve the processing of biomedical signals at (or very close to) the sensor level. In this work, we present a neuromorphic system that combines for the first time a neural recording headstage with a signal-to-spike conversion circuit and a multi-core spiking neural network (SNN) architecture on the same die for recording, processing, and detecting clinically relevant HFOs in iEEG from epilepsy patients. The device was fabricated using a standard 0.18μm CMOS technology node and has a total area of 99 mm2. We demonstrate its application to HFO detection in the iEEG recorded from 9 patients with temporal lobe epilepsy who subsequently underwent epilepsy surgery. The total average power consumption of the chip during the detection task was 614.3 μW. We show how the neuromorphic system can reliably detect HFOs: the system predicts postsurgical seizure outcome with state-of-the-art accuracy, specificity, and sensitivity (78%, 100%, and 33% respectively). This is the first feasibility study towards identifying relevant features in intracranial human data in real-time, on-chip, using event-based processors and spiking neural networks. By providing “neuromorphic intelligence” to neural recording circuits the approach proposed will pave the way for the development of systems that can detect HFO areas directly in the operation room and improve the seizure outcome of epilepsy surgery.

A spiking neural network (SNN) for detecting high frequency oscillations (HFOs) in the intraoperative ECoG

“…Here, we simulated a spiking neural network (SNN) for HFO detection in the intraoperative ECoG. This work builds on a previously validated SNN for HFO detection in the intracranial EEG (iEEG) 23,24 , and extends it by introducing a novel artifact rejection mechanism to reject fast transient artifacts, and by validating it on intraoperative electrocorticography recordings (ECoG). As a computational principle, the SNN emulates the spiking of neurons in small networks 25 so that they can be implemented in low-power and compact neuromorphic hardware that perform real-time computation 26 .…”

“…In this benchmark testing, we were able to correctly predict the postoperative seizure outcome in all 8 patients. This is a further step towards an SNN that may be implemented in a neuromorphic device 23 for standardized and real-time HFO detection during epilepsy surgery.…”

“…1a). Since Butterworth filters showed to be a good approximation of the Tow-Thomas architectures for the hardware SNN filters 23,24,28 , we used a 2nd order filter of this model and the SciPy Python package to simulate the filtering stage [29][30][31] . In the filtered signal, we defined a baseline amplitude that has to be exceeded by a putative HFO event (Fig.…”

“…1b). Following the algorithm implemented in the hardware SNN 23 , we selected a 1 s time window, stored the maximum signal amplitudes of consecutive non-overlapping time windows of 50 ms, and took the mean of the lowest quartile as the baseline amplitude.…”

“…The projections to the second layer neurons are excitatory for UP spikes and inhibitory for DN spikes. We used the Python SNN simulator Brian2 34 , the custom toolbox Teili 35 , and the parameters in Table 1 to simulate an SNN that matches the behavior of the neuromorphic circuits of the hardware SNN 23 . The software simulations take into account the neuromorphic circuit properties.…”