A brain-silicon closed-loop hybrid system with synapse-inspired neuroelectronic links

The SYNCH project (a synaptically connected brain-silicon neural closed-loop hybrid system) aims to create a hybrid system connecting a biological neural network (BNN) in the brain of a living animal and a spiking neural network (SNN) on a chip through memristors, mimicking synapses, enabling co-evolution of connectivity and co-processing of information at the level of neurons. The SNN acts as a supportive network, processing activity read from a neuronal population, and modulating neural behavior through intracortical microstimulation (ICMS). In the context of this project involving many partners and universities, my work and this thesis focus on the real-time bidirectional brain-computer interface and, therefore, on the detection of ongoing intracortical neuronal activity, such as extracellular action potentials (AP) and multiunit activity (MUA), and on the characterization of the feedback available to the SNN via intracortical microstimulation (ICMS). The bidirectional communication consists of developing a system that detects the in-vivo neuronal spiking activity in real time by extracting the relevant features --location and timing-- to be fed as input to the SNN. Using a traditional computer architecture to detect and send events introduces a variable latency of a few tens of milliseconds due to the USB connection with the recording system and the software elaboration time --same is for the stimulation trigger--, while to achieve a reliable closed-loop control and drive properly the neural activity, the latency between an event occurrence and its detection is a crucial point. For this reason, I developed a spike detection algorithm running directly on the FPGA of an electrophisiology recording system, together with the logic required to stimulate the tissue from the same electrode array used to sense the neuronal activity. For the characterization of the effect of the ICMS perception of the brain network, I classified the neural activity evoked by ICMS experiments conducted in the somatosensory cortex of rats during urethane anesthesia. Activity is recorded by a probe that spans the entire depth of the cortex and allows the recording of the local field potentials (LFP), as well as the MUA and the AP of subpopulations of neurons, while also allowing simultaneous stimulation. I found that the ICMS caused an initial inhibition of the activity of about 120-140 ms, followed by spindle-like activity at about 11-12 Hz. The evoked response in its entirety lasts many hundreds of milliseconds, originating from the central layers (upper end of Va and IV) and then spanning to layers III and Vb, with stronger and more widespread activity in the central oscillations. Varying the ICMS parameters --amplitude, duration, and depth --, the response did not show any significant variation in shape. The stimulation amplitude start to be effective from 5 µA with a higher number of successfully evoked trials per experiment when increased, while stimulating in layers between IV to VI did not show any significative difference. Based on these results, the response is modulated by applying pulse trains at different frequencies (5 to 20 Hz). The trains showed to strongly attract the spindles frequency to the frequency of the stimulation. By combining ICMS of different frequencies with different amplitudes (5 to 20 µA), the neuronal activity could be modeled following the pictorical representation of an Arnold’s tongue, highlighting the typical behavior of a robust self-sustained oscillator during the evoked spindle.