Researchers from Belgium have developed a novel open-source software platform for processing streaming experimental data in closed-loop neuroscience experiments, while optimizing CPU resource use through a threaded architecture. The software dubbed "Falcon" has a sub-millisecond intrinsic latency, wide hardware compatibility and high flexibility in the implementation of experimental processing pipelines (J. Neural Eng. 14 045004).

Closed-loop electrophysiology experiments in which voltage measurements are used to record the activity of neural populations within the brain, which is then stimulated in direct response to the activity measured have considerable potential to explore the mysteries of brain dynamics and function.

To control these experiments, software speed and flexibility are of vital importance. Many software solutions, however, constrain themselves to specific experimental setups (such as electroencephalogram-based brain-computer interfaces, or cellular electrophysiology) or are tailored for particular hardware and data types and offer the user little control over the allocation of CPU resources.

To address these issues, Davide Ciliberti and Fabian Kloosterman of the Neuro-Electronic Research Flanders in Belgium have developed Falcon. A client-server application written in C++, Falcon's multi-threaded signal processing pipeline is built around a graph architecture, comprised of individual signal processing nodes, connected by buffered threads. Users can construct new real-time analysis pipelines by connecting basic processing nodes such as spike detectors and digital filters to suit their given experimental design.

Based on the chosen graph structure, Falcon is then able to intuitively map the necessary computations to the available CPUs so as to maximize the overall processing speed and throughput, such as by determining which computations need to be executed prior to, and which during, the actual experiment. Furthermore, on multi-core CPU systems, different processing threads can be executed in parallel across the different cores, reducing processing times.

"Falcon is highly flexible, as it allows the implementation of arbitrary real-time processing pipelines, including those requiring complex data structures like encoding models, and gives the user direct control over the CPU resources," Ciliberti told medicalphysicsweb.

Falcon would be particularly useful, the researchers say, for handling those closed-loop experiments that require complex data structures and the real-time execution of computationally intensive algorithms such as, for example, population neural decoding and encoding from large cell assemblies.

Being open source, Falcon is free-to-use. In addition, Falcon is not limited to specific hardware choices. In their study, for example, the researchers have demonstrated the capacity for Falcon to work successfully with both Neuralynx and Open Ephys hardware - demonstrating round-trip latencies of less than 1ms and less than 15ms, respectively, on both 32- and 4-core workstations, with the software only contributing 0.5ms. These round-trip latencies are at least comparable to those in other closed-loop setups.

"Falcon basically gives the experimenter a free hand over what needs to be implemented for a closed-loop experiment of interest," Ciliberti says, adding: "We will be glad to assist clinical labs that want to push real-time experiments in their research scope or want to use Falcon to test out new algorithms for decoding brain activity as part of brain-computer interfaces."

To demonstrate one such real-time experiment, the researchers describe Falcon's use to successfully detect neural population bursts from the hippocampus of a freely-moving rat with low latencies, averaging at 40ms.

With this initial study complete, the researchers are now working to expand Falcon's compatibilities with different acquisition systems and hardware. Alongside this, they plan to further demonstrate the potential of Falcon, showing how it could be used to decode neural states from the firing of large neuronal ensembles over timescales in the order of tens of milliseconds.

"For the first time, the neuroscientist community will be able to perturbed complex spike patterns at an unprecedented temporal resolution and specificity," Ciliberti says, adding: "For example, by transiently suppressing a given neuronal activity pattern for example, corresponding to a past experience we can investigate its contribution to the formation of a new memory."

Open Ephys enables closed-loop electrophysiologyWearable device monitors driver vigilanceBrain-computer interface adapts to user's moods

See the original post:
Falcon focuses on closed-loop neuroscience - Medical Physics Web (subscription)