Analysis of brain signals
An important part of knowledge about brain function is based on recording neural electrical signals. The signal thus obtained is, however, unclear to interpret because it represents the contribution of several independent events. To interpret the neural signal, it is, therefore, necessary to analyze it with mathematical techniques aimed both at separating the contribution of the different neural phenomena and at interpreting its function in brain dynamics. The most important techniques for the separation of neural contributions are those of identifying the action potentials (impulses) emitted by neurons in the vicinity of the electrode, and the classification of impulses emitted by each neuron. The most relevant techniques for interpreting the function of a brain signal are those of decoding the signal (➔ brain, models for the large-scale activity of neuronal communication code; electrophysiology of the nervous system; neural information; action potential).
The analysis of brain signals obtained by recording neural electrical activity provides information relevant to the knowledge of brain function. Extracellular recordings of the neural signal are usually performed by inserting microelectrodes into the brain and measuring the electric field potential generated by the flow of currents into the extracellular space following various types of neural activation events. The signal thus obtained (extracellular signal) is, however, ambiguous to interpret because it represents the contribution of many potentially different or independent factors, such as the action potentials emitted by different neurons, synaptic activity, and numerous other factors.
The first step to being able to interpret the neural signal is, therefore, to analyze it with mathematical techniques aimed at separating the contribution of the different neural phenomena that generated it.
Since the output of the elementary processing carried out by each individual neuron consists of a temporal sequence of action potentials (which we will call impulses here), a very important type of analysis of the neural signal consists in identifying the impulses emitted by different neurons in the proximity of the electrode. The second step is to understand the meaning of the message that the sequence of impulses emitted by each individual neuron communicates to the rest of the nervous system.
Separation of neural signals into action potentials and local field potentials
The extracellular signal obtained by the microelectrodes is first amplified and then filtered with band-pass filters to obtain two different components of the neural signal, the high-frequency component, and the low-frequency component. The separation of the signal into high and low-frequency bands is performed to separate the action potentials (high-frequency events lasting about 1 ms) from the contributions brought about by other types of slower neural events, such as synaptic activity. The high-frequency component (obtained by filtering the extracellular signal in a frequency band typically delimited between 300 and 3,000 Hz) is the most analyzed as it contains neural impulses. The low-frequency component (obtained by filtering the signal in the frequency band 1 ÷ 200 Hz) is called local field potential (PLC) and is used to obtain an indication of the total synaptic activity in the vicinity of the electrode.
Identification of neural impulses
The first step in the analysis of signals is to identify the presence of neural impulses in the time series of the extracellular signal. To identify the pulses, the signal recorded by the extracellular electrode is first filtered with band-pass filters. Frequencies below 300 Hz are cut to eliminate slow fluctuations unrelated to pulse emissions. Frequencies above 3,000 Hz are cut to visualize pulses better and reduce noise.
The amplitude of impulses in the extracellular signal
The extracellular signal filtered in the high-frequency band has pulses of variable height. The detectable heights of the pulses depend on the distance between the neurons that emitted them and the electrode. Neurons close enough to the tip of the electrode (at a distance of no more than 50 ÷ 100 mm) are able to generate impulses that are picked up by the electrode with sufficient amplitude to guarantee a good signal-to-noise ratio to be able to be then classified and distinguished in the single-neuron activity. Neurons more distant from the tip of the electrode (at distances of about 150 mm) emit impulses that can be identified, but which can no longer be classified as belonging to different neurons due to their lower signal-to-noise ratio.
After filtering, detection is done using a threshold. When the amplitude of the neural signal exceeds a certain threshold (positive or negative), the presence of an impulse is recorded. The choice of the threshold value is a compromise between different needs. If the threshold is too high, the algorithm may not be able to detect impulses present instead. If the threshold is too low, there is the opposite problem that is the erroneous identification as impulses of events, which instead reflect a random fluctuation of the extracellular signal attributable to noise. The threshold can be chosen manually by the user after having visually inspected the data, or it can be determined by an automatic algorithm that calculates the amplitude of the noise, it compares it with that of the highest pulses and chooses the threshold value corresponding to the best compromise among the needs listed above. Automatic selection is preferable, especially when analyzing large amounts of data, e.g., those simultaneously recorded by a large number of different electrodes.
Classification of neural impulses
The second step in analyzing signals is to determine how many different neurons the pulses have been emitted from and to assign each pulse to the neuron that emitted it. In principle, the easiest way to separate the impulses emitted by different neurons is to use a discriminator based on the amplitude of the impulses. The problem with this simple technique, however, is that the impulses of different neurons can often have similar amplitude, but very different shapes. And then separate and classify the pulses into different groups (each corresponding to a different neuron) based on the joint distribution of these multiple features.
The classification of the impulses can be made by tracing outlines that delimit groups of impulses with similar characteristics (e.g., the impulse groups all with approximately the same amplitude or width). Divide them from other groups with values markedly different for these characteristics. More complex strategies can be used to classify impulses. A widespread strategy is based on memorizing the forms of impulses that pass the threshold, then identifying groups of forms of impulses with similar characteristics and calculating the typical shapes for each group of impulses of the putative neuron. Then assign the impulses to the group whose shape is more similar to theirs.
Author: Vicki Lezama