Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Simple Thresholding and NEO Hybrid Brain Spike

Detection Methods
A Literature Review

Harvish Mehta
Pennsylvania State Department of Electrical Engineering
University Park, PA, USA

Abstract— Brain waves are produced by synchronized pulses


from the neurons that communicated with each other. These II. SIMPLE THRESHOLDING
neuron connections dictate our perspectives, thoughts, emotions, In order to achieve a hardware realization, the application
and behaviors. Therefore spike detection from brains can specific integrated circuit (ASIC) must be real-time,
influence the direction that brain interfaced technology is heading
autonomous, and operational with appropriate signal to noise
as well as open a window into the brain to further cognitive
research. This neural processing is processed through spike ratio (SNR). Furthermore, all spike detection methods involve
detection, which determines and sorts the many pulses coming two stages of pre-emphasizing and thresholding. The pre-
from the brain. Spike detection is the first step to gaining emphasizing is exemplified through the detection of the spike
meaningful information from the brain computer interface. Thus, using matched filtering, absolute thresholding, and nonlinear
this paper discusses common algorithms and methods that energy operators, while thresholding is seen as a comparator to
potentially could be used to detect and sort spikes in the brain. sort the signal as valid [5].
Thus, to achieve a hardware realized ASIC, a simple
Index Terms—brain spike detection, sorting spikes, NEO. threshold method with minimal or no preprocessing is ideal. The
advantages of using simple thresholding include lower power
I. INTRODUCTION operation, a low SNR tolerance, and a fast real time response.
However, some of the disadvantages include the need for user
Brain signals are split into many categories based on the
input for the predetermined threshold amplitude. There are three
frequency depending on brain activity level. Due to the infinite
classes of simple threshold methods with minimal computation.
electrical pulse between our billions of neurons, noise becomes
The first is null, where no preprocessing is applied and positive
an interfering and vital factor in classifying the brain signals
threshold is applied to the received raw data. The second is
[4]. Therefore, spike detection is used to determine the
negation where the signal is inverted before a negative threshold
frequency and derive an accurate signal from the brain waves.
is applied. The third is absolute value (ABS) where the signal is
Spike detection is the first step in any analyzation used create a
preprocessed through absolution by applying both a positive and
brain machine interface (BMI). All spike detection methods
negative threshold [2, 5].
revolve around detecting the signal and then preprocessing the
signal for accuracy. This task can be done through an amplitude A. Simple Threshold Methods Comparisons
threshold or even a software based program and firmware. The In [2], simple thresholding methods compared to various
bounds of detecting brain spikes correlate to power nonlinear operator (NEO) and matched filtering (MF) spike
consumption, probability of true detection, and real-time detection techniques. All techniques were tested by using
processing accuracy. This paper will survey the common sample data and action potential waveforms to determine the
algorithms used to analyze the analog feedback received from probability of spike detection (PD) and number of false alarms
the brain. (PFA). To determine the effectiveness and the best spike
The remainder of the paper will break into four sections. detection method, a cost function dependent on PD, PFA, and
Section II will discuss simple threshold, and will derive a preprocessor clock cycles was derived. Because each multiply
reliable and effective simple thresholding technique. Section III and accumulate (MAC) operation creates more clock cycles,
will apply the same the simple thresholding technique to a new computation heavy preprocessors such as the MF receive lower
wireless headstage that combines optogenetics and cost function scores. Therefore, the cost function in [2] is biased
multichannel electrophysiological recording. Section IV will towards heavy preprocessors.
introduce a nonlinear energy operator (NEO) hybrid, which The cost function created the best score for the ABS simple
combines advantages of NEO and thresholding detection thresholding method. To continue, the ABS simple thresholding
methods. Section V will discuss the implications of sections II- method also effectively operated within multiple SNR ranges. In
IV. further comparison of MF, NEO, and simple threshold
technique, a comparison without MAC operations was
considered. In this case, NEO and MF retain higher cost function The hybrid method is proposed in [5] to combine both
scores when the refractory periods (brief moment of time when methods. This two-phase system utilizes ABS to detect a
the neuron slightly pauses) are not present. Thus, when potential spike, and NEO to act as a checking mechanism to
refractory periods are considered the ABS was mostly effective validate the detected spike. Because equation (1) describes ABS
because of its high real time response. simple thresholding technique, equation (2) is a variation of
NEO that can be implemented as a hybrid.
III. APPLICATION OF ABS SIMPLE THRESHOLDING
The next stages of brain spike detection utilize optogenetics. 𝜓[𝑥(𝑛)] = [𝑥(𝑛)]2 − [𝑥(𝑛 − 1) ∗ 𝑥(𝑛 + 1)] (2)
Optogenetics is a new experimental approach that can activate
or deactivate light-sensitized neurons with precision. [1] Equation (2) also has an input data array of 𝑥(𝑛), while
Describes the first attempt at high-resolution neural recording 𝜓[𝑥(𝑛)] is similar to that of 𝑦(𝑛), where 𝜓[𝑥(𝑛)] needs to be
with optogenetics. In [1], a new wireless headstage for combined large to account for a true positive detection. In further
optogenetic and multichannel electrophysiological recoding is comparison of the two equations, the ABS simple thresholding
created to process 32 neural data recording channels, in parallel method detects a spike at any abrupt change or high frequency,
to controlling 32 implantable fiber couple simulation LEDs. Due thereby increase the number of false alarms, while the NEO
to a high resolution neural recoding with optogenetics, results variation extracts and eliminates error due to abrupt changes and
show that simultaneous stimulation and detection lead to results high frequencies. NEO becomes more effective at a cost of more
that are more reliable. computation.
In order to account for the high-resolution accompanying 32
recording channels and 32 stimulation channels, a low power A. Hybrid Spike Detection Technique
digital signal processor is needed. To achieve, a low power Figure 1 portrays the hybrid algorithm presented in [5]. The
digital signal processor, not only are compressors needed but ABS simple thresholding method is used to detect spikes from
effective brain spike detection is also needed. The firmware has the raw data. If a potential spike is detected, then the NEO
core design of a spike detector based off an absolute value algorithm is enabled. This allows less power consumption as the
operator. This way there are lower complexities and high NEO method is reduced from being ON during all clock cycles
performances as seen in the previous section with ABS simple to being enabled during a few clock cycles. As stated earlier the
thresholding. NEO validates the spike as true or false by outputting through a
𝑦(𝑛) = |𝑥(𝑛)| (1) secondary threshold. These thresholds are acquired by using the
variances and medians of the inputted data.
Equation (1) presents the magnitude of an input array, 𝑥(𝑛),
and when 𝑦(𝑛), output array, is greater than the threshold then
only is the spike detected. Although this spike detector is simple,
it is made more effective using an adaptive threshold determined
with the standard deviation of noise. Because electrodes drifting
in the brain and the presence of neural noise, changes in SNR
are originated. Thus, an automatic threshold calculation strategy
is adopted by the use of statistics, in which a PDF of half a
normal distribution models the detected spikes. The
performance of this spike detection method gives a 100% true Fig. 1. Hybrid Algorithm Diagram [5]
positive detection rate at as low as 15dB SNR, and a slightly
lower true positive detection rate, 97.28%, as low as 5dB SNR B. Results of Hybrid Spike Detection Technique
[1]. ABS simple threshold with a combination of an adaptive
threshold serves as a low power, precise and efficient spike Using 65nm CMOS technology the power
detector. consumption of the hybrid improves upon the power by 54.48%
when compared to the NEO. All the while, the hybrid circuit
IV. NEO HYBRID DETECTION METHOD maintains its accuracy and precision. Table 1 displays the
Nonlinear energy operator (NEO) is often compared to the resulting data.
ABS simple thresholding due to their commonality. This is TABLE I. COMPARISON OF POWER CONSUMPTION
another common detection method used in analyzing brain
spikes. NEO is seen to be more effective because it uses Spike Detection Algorithm Total Power (nW)
probability and statistics to determine the probability of NEO 53.6
detection and false alarm rate. Once again, as observed in the Hybrid Spike Detection 24.4
previous section, when applied, an ABS simple thresholding
technique consumes less power and can operate effectively with
low SNR, thus making it an attractive option to process brain V. CONCLUSION AND DISCUSSION
spikes. The goal of this combination is make an accurate and low The objective of this document is to give a brief overview of
power ASIC spike detector. common spike detection methods. The document proposed an
ABS simple thresholding method that consumes low power and [2] Obeid, I., and P.d. Wolf. "Evaluation of Spike-Detection
can operate under low SNR. This method was applied to a Algorithms for a Brain-Machine Interface Application." IEEE
wireless headstage integrated circuit, which utilized the lower Transactions on Biomedical Engineering 51.6 (2004): 905-
power operation and added an adaptive threshold to create an 11. IEEE. Web.
accurate spike detector. The next section was an interesting [3] Tam, Wing-Kin, Rosa So, Cuntai Guan, and Zhi Yang. "EC-PC
hybrid of the two most common algorithms used in spike Spike Detection for High Performance Brain-Computer
Interface." 2015 37th Annual International Conference of the
detection. It incorporated ABS thresholding and NEO. By taking
IEEE Engineering in Medicine and Biology Society
advantages from both technique, a lower power and an accurate (EMBC) (2015): n. pag. IEEE. Web.
detection, method was delivered. Sources [1], [2], and [5] were
[4] "What Are Brainwaves?" What Are Brainwaves? Types of Brain
heavily relied upon to attain simulation results and data. Waves | EEG Sensor and Brain Wave – UK. Symphonic Mind
Ltd, n.d. Web. 19 Mar. 2017.
REFERENCES
[5] Zeinolabedin, Seyed Mohammad Ali, Anh Tuan Do, Kiat Seng
[1] Gagnon-Turcotte, Gabriel, Yoan Lechasseur, Cyril Bories, Yeo, and Tony Tae-Hyoung Kim. "Design of A Hybrid Neural
Younes Messaddeq, Yves De Koninck, and Benoit Gosselin. "A Spike Detection Algorithm for Implantable Integrated Brain
Wireless Headstage for Combined Optogenetics and Circuits." 2015 IEEE International Symposium on Circuits and
Multichannel Electrophysiological Recording." IEEE Systems (ISCAS) (2015): n. pag. IEEE. Web.
Transactions on Biomedical Circuits and Systems 11.1 (2016): 1-
14. Web.

You might also like