Method and apparatus for evaluating state of neural network

Abstract

A method for evaluating a state of a neural network, according to an embodiment, may include the steps of: generating inter-spike interval (ISI) time-series data from a neural signal; identifying extrema including a local maxima and a local minima in the ISI time-series data; calculating at least one interval distribution on the basis of a time interval between the extrema; calculating an attention entropy (AE) value by calculating entropy of the interval distribution; and determining a maturity or an active state of the neural network on the basis of the AE value.

Description

Neural network state evaluation method and device

[0001] The following embodiments relate to neural signal analysis technology, and more specifically, to a method and system for quantitatively evaluating the differentiation, maturity, and activity status of a neural network by analyzing the inter-spike interval (ISI) of neural signals obtained from a microelectrode array (MEA), etc.

[0002]

[0003] Accurately evaluating the maturity of neural networks is essential for understanding the developmental processes of the human brain and nervous system, and for early diagnosis or the development of treatments for neurodevelopmental disorders such as autism spectrum disorder (ASD). Previously, neural activity was analyzed using methods such as Shannon entropy or mean firing rate; however, these methods had limitations in that they could not fully reflect the temporal structure or complexity of time-series data. Furthermore, non-linear analysis methods such as minimum embedding dimension (MED) analysis were computationally intensive and sensitive to data length, which restricted real-time analysis.

[0004] The information described above may be provided as related art for the purpose of aiding understanding of the present disclosure. No claim or determination is made as to whether any of the foregoing may be applied as prior art related to the present disclosure.

[0005]

[0006] A neural network state evaluation method according to one embodiment may include: generating inter-spike interval (ISI) time series data from a neural signal; identifying extrema including local maxima and local minima within the inter-spike interval time series data; calculating four interval distributions based on time intervals between the extrema; calculating an attention entropy (AE) value by calculating the entropy of the interval distributions; and determining the maturity or active state of the neural network based on the attention entropy value.

[0007] The step of calculating the attention entropy value includes the step of generating a plurality of interval series classified according to the types of the extrema, and the plurality of interval series may include at least one of an interval between local maximums, an interval between local minimums, and an intersection interval between local maximums and local minimums.

[0008] The step of calculating the attention entropy value may include: a step of calculating an individual Shannon entropy for each of the plurality of interval time series; and a step of calculating a representative value of the individual Shannon entropies and determining it as the attention entropy value.

[0009] The above plurality of interval time series may include a first interval time series consisting of intervals between consecutive local maximums; a second interval time series consisting of intervals between consecutive local minimums; a third interval time series consisting of intervals from a local maximum to a subsequent local minimum; and a fourth interval time series consisting of intervals from a local minimum to a subsequent local maximum.

[0010] The step of calculating the above representative value may include the step of summing and averaging the Shannon entropy values ​​calculated for each of the first interval time series, the second interval time series, the third interval time series, and the fourth interval time series to derive the attention entropy value.

[0011] A neural network state evaluation method according to one embodiment may further include the step of obtaining a filtered signal by performing bandpass filtering on the neural signal; and the step of detecting a spike point by applying an adaptive threshold to the filtered signal.

[0012] A neural network state evaluation method according to one embodiment may further include the step of generating surrogate data by randomly shuffling the order of the time series data of the interval between spikes; and the step of verifying the significance of the non-linear temporal structure of the neural network activity by comparing the attention entropy value of the time series data of the interval between spikes with the attention entropy value of the surrogate data.

[0013] The step of determining the state of the neural network may include a step of determining that the synaptic connections of the neural network are strengthened and functionally matured if the attention entropy value shows an increasing trend over time.

[0014] The step of determining the state of the neural network may include determining that the neural network has autism spectrum disorder (ASD) or developmental delay characteristics if the attention entropy value maintains a lower value compared to a normal control group during the same culture period.

[0015] The step of determining the state of the neural network may include determining that the higher the attention entropy value, the more dynamic equilibrium the excitatory signal and inhibitory signal within the neural network interact.

[0016] An electronic device according to one embodiment includes at least one processor comprising a processing circuit; and a memory for storing instructions, wherein when the instructions are executed individually or collectively by the at least one processor, the electronic device may generate inter-spike interval (ISI) time series data from a neural signal, identify extrema including local maxima and local minima within the inter-spike interval time series data, calculate at least one interval distribution based on the time interval between the extrema, calculate the entropy of the interval distribution to calculate an attention entropy (AE) value, and determine the maturity or active state of the neural network based on the attention entropy value.

[0017] When the above instructions are executed individually or collectively by the at least one processor, the electronic device generates a plurality of interval series classified according to the types of the extrema, and the plurality of interval series may include at least one of an interval between local maximums, an interval between local minimums, and an intersection interval between local maximums and local minimums.

[0018] When the above instructions are executed individually or collectively by the at least one processor, the electronic device may be configured to calculate an individual Shannon entropy for each of the plurality of interval time series and to calculate a representative value of the individual Shannon entropies and determine it as the attention entropy value.

[0019] The above plurality of interval time series may include a first interval time series consisting of intervals between consecutive local maximums; a second interval time series consisting of intervals between consecutive local minimums; a third interval time series consisting of intervals from a local maximum to a subsequent local minimum; and a fourth interval time series consisting of intervals from a local minimum to a subsequent local maximum.

[0020] When the above instructions are executed individually or collectively by the at least one processor, the electronic device may be configured to sum and average the Shannon entropy values ​​calculated for each of the first interval time series, the second interval time series, the third interval time series, and the fourth interval time series to derive the attention entropy value.

[0021] When the above instructions are executed individually or collectively by the at least one processor, the electronic device may perform bandpass filtering on the neural signal to obtain a filtered signal, and apply an adaptive threshold to the filtered signal to detect the spike point.

[0022] When the above instructions are executed individually or collectively by the at least one processor, the electronic device may generate surrogate data that randomly shuffles the order of the time series data between spike intervals, and compare the attention entropy value of the time series data between spike intervals with the attention entropy value of the surrogate data to verify the significance of the non-linear temporal structure of the neural network activity.

[0023] When the above instructions are executed individually or collectively by the at least one processor, the electronic device may determine that the synaptic connections of the neural network are strengthened and functionally mature if the attention entropy value shows an increasing trend over time.

[0024] When the above instructions are executed individually or collectively by the at least one processor, the electronic device may determine that the neural network has autism spectrum disorder (ASD) or developmental delay characteristics if the attention entropy value remains lower than that of a normal control group during the same culture period.

[0025] The higher the attention entropy value, the more likely it is to be determined that the neural network is in a dynamic equilibrium state where excitatory and inhibitory signals interact.

[0026]

[0027] FIG. 1 is a diagram illustrating the overall processing process of a neural network state evaluation method according to one embodiment.

[0028] FIG. 2 is a diagram specifically illustrating a neural signal measurement and data processing pipeline based on a microelectrode array (MEA) according to one embodiment.

[0029] Figure 3 is a graph showing a comparison of an example of time series data representing an attention entropy (AE) value, a spike train, and the corresponding inter-spike interval (ISI) according to one embodiment.

[0030] FIG. 4 is a conceptual diagram illustrating, step-by-step, the process of extracting and generating time series data of the interval between spikes (ISI) from continuous voltage data according to one embodiment.

[0031] FIG. 5 is a conceptual diagram illustrating the process of identifying local extrema in time series data of interval between spikes (ISI) to calculate attention entropy (AE) according to one embodiment, and extracting four types of interval distributions based thereon.

[0032] FIG. 6 is a diagram illustrating the concept of a surrogate test for verifying the validity of attention entropy according to one embodiment.

[0033] FIG. 7 may be a box plot graph showing the results of a surrogate test performed to verify the statistical significance of the Attention Entropy (AE) indicator according to one embodiment.

[0034] FIG. 8 is a graph showing the results of the analysis of neural signals obtained from a neural network that is a normal control according to one embodiment.

[0035] FIG. 9 is a graph showing the results of signal analysis of a neural network derived from a patient with Autism Spectrum Disorder (ASD) according to one embodiment.

[0036] Figure 10 is a graph showing the trend of changes in the neural network complexity index over the course of the culture period (Days In Vitro, DIV) of nerve cells according to one embodiment.

[0037] Figure 11 is a graph showing the results of a surrogate test to compare and verify the performance difference between Attention Entropy (AE) according to one embodiment and the conventional Shannon Entropy (SE).

[0038] FIG. 12 is a diagram showing experimental results for verifying the biological validity of the attention entropy (AE) index in a mouse primary neuron model according to one embodiment.

[0039] FIG. 13 is a flowchart illustrating the procedure of a neural network state evaluation method performed by a computing device according to one embodiment.

[0040] FIG. 14 is a block diagram of an electronic device according to one embodiment.

[0041]

[0042] The specific structural or functional descriptions disclosed in this specification are illustrative of embodiments according to technical concepts only, and the actual implemented form may take various other forms and is not limited to the embodiments described in this specification.

[0043] Terms such as "first" or "second" may be used to describe various components, but these terms should be understood solely for the purpose of distinguishing one component from another. For example, the first component may be named the second component, and similarly, the second component may be named the first component.

[0044] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. Conversely, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between. Expressions describing the relationships between components, such as "between" and "directly between," or "adjacent to" and "directly adjacent to," should be interpreted in the same way.

[0045] The singular expression includes the plural expression unless the context clearly indicates otherwise. In this specification, terms such as “comprising” or “having” are intended to specify the existence of the implemented features, numbers, steps, actions, components, parts, or combinations thereof, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0046] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this specification.

[0047] The embodiments can be implemented in various forms of products, such as personal computers, laptop computers, tablet computers, smartphones, televisions, smart home appliances, intelligent vehicles, kiosks, and wearable devices. The embodiments will be described in detail below with reference to the attached drawings. Identical reference numerals in each drawing indicate identical components.

[0048] FIG. 1 is a diagram illustrating the overall processing steps of a neural network state evaluation method according to one embodiment. A neural network state evaluation method according to one embodiment may largely consist of the steps of generating Inter-Spike Interval (ISI) time series data, calculating Attention Entropy (AE), and evaluating the neural network state based thereon.

[0049] ISI time series data according to one embodiment may be data that quantifies electrical activity occurring in a neuron or neural network over time. Neurons can generate electrical signals called spikes or action potentials in response to specific stimuli or voluntary activities. In this case, the time difference between two consecutively occurring spikes can be defined as the Inter-Spike Interval (ISI), which may be referred to as the firing interval. Unlike a histogram that simply lists the frequency of intervals, data that lists spikes while maintaining the order in which they occurred can be referred to as ISI time series data. Since ISI time series data contains information regarding the information processing process or temporal patterns of the neural network, it can be utilized as basic data for analyzing the dynamic characteristics of the neural network.

[0050] Attention Entropy (AE) according to one embodiment may be a quantitative indicator representing how complex and irregular the structure of input ISI time series data is. Unlike conventional Shannon Entropy, which considers only the overall frequency distribution of the data, Attention Entropy can be calculated by focusing on the distribution of extrema, such as local maxima and local minima, that appear within the waveform of the time series data. Specifically, Attention Entropy (AE) can be understood as a value that quantifies the uncertainty or complexity inherent in the change in intervals between extrema. Since Attention Entropy (AE) can sensitively capture non-linear and dynamic changes in neural signals, it can function as a complexity index of the neural network.

[0051] Assessment of neuronal network status according to one embodiment may be a process of determining the biological or functional state of a target neural network based on a calculated attention entropy (AE) value. The state of the neural network may include maturity, which refers to the degree to which neurons develop structurally and functionally over a culture period. Additionally, the state of the neural network may include an activity level, which indicates how densely and efficiently synaptic connectivity is formed. Furthermore, the assessment of the neural network's state may include a process of diagnosing the presence or absence of neurodevelopmental disorders such as Autism Spectrum Disorder (ASD) or monitoring changes in the neural network's response to drug administration. For example, if the attention entropy (AE) value is measured to be high, the neural network may be assessed as being in a mature and balanced state capable of responding flexibly to various stimuli. Conversely, if the attention entropy (AE) value is measured to be low, the neural network may be evaluated as being in a pathological state where development is delayed or specific patterns are fixed.

[0052] FIG. 2 is a diagram specifically illustrating a neural signal measurement and data processing pipeline based on a microelectrode array (MEA) according to one embodiment. The entire process can proceed in the steps of signal acquisition, preprocessing, spike detection, and attention entropy calculation.

[0053] Referring to drawing (210), extracellular potential or electrical activity signals from neurons cultured on a microelectrode array (MEA) plate can be detected non-invasively. The microelectrode array has a number of microelectrodes arranged in a grid pattern, allowing for the simultaneous recording of network activity of a neuronal population.

[0054] As shown in drawing (220), in the signal recording step, the voltage signal can be digitized and recorded at a sampling frequency of 12.5 kHz to secure high-resolution temporal resolution. To ensure the stability of the recording, measurements are performed for a total of 15 minutes, but the data from the subsequent 10 minutes can be used for analysis, excluding the initial 5 minutes of stabilization time. This may be to exclude temporary noise or instability caused by changes in the culture environment.

[0055] Referring to drawing (230), a preprocessing process may be performed to remove background noise from the acquired raw voltage signal and extract only the neural activity component. The preprocessing may include bandpass filtering that passes only a specific frequency band, and specifically, a filter in the 10 Hz to 2.5 kHz band may be applied. Through this, low-frequency drift or high-frequency electrical noise can be effectively attenuated.

[0056] As shown in drawing (240), a step of detecting spikes, which are significant action potentials, from the filtered signal can be performed. For spike detection, an adaptive threshold method that changes dynamically according to the noise level of the signal can be applied. Specifically, based on the mean or root mean square (RMS) noise level of the signal, a point that is 5.5 times or more of the standard deviation can be identified as the point of spike occurrence. The spikes detected in this way can be arranged over time for each electrode and visualized in the form of a raster plot, which can show a simultaneous firing pattern at multiple electrodes.

[0057] Referring to the drawing (250), the interval between spikes (ISI) is calculated from the detected spike times, and the attention entropy (AE) can be calculated through statistical distribution analysis. This process may be a step for evaluating the dynamic characteristics of the neural network by quantifying the variability and randomness of spike timing, going beyond simply counting the frequency of spikes.

[0058] Figure 3 is a graph showing a comparison of an example of time series data representing an attention entropy (AE) value, a spike train, and the corresponding inter-spike interval (ISI) according to one embodiment.

[0059] Referring to Figure 3, the top of each graph represents the point in time when a spike occurs over time, and the bottom represents the change in the interval between the spikes as a time series.

[0060] Drawing (310) may represent a case where the attention entropy (AE) value is maximum among the analyzed data. In this case, the timing of spike occurrence may exhibit a very irregular and complex pattern, and accordingly, the range of variation of the interval between spikes (ISI) may be large and have a non-linear structure. A high attention entropy (AE) value may imply that changes in the spike firing rate within the neural network occur unpredictably and complexly, which may correspond to a state where the neural network processes various information or secures a high level of functional connectivity.

[0061] Drawing (320) may indicate a case where the attention entropy (AE) value is minimum. Spikes may simply repeat at specific intervals or show a low firing frequency. A low attention entropy (AE) value may suggest that the change in spike intervals is regular and predictable, which may be associated with a state where the activity of the neural network is monotonic or not yet sufficiently mature, or an unbalanced state where either excitatory or inhibitory is dominant.

[0062] Drawing (330) is a case where it has an intermediate level of attention entropy (AE) value and may show a pattern mixed with randomness and regularity.

[0063] Drawing (340) may represent virtual spike sequences and ISI data simulated from random noise rather than biological signals.

[0064] The attention entropy (AE) according to one embodiment can be used as an indicator to quantitatively and clearly distinguish between a signal with biological complexity as in Figure (310), simple random noise as in Figure (340), or a simple repeating pattern as in Figure (320).

[0065] FIG. 4 is a conceptual diagram illustrating, step-by-step, the process of extracting and generating time series data of the interval between spikes (ISI) from continuous voltage data according to one embodiment.

[0066] Referring to the drawing (410), the waveform of the raw voltage signal (raw voltage data) recorded through the microelectrode array (MEA) can be observed. This waveform continuously shows changes in voltage over time and may include spikes corresponding to the action potentials of nerve cells.

[0067] As illustrated in drawing (420), when spike points exceeding a threshold are detected in a continuous voltage signal, an interval, which is the time difference between two adjacent spikes, can be calculated. For example, the interval between the first spike and the second spike can be defined sequentially as y1, and the interval between the second and the third as y2. 2 .

[0068] Referring to the drawing (430), the calculated interval values ​​can be arranged according to the spike occurrence order (spike index) and converted into a single time series data. In this case, the X-axis can represent the spike order (index), and the Y-axis can represent the interval time at that order.

[0069] According to one embodiment, unlike simply measuring the frequency of spike occurrences, this transformation process can preserve the temporal order and change patterns of interval values. The ISI time series data generated in this way can be used as basic input data for subsequent attention entropy (AE) calculations, thereby contributing to the quantification of the dynamic characteristics of neural network activity.

[0070] FIG. 5 is a conceptual diagram illustrating the process of identifying local extrema in time series data of interval between spikes (ISI) to calculate attention entropy (AE) according to one embodiment, and extracting four types of interval distributions based thereon.

[0071] The input ISI time series data can form a waveform that increases or decreases over time. As shown in the figure, local maxima (Peak), which is a point within the waveform where the value increases and then decreases, can be defined as p1, p2, p3, ..., and conversely, local minima (Trough), which is a point where the value decreases and then increases, can be defined as t1, t2, t3, ....

[0072] According to one embodiment, the time intervals between identified extreme values ​​can be classified into the following four types of interval series depending on their connection type. 4 .

[0073] 1. Pxx(Peak-to-Peak): Can be a time series of intervals between consecutive local maximums.

[0074] 2. Pnn(Trough-to-Trough): Can be a time series of intervals between consecutive local minimums.

[0075] 3. Pxn(Peak-to-Trough): This can be a time series of intervals from a local maximum to the immediately following local minimum.

[0076] 4. Pnx(Trough-to-Peak): This can be a time series of intervals from a local minimum to the immediately following local maximum.

[0077] Finally, the attention entropy (AE) value can be calculated by determining the probability distribution for each of the four interval time series mentioned above, calculating the Shannon entropy of each distribution, and then taking the average of these values. This can serve as an indicator that comprehensively reflects not only the frequency of simple intervals but also the pattern of change between extreme values ​​and structural complexity.

[0078] FIG. 6 is a diagram illustrating the concept of a surrogate test for verifying the validity of attention entropy according to one embodiment.

[0079] Referring to Fig. 6, the top graph may represent original inter-spike interval (ISI) time series data obtained from neural signals, and the bottom graph may represent surrogate data generated by randomly shuffling the temporal order of the original ISI data.

[0080] According to one embodiment, the surrogate test may be a verification method that destroys the inherent temporal pattern or correlation of the data, while maintaining the overall statistical distribution or histogram of the data values ​​identical to the original.

[0081] Attention entropy (AE) according to one embodiment has the characteristic that the interval distribution changes depending on the arrangement order of local extrema, so the calculated value may change if the order of the data changes. On the other hand, conventional Shannon entropy considers only the frequency of data values, so it has the characteristic of maintaining the same value even if the order changes.

[0082] Therefore, by comparing the attention entropy values ​​of the original data and the surrogate data, it can be verified that the measured neural activity is not a simple random signal but contains a biologically significant non-linear temporal structure.

[0083] FIG. 7 may be a box plot graph showing the results of a surrogate test performed to verify the statistical significance of the Attention Entropy (AE) indicator according to one embodiment.

[0084] Referring to Figure 7, the red square points (orig) in each graph represent the attention entropy values ​​of the original inter-spike interval (ISI) time series data, and the box plot can represent the attention entropy distribution range of 100 surrogate data generated by randomly shuffling the order of the original data.

[0085] Referring to drawing (710), the analysis results of a complex signal having the highest attention entropy (Max AE) value can be illustrated. The attention entropy value of the original data may be located at a position significantly higher than the distribution range of the surrogate data, which may suggest that the original data has a statistically distinct non-linear temporal structure compared to randomly mixed data. An asterisk (***) at the top of the graph indicates that the statistical significance probability (p-value) is less than 0.001, which may show that the difference between the two data groups is very significant.

[0086] Drawing (720) may represent the case of a simple or regular signal having the lowest attention entropy (Min AE) value. Even in this case, the value of the original data can be clearly distinguished from the distribution of the surrogate data. This can serve as evidence to prove that the low attention entropy value is not simply due to a lack of data or errors, but is due to the inherent regularity or structural characteristics of the signal.

[0087] Drawing (730) can represent the case of a signal having an intermediate level of attention entropy (Intermediate AE) value, and a significant difference between the original data and the surrogate data can also be observed.

[0088] On the other hand, drawing (740) may show the results of the analysis of a signal generated from random noise. Referring to drawing (740), it can be seen that the attention entropy value of the original data is located within the distribution range (box plot) of the surrogate data. This may mean that the original signal itself has random characteristics, so there is no significant change in the entropy value even if the order is mixed.

[0089] Consequently, the experimental results of Figure 7 can be used to verify that attention entropy (AE) is not merely a measure of the randomness of data, but an indicator capable of sensitively capturing temporal correlations and sequential patterns inherent in neural signals.

[0090] FIG. 8 is a graph showing the results of the analysis of neural signals obtained from a neural network that is a normal control according to one embodiment.

[0091] Referring to drawing (810), a raster plot visualizing the timing of spike occurrences recorded in a population of normally differentiated neurons can be shown. In the case of a normal control group, frequent and dense spike firing is observed at multiple electrodes, which may suggest that active interaction between neurons is taking place.

[0092] Drawing (820) may represent time series data of the interval between spikes (ISI) extracted from the spike column of drawing (810). The Y-axis of the graph represents the size of the interval according to the spike index, and a pattern in which the interval values ​​change in various and irregular ways over time can be observed. This high variability may indicate that the neural network is not repeating a fixed pattern, but is in a complex dynamic state capable of processing various information.

[0093] Drawing (830) may be a histogram showing four interval distributions (Pxx, Pnn, Pnx, Pxn) between local extrema identified in ISI time series data. In all four distributions, a relatively wide spread without being skewed toward specific values ​​may be observed, which may be the cause of high Shannon entropy values ​​for each distribution.

[0094] FIG. 9 is a graph showing the results of signal analysis of a neural network derived from a patient with Autism Spectrum Disorder (ASD) according to one embodiment.

[0095] Referring to Figure (910), a raster plot showing the timing of spike occurrences recorded in ASD-derived neurons may be illustrated. Compared to the normal control group illustrated in Figure 8, Figure (910) may show a relatively low frequency of spike events and a pattern in which the spikes are sporadic or dispersed over time. This may suggest that signal transmission or network synchronization between neurons is weaker than in the normal group.

[0096] Drawing (920) may represent time series data of the interval between spikes (ISI) corresponding to the spike column of drawing (910). Looking at the graph, it can be seen that the change in interval values ​​is monotonous or that specific patterns are poorly observed. This may mean that the complex and diverse temporal variability observed in the normal control group is significantly reduced in the ASD neural network.

[0097] Figure (930) may be a histogram representing four interval distributions (Pxx, Pnn, Pnx, Pxn) derived from ISI time series data. Unlike the normal control distribution in Figure 8, the distributions in Figure (930) may show a tendency to have low frequencies or be concentrated in specific intervals. In particular, the overall disorder may appear low, such as when the entropy is measured to be low in the distribution of the crossover interval (Pxn) between local maximums and local minimums.

[0098] Consequently, in the case of the ASD neural network illustrated in Fig. 9, the attention entropy (AE) value can be calculated to be significantly lower than that of a normal control group. This low attention entropy value serves as an indicator that quantitatively represents the reduction in complexity and functional connectivity of the neural network, and can be utilized for the diagnosis of neurodevelopmental disorders or the identification of pathological conditions.

[0099] Figure 10 is a graph showing the trend of changes in the neural network complexity index over the course of the culture period (Days In Vitro, DIV) of nerve cells according to one embodiment.

[0100] Referring to Figure 10, the graph on the left shows the change in attention entropy (AE), and the graph on the right may show the change in minimum embedding dimension (MED), which is an existing non-linear analysis method, for comparison.

[0101] Referring to the graph on the left, the blue line corresponding to the normal control group (CON) may show a tendency for attention entropy (AE) values ​​to rise sharply from day 4, the early stage of culture, to day 17. This may quantitatively reflect the maturation process, in which synaptic connections are formed and the activity patterns of the neural network become increasingly complex as neurons differentiate. On the other hand, the red line corresponding to the autism spectrum disorder (ASD) group can be observed to maintain significantly lower attention entropy values ​​compared to the normal control group throughout the overall culture period.

[0102] Referring to the graph on the right, the existing complexity analysis indicator MED can also exhibit a time-series change pattern very similar to AE. Both indicators tend to reach a peak in the mid-stage of differentiation (around day 17) and then stabilize or decrease.

[0103] Consequently, Figure 10 may serve as evidence that attention entropy (AE) can detect the developmental process of neural networks and pathological differences at a level equivalent to existing MED analysis methods that require complex computations. Furthermore, unlike MED, attention entropy does not require parameter setting and has a low computational load, suggesting that it can be utilized as an efficient analysis tool.

[0104] Figure 11 is a graph showing the results of a surrogate test to compare and verify the performance difference between Attention Entropy (AE) according to one embodiment and the conventional Shannon Entropy (SE).

[0105] Referring to Figure 11, the graph on the left may be the result of analysis based on data from the normal control group, and the graph on the right may be the result based on data from the autism spectrum disorder (ASD) group.

[0106] In each graph, the horizontal axis represents the entropy indices (AE, AExx, AEnn, AExn, AEnx, SE) used in the analysis, and the vertical axis represents the calculated entropy values. Additionally, the red marker (orig) represents the entropy values ​​of the original data, and the blue box plot represents the entropy distribution of surrogate data generated by randomly shuffling the order of the original data 100 times.

[0107] Referring to the graph, it can be observed that for Attention Entropy (AE) and its detailed components, the values ​​of the original data (red markers) are distinctly separated from the distribution range of the surrogate data (box plot). Statistical analysis suggests that this difference may be highly significant. This implies that the value of Attention Entropy changes when the order of the data is mixed, suggesting that it sensitively captures the fact that the original neural signal possesses a unique temporal structure rather than being random noise.

[0108] On the other hand, if we refer to the Shannon entropy (SE) item located on the far right of the graph, we can observe that the values ​​of the original data and the distribution of the surrogate data are nearly identical or overlap. Since Shannon entropy is permutation-invariant, meaning it considers only the frequency of values ​​regardless of the temporal order of the data, it may exhibit limitations in that it cannot distinguish between the complexity of the original signal and the complexity of the randomly mixed signal.

[0109] FIG. 12 is a diagram showing experimental results for verifying the biological validity of the attention entropy (AE) index in a mouse primary neuron model according to one embodiment.

[0110] Referring to Drawing (1210), immunofluorescence staining images of neurons according to culture periods (1 week, 2 weeks, 3 weeks) may be shown. In the images, green may represent GFAP (Glial Fibrillary Acidic Protein), an astrocyte marker; red may represent MAP2 (Microtubule-Associated Protein 2), a marker of mature neurons; and blue may represent DAPI, which stains the nuclei. As shown in Drawing (1210), as culture progresses from week 1 to week 3, a pattern can be observed in which the expression of MAP2 (red) gradually increases and the structure of the neural network becomes denser. This can visually demonstrate that the neurons are physically growing and forming a structural network over time.

[0111] Drawing (1220) may be a graph showing the number of MAP2-positive cells by quantitatively analyzing the image of Drawing (1210). Looking at the graph, the number of MAP2-positive cells may show a statistically significant increasing trend from week 1 to week 2 and week 3. This can be quantitatively proven that the differentiation and maturation of nerve cells are progressing as the culture period progresses.

[0112] Drawing (1230) may be a graph showing the trend of changes in attention entropy (AE) values ​​calculated based on neural signals measured under the same culture conditions. In the graph, the attention entropy values ​​may show a tendency to gradually increase from the beginning of culture and reach a peak around day 17. This pattern of increasing attention entropy values ​​may show a high correlation with the biological maturation process (increase in MAP2 expression) identified in Drawing (1210) and Drawing (1220).

[0113] FIG. 13 is a flowchart illustrating the procedure of a neural network state evaluation method performed by a computing device according to one embodiment.

[0114] In step (1310), the computing device may perform the step of generating time series data of the interval between spikes (ISI) from the neural signal. Specifically, step (1310) may include a process of removing noise by performing bandpass filtering on a raw voltage signal obtained through a microelectrode array (MEA), etc. Additionally, it may include a process of detecting spike timings by applying an adaptive threshold based on the mean or standard deviation of the signal to the filtered signal, and calculating the time difference between adjacent spikes and arranging them in a time series form.

[0115] In step (1320), a step of identifying extrema including local maxima and local minima within the generated time series data between spike intervals may be performed. This may be a process of analyzing the overall waveform of the time series data to find peaks where values ​​increase and then decrease, and troughs where values ​​decrease and then increase.

[0116] Referring to step (1330), a step of calculating at least one interval distribution based on the time interval between the identified extreme values ​​may be performed. According to one embodiment, this step may include the process of generating a plurality of interval time series according to the type of extreme values. Specifically, a first interval time series consisting of intervals between consecutive local maximums, a second interval time series consisting of intervals between consecutive local minimums, a third interval time series consisting of intervals from a local maximum to a subsequent local minimum, and a fourth interval time series consisting of intervals from a local minimum to a subsequent local maximum may be generated.

[0117] In step (1340), the attention entropy (AE) value may be calculated by computing the entropy of the calculated interval distribution. Step (1340) may include the process of calculating individual Shannon entropy for each of the first to fourth interval time series, and the process of summing and averaging the four calculated Shannon entropy values ​​to derive a single representative value, the attention entropy. Through this, complexity based on temporal structure, rather than simple randomness of the signal, can be quantified.

[0118] Finally, referring to step (1350), a step of determining the maturity or active state of the neural network based on the calculated attention entropy value may be performed. According to one embodiment, if the attention entropy value shows an increasing trend over time, it can be determined that the synaptic connections of the neural network are strengthened and functionally mature. In another embodiment, if the attention entropy value maintains a lower value compared to a normal control group during the same culture period, the neural network may be determined to have autism spectrum disorder (ASD) or developmental delay characteristics. Additionally, the higher the attention entropy value, the more it can be interpreted as a dynamic balance state (E / I balance) in which excitatory and inhibitory signals within the neural network interact.

[0119] FIG. 14 is a block diagram of an electronic device according to one embodiment. The configuration and operation described with reference to FIG. 1 to 13 may be applied to the embodiment of FIG. 14 in a similar manner or with some modifications, and therefore, in the description of FIG. 14, detailed descriptions of parts that are substantially similar to those already described with reference to FIG. 1 to 13 may be omitted.

[0120] Referring to FIG. 14, an electronic device (1400) according to one embodiment may include a memory (1410) and a processor (1430).

[0121] The memory (1410) can store instructions (e.g., programs) executable by the processor (1430). For example, the instructions may include instructions for executing the operation of the processor (1430) and / or the operation of each component of the processor (1430).

[0122] The memory (1410) can be implemented as a volatile memory device or a non-volatile memory device.

[0123] Volatile memory devices can be implemented as DRAM (dynamic random access memory), SRAM (static random access memory), T-RAM (thyristor RAM), Z-RAM (zero capacitor RAM), or TTRAM (Twin Transistor RAM).

[0124] Non-volatile memory devices can be implemented as EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, MRAM (Magnetic RAM), Spin-Transfer Torque (STT)-MRAM, Conductive Bridging RAM (CBRAM), FeRAM (Ferroelectric RAM), PRAM (Phase change RAM), Resistive RAM (RRAM), Nanotube RRAM, Polymer RAM (PoRAM), Nano Floating Gate Memory (NFGM), holographic memory, Molecular Electronic Memory Device, or Insulator Resistance Change Memory.

[0125] The processor (1430) can process data stored in memory (1410). The processor (1430) can execute computer-readable code (e.g., software) stored in memory (1410) and instructions triggered by the processor (1430).

[0126] The processor (1430) may be a data processing device implemented in hardware having a circuit having a physical structure for executing desired operations. For example, the desired operations may include code or instructions included in a program.

[0127] For example, a data processing device implemented in hardware may include a microprocessor, a central processing unit, a processor core, a multi-core processor, a multiprocessor, an Application-Specific Integrated Circuit (ASIC), and a Field Programmable Gate Array (FPGA).

[0128] A processor (1430) can generate inter-spike interval (ISI) time series data from a neural signal, identify extrema including local maxima and local minima within the inter-spike interval time series data, calculate at least one interval distribution based on the time interval between the extrema, calculate the entropy of the interval distribution to calculate an attention entropy (AE) value, and determine the maturity or active state of the neural network based on the attention entropy value.

[0129] The embodiments described above may be implemented as hardware components, software components, and / or combinations of hardware and software components. For example, the devices, methods, and components described in the embodiments may be implemented using a general-purpose computer or a special-purpose computer, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing unit may execute an operating system (OS) and software applications executed on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing unit may be described as being used as a single unit, but those skilled in the art will understand that the processing unit may include multiple processing elements and / or multiple types of processing elements. For example, the processing unit may include multiple processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are also possible.

[0130] Software may include computer programs, code, instructions, or a combination thereof, and may configure a processing unit to operate as desired or command the processing unit independently or collectively. Software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium or device, or transmitted signal wave in order to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over networked computer systems and may be stored or executed in a distributed manner. Software and data may be stored on computer-readable recording media.

[0131] The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either alone or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the embodiment, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc.

[0132] Although the embodiments have been described above with reference to the limited drawings, those skilled in the art can apply various technical modifications and variations based on the above. For example, suitable results may be achieved even if the described techniques are performed in a different order than described, and / or if the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0133] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below.

Claims

1. A step of generating inter-spike interval (ISI) time series data from neural signals; A step of identifying extreme values ​​including local maxima and local minima within the above-mentioned time series data of the interval between spikes; A step of calculating four interval distributions based on the time intervals between the above extreme values; A step of calculating the attention entropy (AE) value by calculating the entropy of the above interval distribution; and A step of determining the maturity or active state of a neural network based on the above attention entropy value A neural network state evaluation method including 2. In Paragraph 1, The step of calculating the attention entropy value above is, A step of generating a plurality of interval series classified according to the types of the above extreme values Includes, The above plurality of interval time series are including at least one of the interval between local maximums, the interval between local minimums, and the intersection interval between local maximums and local minimums. Neural network state evaluation method.

3. In Paragraph 2, The step of calculating the above attention entropy value A step of calculating individual Shannon entropy for each of the plurality of interval time series; and A step of calculating representative values ​​of the individual Shannon entropies and determining them as attention entropy values. A neural network state evaluation method including 4. In Paragraph 3, The above plurality of interval time series are A first interval time series consisting of intervals between consecutive local maximums; A second interval time series consisting of intervals between consecutive local minimums; A third interval time series consisting of intervals from a local maximum to a subsequent local minimum; and A fourth interval time series consisting of intervals from a local minimum to a subsequent local maximum A neural network state evaluation method including 5. In Paragraph 4, The step of calculating the above representative value A step of deriving the attention entropy value by summing and averaging the Shannon entropy values ​​calculated for each of the first interval time series, the second interval time series, the third interval time series, and the fourth interval time series. A neural network state evaluation method including 6. In Paragraph 1, A step of obtaining a filtered signal by performing bandpass filtering on the above neural signal; and A step of detecting spike timing by applying an adaptive threshold to the filtered signal above. A neural network state evaluation method that further includes 7. In Paragraph 1, A step of generating surrogate data by randomly shuffling the order of the time series data of the above-mentioned spike intervals; and A step of verifying the significance of the non-linear temporal structure of the neural network activity by comparing the attention entropy value of the time series data of the interval between spikes with the attention entropy value of the surrogate data. A neural network state evaluation method that further includes 8. In Paragraph 1, The step of determining the state of the above neural network is A step of determining that the synaptic connections of the neural network are strengthened and functionally matured when the attention entropy value shows an increasing trend over time. A neural network state evaluation method including 9. In Paragraph 1, The step of determining the state of the above neural network is A step of determining that the neural network has autism spectrum disorder (ASD) or developmental delay characteristics if the attention entropy value maintains a lower value compared to a normal control group of the same culture period. A neural network state evaluation method including 10. In Paragraph 1, The step of determining the state of the above neural network is A step of determining that the higher the attention entropy value, the more dynamic equilibrium the excitatory and inhibitory signals within the neural network interact. A neural network state evaluation method including 11. In an electronic device, At least one processor including a processing circuit; and memory that stores instructions Includes, When the above instructions are executed individually or collectively by the at least one processor, the electronic device, Generates inter-spike interval (ISI) time series data from neural signals, and Identify extreme values ​​including local maxima and local minima within the above spike interval time series data, and Calculate at least one interval distribution based on the time interval between the above extreme values, and Calculate the entropy of the above interval distribution to calculate the Attention Entropy (AE) value, and An electronic device for determining the maturity or active state of a neural network based on the above attention entropy value.

12. In Paragraph 11, When the above instructions are executed individually or collectively by the at least one processor, the electronic device, Generate multiple interval series classified according to the types of the above extreme values, and The above plurality of interval time series are An electronic device comprising at least one of a gap between local maximums, a gap between local minimums, and a gap between local maximums and local minimums.

13. In Paragraph 12, When the above instructions are executed individually or collectively by the at least one processor, the electronic device, Calculate the individual Shannon entropy for each of the above plurality of interval time series, and An electronic device that calculates a representative value of the individual Shannon entropies and determines the attention entropy value.

14. In Paragraph 13, The above plurality of interval time series are A first interval time series consisting of intervals between consecutive local maximums; A second interval time series consisting of intervals between consecutive local minimums; A third interval time series consisting of intervals from a local maximum to a subsequent local minimum; and A fourth interval time series consisting of intervals from a local minimum to a subsequent local maximum An electronic device including 15. In Paragraph 14, When the above instructions are executed individually or collectively by the at least one processor, the electronic device, An electronic device that derives the attention entropy value by summing and averaging the Shannon entropy values ​​calculated for each of the first interval time series, the second interval time series, the third interval time series, and the fourth interval time series.

16. In Paragraph 11, When the above instructions are executed individually or collectively by the at least one processor, the electronic device, Bandpass filtering is performed on the above neural signal to obtain a filtered signal, and An electronic device that detects spike timing by applying an adaptive threshold to the filtered signal.

17. In Paragraph 11, When the above instructions are executed individually or collectively by the at least one processor, the electronic device, Generate surrogate data by randomly shuffling the order of the above spike interval time series data, and An electronic device that verifies the significance of the non-linear temporal structure of the neural network activity by comparing the attention entropy value of the time series data of the interval between the spikes with the attention entropy value of the surrogate data.

18. In Paragraph 11, When the above instructions are executed individually or collectively by the at least one processor, the electronic device, An electronic device that determines that the synaptic connections of the neural network are strengthened and functionally mature when the attention entropy value shows an increasing trend over time.

19. In Paragraph 11, When the above instructions are executed individually or collectively by the at least one processor, the electronic device, An electronic device that determines that the neural network has autism spectrum disorder (ASD) or developmental delay characteristics when the attention entropy value maintains a lower value compared to a normal control group of the same culture period.

20. In Paragraph 11, An electronic device that determines that the higher the attention entropy value, the more likely it is to be in a dynamic equilibrium state in which excitatory signals and inhibitory signals within the neural network interact.

Images