Big data visual function intelligent analysis method and system

Abstract

The application relates to the technical field of neurosciences, in particular to a big data visual function intelligent analysis method and system. The system comprises a cascade action potential extraction module, an action potential separation module, a function analysis module and an integration and statistics module; the action potential extraction module is used for screening effective recording channels, extracting action potential sequence information and analyzing visual stimulation information; the action potential separation module is used for integrating single big data type electrophysiological data of multiple effective channels, classifying and obtaining action potential sequence information of a single neuron; further outputting action potential sequence information of a single type neuron; the function analysis module is used for visual function analysis of the neuron; and the integration and statistics module is used for constructing a database and performing analysis and evaluation. The method can solve the problem that the prior art lacks an analysis method and system for high-throughput electrophysiological data for visual research, and has the characteristics of high efficiency and high precision.

Description

Technical Field

[0001] This application relates to the field of neuroscience technology, and in particular to a method and system for intelligent analysis of visual functions in big data. Background Technology

[0002] The analysis of visual neuron functional encoding, the construction of functional classification maps, and the assessment of differences in disease and treatment models all require the collection and analysis of large-scale, high-throughput visual neuronal electrophysiological data. Recently, with the innovation and development of microelectrode technology, high-throughput electrophysiological recording technology (with over 4000 channels) can efficiently collect large-scale electrophysiological data of neurons compared to conventional low-throughput electrophysiological recording techniques. It has become the most cutting-edge technology in the field of electrophysiological research and has been successfully applied to various studies in neuroscience, such as retinal, hippocampal, stem cell, central neural network, and pharmacological research. This technology can simultaneously stimulate and record the electrophysiological signals and network information of multiple neurons within neural tissue. This not only allows us to understand the electrophysiological characteristics of individual types of neurons within neural tissue but also to analyze the interrelationships between neurons in neural circuits. Therefore, the analysis of this high-throughput electrophysiological data will contribute to a high-resolution understanding of the fundamentals of visual neural circuits and provide a new foundation for the accurate evaluation of various visual disease-related models and their treatment effects.

[0003] Given the high resolution and high data acquisition throughput of high-throughput electrophysiological recording technology, there is still a lack of matching big data neuroscience analysis systems, especially big data visual coding analysis systems that are combined with visual research. Therefore, how to invent a visual function intelligent analysis method and system for high-throughput electrophysiological data suitable for visual research is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] Therefore, it is necessary to provide a high-throughput intelligent analysis method and system for visual function of electrophysiological data suitable for visual research, which is characterized by high efficiency, high accuracy, strong intelligence and user-friendly visualization, to address the above-mentioned technical problems.

[0005] To achieve the above-mentioned objectives of this invention, the technical solution adopted is as follows:

[0006] We acquire and filter effective recording channels from high-throughput electrophysiological big data, extract action potential sequence information, and parse visual stimulus information based on the duration information of simulated input signals.

[0007] Action potential sequence information from multiple electrophysiological data is integrated, and non-visual stimulus time periods are identified and removed based on visual stimulus information to obtain a single integrated electrophysiological big data. The single integrated electrophysiological big data is classified through principal component analysis and multiple clustering methods to obtain action potential sequence information of a single neuron under different visual stimuli. Repeatedly collected and non-target neuronal signals are intelligently identified and automatically removed from the action potential sequence information of a single neuron to obtain non-repetitive action potential sequence information of a single type of neuron.

[0008] Visual features of non-repetitive action potential sequences of a single type of neuron are analyzed, and spatiotemporal visual function is analyzed to obtain analytical data of that type of neuron, including visual function indicators and feature curves.

[0009] Integrate and statistically analyze the parsed data of all types of neurons, and output the analysis results.

[0010] A big data visual function intelligent analysis system, the system comprising a cascaded action potential extraction module, an action potential separation module, a function analysis module, and an integration and statistics module;

[0011] The action potential extraction module is used to screen effective recording channels from high-throughput electrophysiological data, extract action potential sequence information, and parse visual stimulus information based on the duration difference information of the simulated input signal, and input it into the action potential separation module.

[0012] The action potential separation module receives and integrates action potential sequence information from multiple electrophysiological data. Based on visual stimulus information, it identifies and removes non-visual stimulus time periods to obtain a single integrated electrophysiological big data. The single integrated electrophysiological big data is classified using principal component analysis and multiple clustering methods to obtain action potential sequence information of a single neuron integrated under multiple visual stimuli or different experimental conditions. The integrated action potential sequence information is then automatically separated into action potential sequence information for a single visual stimulus type. The module intelligently identifies and automatically removes repetitive and non-target neuronal signals from the action potential sequence information of a single neuron, and outputs the non-repetitive action potential sequence information of a single type of neuron to the functional analysis module.

[0013] The functional parsing module is used to receive and parse the visual features of the action potential sequence information of a single type of non-repetitive neuron, perform visual spatiotemporal functional analysis, and obtain and output the parsing data of this type of neuron, including visual function indicators and feature curves, to the integrated statistics module.

[0014] The integrated statistics module is used to receive, integrate, and statistically analyze the parsed data of all types of neurons, establish a visual function database, and visualize the parsed results.

[0015] Preferably, the visual stimulus information includes visual stimulus type, stimulus start and end time, key time points, the start and end times of light application and removal in flashing stimuli and their stimulation time, the stimulation time of each frame of white noise, the direction of each random run of the bar stimulus, the spectral information and order of each random sequence of color stimuli, and other visual stimulus information.

[0016] Furthermore, the action potential extraction module includes an effective channel fast detection submodule and an action potential detection submodule, with the output of the effective channel fast detection submodule serving as the input of the action potential detection submodule;

[0017] The effective channel rapid detection submodule is used to access the electrophysiological data of the recording channels with specified names on the hard disk through memory mapping; generate image format information for each recording channel to facilitate offline screening and calibration; perform hierarchical integration and screening based on the magnitude of action potential amplitude, and combine the functions of channel selection, time period selection and visual stimulus signal simulation to screen out the recording channels with effective photoresponse, thereby reducing the computational resource consumption of invalid information in the subsequent analysis module;

[0018] The action potential detection submodule is used to automatically eliminate invalid recording channels by parallel computing and clustered reading of the effective photoresponse recording channels, setting a detection standard of action potential emission frequency threshold, thereby obtaining action potential information in the effective recording channels within each original high-throughput data. This can eliminate invalid recording channels, thereby reducing the waste of subsequent computing and storage resources and parsing the action potential information contained in all recording channels within each data. It also extracts action potential sequence information from the original electrophysiological data at high speed by selecting bandpass filtering and threshold detection methods. Based on the duration information of the simulated input signal, it parses the individual visual stimulus information features and automatically outputs the electrophysiological data of a single visual stimulus from the integrated recording of multiple visual stimuli.

[0019] Furthermore, the extraction of action potential sequence information from the raw electrophysiological data using bandpass filtering and threshold detection is specifically as follows: First, bandpass filtering is performed within a certain frequency range by setting low-frequency and high-frequency thresholds to obtain filtered data containing action potential characteristics; then, threshold standards are set to extract action potentials from the filtered data; next, outliers in the action potential amplitude distribution characteristics are analyzed, and noise waveforms contained in the action potential sequence information are filtered and removed based on the outliers to extract valid action potential sequence information; the data characteristics containing action potential sequences after filtering in a single recording channel are intelligently visualized, as well as the extracted action potential waveforms and quantity characteristics, and the operation of recalibrating and deleting action potential information in a single channel can be performed again.

[0020] Furthermore, the action potential separation module specifically includes a multi-data integration submodule, a big data action potential high-speed classification submodule, an integrated action potential separation submodule, and an action potential information recognition submodule;

[0021] The multi-data integration submodule is used to receive the action potential sequence information of a single visual stimulus output by the action potential extraction module. For multiple electrophysiological big data, it identifies and removes the action potential sequence information of non-stimulation time periods in the action potential sequence information. Then, according to the data time interval setting, it calibrates and adds the multiple electrophysiological big data. Through parallel computing, it obtains the re-integration of the action potential time series to get the single integrated electrophysiological big data, which contains the start time of each data and its recording information.

[0022] The high-speed classification submodule for big data action potentials is used to classify the action potential sequences of multiple neurons in each electrode according to principal component analysis and clustering methods to obtain the action potential sequence of a single neuron; based on the information features of the same array of neurons under different conditions, it intelligently calculates and automatically matches the action potential sequence information of a single neuron in different data under different conditions.

[0023] The integrated action potential separation submodule is used to separate and calibrate the action potential time series of the individual neurons according to the data name before integration, stimulation start time, stimulation end time, and sampling frequency of each data, thereby obtaining electrophysiological data with different sampling frequencies and action potential classification information, and finally realizing the synchronization and parsing of action potential separation in all recording channels of the multiple electrophysiological big data.

[0024] The action potential information recognition submodule is used to identify the electrophysiological data output by the integrated action potential separation submodule, intelligently identify the source of neuronal cell body, axon, or dendrite signals and remove redundant signals repeatedly collected from the same neuron, intelligently identify and remove signals from cells without long processes that have action potential firing characteristics, and retain only non-repetitive action potential sequence information of a single type of neuron.

[0025] Furthermore, the high-speed classification submodule for big data action potentials performs action potential classification, specifically including the following steps:

[0026] Principal component analysis is used to reduce the dimensionality of the high-dimensional original action potential sequence information contained in each electrode, transforming the classification of action potential sequence information into a clustering problem of points in a two-dimensional plane. By combining Valley clustering and K-means clustering methods, the integrated waveform of action potential sequence information is intelligently clustered into multiple action potential point sets of neurons in multiple recording channels based on the differences in amplitude and time characteristics through parallel computing and memory mapping.

[0027] Based on the peak potential interval standard, noise waveforms deviating from the average waveform are automatically removed to obtain the action potential sequence information of the individual neuron. It also incorporates an action potential waveform classification and comparison function. Through progressively increasing clustering parameters, matching the maximum correlation, matching the shortest distance, and the lowest noise level threshold, the best matching result is selected from multiple clustering calculations. This allows for the analysis of the one-to-one correspondence between subclass neurons in each recording channel within multiple electrophysiological big data sets induced by different visual stimuli and pharmacological conditions in the same neural tissue, enabling synchronous comparison of the same neuron under different conditions. Based on the number of action potentials after clustering, noise waveforms are automatically identified and parameters are intelligently adjusted to obtain the best neuron classification result. The classification results of neurons in each recording channel are intelligently visualized.

[0028] Furthermore, the functional analysis module specifically includes a flashing stimulus analysis submodule, a color stimulus analysis submodule, a Bar stimulus analysis submodule, a Chirp stimulus analysis submodule, a spatiotemporal characteristic analysis submodule, and an ipRGCs characteristic analysis submodule;

[0029] The flickering stimulus analysis submodule is used to perform white light flickering stimulus analysis on the non-repetitive action potential sequence information of the single type of neuron. Through parallel computation and pre- and post-stimulation response histogram plotting, it evaluates visual functional indicators such as maximum light response intensity, average light response intensity, delay time, and signal-to-noise ratio. Based on the characteristics of light-on and light-off stimuli, it calculates the dominant response index, automatically classifying neurons into light-on, light-on-off, and light-off types. Based on the response time-course characteristics, it calculates the response time-course index, automatically classifying neurons into persistent and transient types. Based on the light response intensity and signal-to-noise ratio, it automatically identifies neurons with light responses. It intelligently obtains light-evoked response curves through a fitting algorithm and extracts visual features for functional subtype classification. It is also used to analyze different light response stimulus intensities and quickly calculate the photosensitivity value of neurons through curve fitting.

[0030] The color stimulus analysis submodule is used to perform color-coded feature analysis on the non-repetitive action potential sequence information of a single type of neuron output by the action potential separation module, identify the spectral information and order of each random sequence of color stimulus, analyze and visualize the histogram of the neuron's response before and after stimulation under each color spectrum stimulus through parallel computing, automatically analyze the visual functional response characteristics of the neuron under each color, obtain the color response feature curve through curve fitting algorithm, and evaluate its color selection characteristics through light response indicators such as average light response intensity and maximum light response intensity.

[0031] The Bar stimulation analysis submodule is used to analyze the direction selectivity and directional characteristics of the non-repetitive action potential sequence information of a single type of neuron output by the action potential separation module. It intelligently identifies the information of randomly moving direction sequences, duration segments, and blank stimuli, and analyzes the response histograms of multiple neurons before and after stimulation in each direction of motion through parallel computation. Then, it analyzes the response curves of neurons in each direction through curve fitting, and intelligently analyzes the optimal direction information of neurons through the maximum light response intensity and average light response intensity. Then, it analyzes the light response characteristic curves of neurons in each direction from 0 to 360 degrees through interpolation fitting algorithm, thereby calculating the direction selectivity and directional evaluation index of neurons. The average light response intensity is calculated by the total average value, half-saturation value, and automatically finding the area of ​​the response curve. The light response characteristic curves of neurons in each direction of motion are intelligently visualized, and the optimal stimulation curve and the response curve and Tunning curve under blank stimulation are automatically presented.

[0032] The Chirp stimulation analysis submodule is used to analyze the non-repetitive action potential sequence information of the single type of neuron under white light flickering stimulation, sinusoidal frequency stimulation, and sinusoidal contrast stimulation, intelligently identify the onset time information of flickering, frequency and contrast stimulation, analyze visual function features through parallel computing and curve fitting, and intelligently visualize the light response feature curves of multi-neurons.

[0033] The spatiotemporal characteristic analysis submodule is used to analyze the spatiotemporal receptive field characteristics of neurons from the non-repetitive action potential sequence information of the single type of neuron. It calculates the peak trigger average sequence by using the visual stimulus intensity and the number of action potentials fired in each frame, gradually increases the detection window size and adopts a strategy that does not directly locate the neuron cell body, so as to accurately and quickly analyze the spatial receptive field and temporal response characteristics. Through threshold evaluation, it intelligently screens out effective neurons with spatiotemporal characteristics and intelligently visualizes them as the spatial response characteristics of multiple neurons as a group and the temporal characteristic curve of a single neuron.

[0034] The ipRGCs characteristic analysis submodule is used to analyze the light response characteristics of autonomous photosensitive ganglion cells from the non-repetitive action potential sequence information of the single type of neuron. It calculates the light response index at high speed and visualizes the light response characteristic curve through parallel computing and curve fitting. It automatically identifies autonomous photosensitive ganglion cells from multiple neurons rather than other neurons through threshold evaluation.

[0035] Furthermore, the integrated statistics module includes a database construction submodule and a statistical analysis submodule. The database construction submodule is used to process the visual function analysis data output by the function analysis module. Through memory mapping, it reads and parses the visual feature curves in the big data neural decoding information of multiple neurons in all data. By screening neurons with light response characteristics and performing big data merging, hierarchical clustering information of neuron functional subtype classification is obtained through hierarchical clustering.

[0036] Furthermore, the integrated statistics module also includes a statistical analysis submodule, with the output of the database construction submodule serving as the input of the statistical analysis submodule;

[0037] The statistical analysis submodule is used to process the hierarchical clustering information output by the database construction submodule, import multiple visual function analysis data, intelligently select statistical analysis methods according to the number of data groups, and perform functional assessments of all neurons within the same neural tissue and all neurons between different tissues. Based on the cell function subtype, it intelligently assesses whether there are statistical differences between groups within each statistical indicator of each subtype of neurons, and draws a bar chart of the differences, thereby achieving accurate assessment of the functional differences of each subtype of neurons and their disease models.

[0038] The beneficial effects of this invention are as follows:

[0039] This invention designs a big data visual function intelligent analysis system comprising a cascaded action potential extraction module, an action potential separation module, a functional analysis module, and an integrated statistics module. It can extract, classify, identify, analyze, and statistically analyze visually evoked action potentials from tens of thousands of neurons within high-throughput electrophysiological data with more than 4,000 channels or even tens of thousands of channels, meeting the requirements for accurate evaluation in applications such as visual neural coding mechanisms, functional classification maps, and disease and treatment models. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the intelligent analysis method for big data visual functions according to the present invention.

[0041] Figure 2 This is a schematic diagram of the composition of the big data visual function intelligent analysis system of the present invention;

[0042] Figure 3 This is a schematic diagram of the action potential detection submodule;

[0043] Figure 4 This is an effective channel rapid detection submodule;

[0044] Figure 5 This is a schematic diagram of a multi-data integration submodule;

[0045] Figure 6 A schematic diagram of the high-speed classification submodule for big data action potentials;

[0046] Figure 7 A schematic diagram of the integrated action potential separation module;

[0047] Figure 8 A schematic diagram of the action potential information recognition submodule;

[0048] Figure 9 This is a schematic diagram of the flickering stimulus analysis submodule;

[0049] Figure 10 This is a schematic diagram of the color stimulus analysis submodule;

[0050] Figure 11 This is a schematic diagram of the Bar stimulus analysis submodule;

[0051] Figure 12 This is a schematic diagram of the Chirp stimulus analysis submodule;

[0052] Figure 13 This is a schematic diagram of the spatiotemporal characteristic analysis submodule;

[0053] Figure 14 A schematic diagram of the ipRGCs feature analysis submodule;

[0054] Figure 15 Diagram illustrating the construction of sub-modules for the database;

[0055] Figure 16 This is a schematic diagram of the statistical analysis submodule.

[0056] Figure 17 This is a schematic diagram of large-scale electrophysiological data of retinal neurons. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] Example 1

[0059] like Figure 1 As shown, a big data visual function intelligent analysis method includes the following steps:

[0060] We acquire and filter effective recording channels from high-throughput electrophysiological big data, extract action potential sequence information, and parse visual stimulus information based on the duration information of simulated input signals.

[0061] Action potential sequence information from multiple electrophysiological data is integrated, and non-visual stimulus time periods are identified and removed based on visual stimulus information to obtain a single integrated electrophysiological big data. The single integrated electrophysiological big data is classified through principal component analysis and multiple clustering methods to obtain action potential sequence information of a single neuron under different visual stimuli. Repeatedly collected and non-target neuronal signals are intelligently identified and automatically removed from the action potential sequence information of a single neuron to obtain non-repetitive action potential sequence information of a single type of neuron.

[0062] Visual features of non-repetitive action potential sequences of a single type of neuron are analyzed, and spatiotemporal visual function analysis is performed to obtain analytical data of that type of neuron, including visual function indicators and feature curves.

[0063] Integrate and statistically analyze the parsed data of all types of neurons, and output the analysis results.

[0064] Example 2

[0065] like Figure 2 As shown, a big data visual function intelligent analysis system includes a cascaded action potential extraction module, an action potential separation module, a function analysis module, and an integration and statistics module.

[0066] The action potential extraction module is used to screen effective recording channels from high-throughput electrophysiological data, extract action potential sequence information, and parse visual stimulus information based on the duration difference information of the simulated input signal, and input it into the action potential separation module.

[0067] The action potential separation module receives and integrates action potential sequence information from multiple electrophysiological data. Based on visual stimulus information, it identifies and removes non-visual stimulus time periods to obtain a single integrated electrophysiological big data. The single integrated electrophysiological big data is classified using principal component analysis and multiple clustering methods to obtain action potential sequence information of a single neuron integrated under multiple visual stimuli or different experimental conditions. The integrated action potential sequence information is then automatically separated into action potential sequence information for a single visual stimulus type. The module intelligently identifies and automatically removes repetitive and non-target neuronal signals from the action potential sequence information of a single neuron, and outputs the non-repetitive action potential sequence information of a single type of neuron to the functional analysis module.

[0068] The functional parsing module is used to receive and parse the visual features of the action potential sequence information of a single type of non-repetitive neuron, perform visual spatiotemporal functional analysis, and obtain and output the parsing data of this type of neuron, including visual function indicators and feature curves, to the integrated statistics module.

[0069] The integrated statistics module is used to receive, integrate, and statistically analyze the parsed data of all types of neurons, establish a visual function database, and visualize the parsed results.

[0070] In this embodiment, the high-throughput electrophysiological data specifically refers to electrophysiological data with more than 4000 channels.

[0071] In one specific embodiment, the visual stimulus information includes visual stimulus type, stimulus start and end time, key time points, the start and end times of light application and removal in flashing stimuli and their stimulation time, the stimulation time of each frame of white noise, the direction and start time of each random run of bar stimuli, and the spectral information and order of each random sequence of color stimuli, etc.

[0072] like Figure 3 As shown, in one specific embodiment, the action potential detection submodule is mainly responsible for the high-speed detection and extraction of action potential sequences contained in each recording channel of high-throughput (more than 4000 recording channels) and large-scale electrophysiological data, as well as the intelligent analysis of visual stimulus type information. Given the extremely large data capacity of neural signals acquired using high-throughput technology (up to 10Gb per minute), and the combined use of various types of visual stimuli and pharmacological detection, the entire recording time will exceed 4 hours. This also results in an extremely large storage capacity for electrophysiological data, making successful analysis of this data impossible using conventional analysis methods and commercial software.

[0073] This module primarily employs strategies such as parallel computing and intelligent clustering of recording channels, utilizing a controllable number of CPU cores to rapidly read and analyze the action potential information contained in each recording channel within the raw high-throughput data. The data acquired from each recording channel contains electrophysiological signals with a bandwidth of 1–5000 Hz, mainly including low-frequency field potentials, high-frequency action potentials, and background noise.

[0074] Therefore, bandpass filtering and threshold detection are selected to effectively extract the encoded information stored in action potentials from the raw electrophysiological data. The extraction strategy is as follows: First, bandpass filtering is performed within a specific frequency range using controllable low-frequency and high-frequency thresholds to obtain filtered data containing action potential characteristics. Then, a controllable threshold standard (the standard deviation multiple of baseline noise) is selected to extract the action potential information contained therein. Finally, outliers in the action potential amplitude distribution characteristics are analyzed to filter and remove noise waveforms contained in the action potential sequence. Given the large number of recording channels in high-throughput data, it is necessary to further efficiently filter out effective recording channels in practical applications. Therefore, this module automatically removes invalid recording channels by setting a detection standard for the action potential emission frequency threshold. The raw electrophysiological data contains simulated input signals with visual stimuli. This module can intelligently parse various visual stimulus types (such as WhiteLight, Bars, Chirp, WhiteNoise, ColorFlash, ipRGCs, etc.), stimulus start and end times, key time points, the start and end times of light application and removal in flashing stimuli and their stimulation times, the stimulation time of each frame of WhiteNoise, the direction of each random movement of Bars, and the spectral information and order of each random sequence of color stimuli, etc. This will ensure the accurate parsing of various visual stimulus codes in the future.

[0075] To visualize and evaluate the intensity of photoevoked responses in each recording channel of high-throughput electrophysiological data, this module can present the results of group channel responses based on the amplitude and frequency of action potentials extracted from each channel. It can also classify and filter the matching recording channel names based on the magnitude of action potential amplitude and automatically view the original filtered photoevoked responses and action potential sequence information. Furthermore, it allows manual adjustment of the detection threshold for individual channels to re-extract and analyze action potentials. In addition, large-scale electrophysiological studies of the same neural tissue often increase with the number of visual stimulus types and pharmacological needs. The extraction of action potential information from each recording channel in each dataset should have the same detection threshold, and invalid recording channels should be eliminated to ensure the smooth progress of subsequent efficient data analysis. Therefore, this module can use the analyzed data of the selected valid recording channels as a template to quickly and effectively extract action potential sequence information within the same recording channel at an equivalent detection threshold level.

[0076] like Figure 4As shown, the effective channel rapid detection submodule is mainly responsible for quickly filtering effective recording channels from tens of thousands of recording channels, thereby enabling high-speed and rapid analysis of multiple large-scale electrophysiological data under multiple visual stimulation patterns in the same neural tissue. Given the numerous and complex recording channels in high-throughput big data, it is necessary to further filter out efficient recording channels with photoreactivity to reduce data storage and improve computing speed. This module, through memory mapping technology, successfully and rapidly accesses electrophysiological data of recording channels with specified names on the hard drive without writing it into memory all at once, solving the problem of data being too large to be successfully imported into the analysis workstation. This allows ordinary computers to filter out efficient recording channels. Furthermore, it performs hierarchical integration and filtering based on the magnitude of action potential amplitude, adding functions such as channel selection, time period selection, and visual stimulus signal simulation. Multiple users can quickly filter out efficient recording channels with photoreactivity, significantly improving the efficiency of visual encoding analysis. It can also intelligently and rapidly generate image format information for each recording channel, facilitating offline filtering and calibration.

[0077] In one specific embodiment, the action potential separation module specifically includes a multi-data integration submodule, a big data action potential high-speed classification submodule, an integrated action potential separation submodule, and an action potential information recognition submodule;

[0078] like Figure 5 As shown, the multi-data integration submodule is mainly responsible for removing action potential sequence information from non-visual stimulus time periods, and then intelligently merging multiple electrophysiological big data according to certain time rules. This will significantly reduce the storage volume of big data and improve the efficiency of action potential classification. The parsing of neuronal visual information encoding mainly relies on data from the time period of visual stimulus delivery. Non-visual stimulus time (such as dark adaptation time, waiting time between different stages of stimulus, etc.) contains almost no visual encoding information and increases the data storage volume by at least twice, bringing unnecessary and ineffective computation to subsequent action potential classification and functional analysis. Therefore, this module automatically identifies and removes data from non-stimulation time periods in action potential information for multiple electrophysiological big data of various visual stimulus types or pharmacological studies, minimizing the data size it occupies; then, it automatically calibrates and adds multiple data according to the data time interval, and quickly obtains a single electrophysiological big data that is re-integrated from the action potential time series through parallel computing, containing the start time and recording information of each data.

[0079] like Figure 6As shown, the high-speed action potential classification submodule for big data is mainly responsible for intelligently, quickly, efficiently, and accurately classifying the raw action potential sequences to obtain the action potential firing characteristics of individual neurons. Since a single electrode often simultaneously collects integrated features of action potential firing from multiple neurons, the action potential sequences collected by each recording channel in electrophysiological big data are integrated information from multiple neurons. However, further separation is needed to study the visual function coding features of a single neuron. This module uses strategies such as principal component analysis (PCA) and clustering to classify the comprehensive information emitted by multiple neurons in each electrode into individual neurons. The action potential classification principle is as follows: PCA is used to reduce the dimensionality of the high-dimensional original action potential sequence information contained in each electrode, transforming the action potential classification into a two-dimensional point clustering problem. Then, two clustering methods are used in combination: the precise Valley method and the high-speed K-means method. The integrated waveform of the action potential sequence is clustered into point sets based on differences in amplitude, time, and other characteristics, achieving both accuracy and speed in neuronal action potential classification. Finally, based on the screening criterion that the absolute refractory period (peak potential interval) of the neuron is greater than 1 millisecond, noise information that deviates far from the average waveform in each subclass is automatically removed, thus accurately obtaining the action potential encoding information of individual neurons. Given the randomness of the results of each clustering method, this module allows setting the number of calculations for each clustering method, ultimately evaluating their Euclidean distance and retaining the clustering results with the greatest differences, thereby significantly improving the efficiency of subsequent result calibration. To reduce the inefficiency of manual operation, this module incorporates an intelligent action potential waveform comparison function. Through progressively increasing clustering parameters and threshold criteria such as matching maximum correlation, shortest matching distance, and lowest noise level, it automatically selects the best matching result from multiple clustering calculations. This allows for the intelligent and precise one-to-one correspondence between subclass neurons in each recording channel within multiple electrophysiological datasets induced by different visual stimuli or conditions in the same neural tissue. Leveraging the large-scale nature of high-throughput electrophysiology data, this module employs memory mapping and parallel computing technologies. By utilizing a controllable number of CPU cores, it intelligently and rapidly parses action potential sequence information from specified channels within large-scale hard disk datasets, overcoming the memory limitations faced by large-scale neuroscience computing. To improve the efficiency of action potential classification, this module allows multiple users to quickly calibrate the accuracy of classification results for each recording channel and select appropriate clustering methods. Users can also reselect clustering parameters for re-analysis and manual clustering to improve classification efficiency. Furthermore, it allows for the import of inaccurately classified recording channel results for secondary checks and manual calibration of action potential classification results. In addition, this module features a fast channel switching function and clear visualization of action potential information, enabling high-speed classification and analysis of action potentials in thousands of recording channels.In summary, this module addresses the limitations of insufficient memory and prolonged processing time in big data neuroscience computing, enabling multi-core, high-speed, and efficient analysis of action potentials in various recording channels of electrophysiological big data for classification.

[0080] like Figure 7 As shown, the integrated action potential separation submodule is mainly responsible for automatically separating the classified action potential information based on the characteristics of the data before integration, thereby obtaining time-series calibrated classification data under various visual stimuli or pharmacological tests. After the integrated electrophysiological big data completes the action potential classification, it can separate and automatically calibrate the action potential time series one by one according to parameters such as the data name before integration, stimulus start time, stimulus end time, and sampling frequency of each data, thereby obtaining electrophysiological data with accurate action potential classification information at different sampling frequencies under various visual stimuli or pharmacological tests. Ultimately, it achieves synchronous and efficient analysis of action potential separation in all recording channels of multiple electrophysiological big data.

[0081] like Figure 8As shown, the action potential information recognition submodule is mainly responsible for identifying the source of signals from the cell body, axon, or dendrites of visual neurons and removing repetitive signals from the same neuron. It accurately identifies and removes signals from cells without long processes that exhibit action potential firing characteristics, ultimately retaining only non-repetitive action potential sequence information from a single type of neuron. Based on the characteristics of high-throughput electrophysiological recording, multiple recording electrodes can simultaneously acquire action potential firing characteristics from the cell body, axon, or dendrites of a single neuron (such as a retinal ganglion cell). Therefore, the identification and deduplication of action potential signals are particularly important. Action potentials fired by the neuronal cell body begin negatively and ultimately form a biphasic waveform, while signals generated by its axon and dendrites begin positively and ultimately form a triphasic waveform. These three waveforms differ significantly. This module accurately determines neuronal cell body signals and non-cell body signals by detecting the amplitude ratio of the first wavelet in the triphasic waveform and filters out extremely low-amplitude noise signals. To eliminate repetitive neuronal signals acquired by multi-electrode datasets, this module utilizes correlation analysis and parallel computing to rapidly calculate the pairwise correlations between all subclass neuronal signals across thousands of recording channels. By using a correlation ratio threshold, it quickly analyzes neural signals from multiple locations within the same neuron, then uses algorithms to identify the neural signal with the optimal amplitude and remove other repetitive signals. Ultimately, it obtains the non-repetitive action potential encoding information for a single neuron. Given that a very small proportion of amygdalae (cells without long processes) with action potential firing characteristics exist in retinal tissue, this module identifies amygdalae signals by evaluating the angular and trend relationships between neuronal cell bodies, axons, and dendrites, automatically removing them to avoid interfering with the evaluation of visual function encoding features of retinal ganglion cells. Furthermore, this module can intelligently view the signal distribution, correlation relationships, matrix distribution, and waveform overlap of various neuronal types, and can quickly remove other repetitive signals and amygdalae signals from different visual stimulus types or pharmacological test data of the same neural tissue.

[0082] In one specific embodiment, the functional parsing module specifically includes a blinking stimulus analysis submodule, a color stimulus analysis submodule, a Bar stimulus analysis submodule, a Chirp stimulus analysis submodule, a spatiotemporal characteristic analysis submodule, and an ipRGCs characteristic analysis submodule;

[0083] like Figure 9As shown, the flickering stimulation analysis submodule is mainly responsible for the comprehensive evaluation of visual function and the calculation of light sensitivity of neurons under white light flickering stimulation. To accurately assess the strength of neuronal visual function and its functional type identification, this module employs parallel computing and pre- and post-stimulus response histograms to rapidly analyze the light response characteristics of tens of thousands of neurons. Multiple fitting algorithms are used to plot their light response curves. The principle is as follows: by plotting pre- and post-stimulus response histograms, the relationship between the frequency of action potential firing by neurons and external visual stimuli within a specific time period is quantitatively evaluated, such as maximum light response intensity, average light response intensity, delay time, and signal-to-noise ratio. This allows for accurate assessment of neuronal cell types and their differences in visual function. By calibrating the light response delay time, the start and end times of the visual response can be accurately identified, thereby precisely calculating various indicators of the light response. Based on the different response characteristics of neurons to light-induced and light-withdrawn stimuli, neurons can be automatically subdivided into light-induced, light-induced-withdrawn, and light-withdrawn types by calculating the dominant response index. Based on the different response time-course characteristics of neurons to light stimuli, neurons can be automatically subdivided into two main categories: persistent and transient types by calculating the response time-course index. Various fitting algorithms are used to accurately obtain light-evoked response curves, and then their visual features are extracted to systematically analyze the classification of neuronal functional subtypes. To accurately analyze the photosensitivity of neurons, this module can automatically analyze the intensity of their photoreactive stimulation and accurately obtain the photosensitivity values ​​of each neuron through curve fitting, enabling the assessment of photosensitivity in different disease models or different types of neurons. Furthermore, to efficiently filter out neuronal information with photoreactive characteristics, this module intelligently filters out effective neurons based on photoreactive intensity, signal-to-noise ratio, and statistical differences. It also allows for intelligent viewing of information such as the functional classification type, photoreactive curve, and photosensitivity curve of each neuron.

[0084] like Figure 10 As shown, the color stimulus analysis submodule is primarily responsible for intelligently parsing the color coding features of neurons and acquiring their color selection characteristics. To accurately assess the selectivity of neurons to different spectral visual stimuli, this module can intelligently identify their spectral information (such as green 530nm, blue 430nm, blue-green 490nm, and yellow 565nm, etc.) and, through parallel computing methods, rapidly and intelligently plot the pre- and post-stimulation response histograms of tens of thousands of neurons under various color spectral stimuli. Furthermore, it evaluates their color selection characteristics using light response indicators such as average light response intensity and maximum light response intensity. Simultaneously, this module intelligently acquires color response feature curves through various fitting algorithms, thereby enabling color classification research of neurons. In addition, this module provides intelligent visualization of the color feature information of each neuron, enabling accurate acquisition of information such as color selectivity, functional subtype classification, and signal-to-noise ratio.

[0085] like Figure 11As shown, the Bar stimulation analysis submodule is primarily responsible for intelligently analyzing the directional selectivity and directivity characteristics of neurons. Based on the inconsistent tendency of neurons to move in different directions, this module can intelligently identify information such as the direction, duration, and blank stimuli of random movement sequences. Through parallel computing, it can rapidly generate histograms of the responses of tens of thousands of neurons before and after stimulation in each movement direction. Then, through curve fitting, it can automatically analyze the response curves in each direction and intelligently analyze the optimal directional information of neurons using the maximum and average light response intensities. The average light response intensity can be intelligently evaluated using algorithms such as the total average value, half-saturation value, and automatic area under the response curve. Then, through interpolation fitting algorithms, it analyzes the light response characteristic curves of neurons in each direction from 0 to 360 degrees, thereby accurately calculating the directional selectivity and directivity evaluation indicators of neurons. Furthermore, this module can automatically calibrate the light response start time of neurons in each movement direction. Through parallel computing, singular value decomposition algorithms, and other algorithms, it accurately analyzes the temporal and spatial characteristics of neurons in the direction of movement, thus enabling the classification study of neurons in the direction of movement.

[0086] like Figure 12 As shown, the Chirp stimulation analysis submodule is primarily responsible for analyzing the light response characteristic curves of neurons under white light flickering stimulation, sinusoidally varying frequency stimulation, and sinusoidally varying contrast stimulation. To further evaluate the differences in the encoding characteristics of visual neurons under frequency and contrast visual stimuli, this module can intelligently identify the onset time information of flickering, frequency, and contrast stimuli under Chirp stimulation. Through algorithms such as parallel computing, curve fitting, and pre- and post-stimulation response histogram plotting, it accurately and quickly analyzes the light response characteristic curves of tens of thousands of neurons and achieves functional subtype classification based on frequency and contrast characteristics. Furthermore, the intelligent visualization interface allows for effective viewing of the light response characteristic curves of each neuron and their differences.

[0087] like Figure 13As shown, the spatiotemporal characteristic analysis submodule is mainly responsible for batch intelligent analysis of the spatiotemporal receptive field characteristics of neurons. Based on the differences in the size and temporal characteristics of the spatial receptive field of each neuron, this module can perform high-speed analysis of the spatiotemporal characteristics of tens of thousands of neurons in a single or two simultaneous WhiteNoise electrophysiological datasets through parallel computing. The principle is mainly as follows: by importing WhiteNoise visual stimulus information, it can intelligently identify the brightness values ​​and visual stimulus parameters that vary normally in each frame and achieve rapid analysis; alternatively, it can achieve slow analysis by intelligently calculating stimulus parameters such as brightness values ​​without importing visual stimulus information. It analyzes the peak trigger average sequence by analyzing the visual stimulus intensity and action potential firing number in each frame, and then uses strategies such as gradually increasing the detection window size and not directly locating the neuron cell body to efficiently analyze the spatial receptive field and temporal response characteristics of a high proportion of neurons. In order to efficiently screen out neurons with effective spatiotemporal characteristics and their group spatial characteristics from tens of thousands of neurons, this module automatically and quickly screens out effective neurons by evaluating the spatial and temporal characteristics of the receptive field and successfully draws the group characteristics of the spatiotemporal receptive fields of thousands of neurons. In addition, this module can intelligently view the spatiotemporal receptive field characteristics of each neuron and the differences in their characteristic curves, and can realize the analysis of the spatiotemporal receptive field characteristic classification map of neurons;

[0088] like Figure 14 As shown, the ipRGCs characteristic analysis submodule mainly focuses on rapidly screening and analyzing the light response characteristics of a low proportion of autonomous photoreceptor ganglion cells (ipRGCs) within the retina. The retina also contains autonomous photoreceptor ganglion cells that can directly sense light without receiving signal input from photoreceptor cells. The visual function encoding characteristics of this type of neuron differ from those of conventional ganglion cells, exhibiting greater light response delay and duration, but its proportion is less than 0.2% of the total number of retinal ganglion cells. Therefore, research on autonomous photoreceptor ganglion cells first requires screening this type of neuron from tens of thousands of neurons to effectively assess its functional encoding characteristics. This module uses parallel computing, curve fitting, and pre- and post-stimulation response histogram plotting to rapidly analyze the light response characteristics of tens of thousands of neurons; then, based on the differences in their light response characteristics, it intelligently screens out a low proportion of autonomous photoreceptor ganglion cells and visualizes their light response characteristic curves and functional evaluation indicators.

[0089] In one specific embodiment, the integrated statistics module includes a database construction submodule and a statistical analysis submodule;

[0090] like Figure 15As shown, the database construction submodule is primarily responsible for integrating the analytical data from various visual stimuli or pharmacological studies of all neural tissues and drawing their clustering maps. Since the data from various types of visual functions or pharmacological analyses of each neural tissue exceeds 10GB, ordinary algorithms cannot systematically integrate multiple neural datasets. Therefore, this module employs memory mapping technology to intelligently identify the analytical data and information from each visual stimulus or pharmacological study within each neural tissue, gradually adding new data without needing to load all stored data, thus solving the problem of database creation failure due to insufficient memory. To clearly visualize the classification results of neuronal visual function subtypes, this module uses algorithms such as parallel computing, interpolation fitting, singular value decomposition, and normalization to rapidly read and analyze the visual feature curves from the large-scale neural decoding information of tens of thousands of neurons in all data. It automatically filters neurons with obvious light response characteristics and performs rapid large-scale merging, then obtains hierarchical clustering information for the functional subtype classification of tens of thousands of neurons through hierarchical clustering. Furthermore, this module can also export large datasets in Mat format, compatible with the application requirements of other software, such as Python and R.

[0091] In one specific embodiment, the integrated statistics module further includes a statistical analysis submodule, and the output of the database construction submodule serves as the input of the statistical analysis submodule;

[0092] like Figure 16 As shown, the statistical analysis submodule is mainly responsible for statistically analyzing and evaluating the differences in various visual function indicators of different types of neurons. To accurately assess the differences in visual function among different neuronal subtypes, this module can import multiple results from the aforementioned visual function analysis. It intelligently selects statistical analysis methods based on the number of data groups, ensuring compatibility with functional assessments of all neurons within the same neural tissue and across different tissues. Based on the cell function subtype, it intelligently assesses whether there are statistical differences between groups within each statistical indicator of each neuronal subtype, and plots the results in a bar chart, thus achieving accurate assessment of the functional differences among neuronal subtypes and their disease models.

[0093] In this embodiment, the present invention has many irreplaceable advantages, specifically as follows: ① Given that conventional low-throughput recording channel extraction and classification techniques for action potentials no longer support the parsing of high-throughput, large-scale electrophysiological information, the present invention employs parallel computing, memory mapping, machine learning, and other big data neuroscience computing methods to accurately identify and extract action potential sequence information of tens of thousands of neurons from visual electrophysiological data with recording channels of up to 4000 or even tens of thousands, or data collection and storage capacities of up to 1Tb or more; Unlike the singularity of conventional action potential classification methods, the present invention also employs two clustering methods, Valley and K-Means, to intelligently identify the optimal classification parameters, balancing the accuracy and speed of action potential classification, and efficiently achieving the classification of action potentials in tens of thousands of recording channels. ② Due to the limitations of conventional techniques, such as slow action potential parsing speed, memory overload leading to parsing failure, and low parsing efficiency, they are no longer suitable for big data neuroscience computing. This invention employs novel algorithms such as controllable multi-core computing, intelligent screening of effective recording channels and batch template application of multiple data, intelligent removal of invalid action potential time series, integration of multiple data for single action potential classification, mapping of big data to the hard disk for stepwise extraction and analysis, and automatic classification of multiple data using machine learning of classified action potential features. These algorithms can quickly and efficiently parse action potential sequences and their classification information of tens of thousands of neurons from tens of thousands of recording channels in electrophysiological big data. ③ Multi-electrode recording technology can acquire repetitive neuronal signals (mainly including cell body signals and non-cell body signals), and can also acquire a very low proportion of signals without long-processed cells that have action potential firing characteristics. Currently, there is no commercial software that can intelligently identify neuronal action potential signals and automatically remove signals without long-processed cells. This invention uses parallel computing to quickly analyze the correlation and waveform feature differences between action potential sequences of tens of thousands of neurons, accurately identify neuronal cell body signals and non-cell body signals, and intelligently remove repetitive signals based on waveform differences. It can also efficiently identify and intelligently remove action potential signals without long-processed cells. ④ The extraction and classification of action potential sequences require comprehensive analysis of visual stimuli to achieve successful parsing of visual information functional encoding. Currently, no software has achieved an effective and systematic combination of the two. This invention can intelligently and quickly complete the extraction, classification, identification, parsing, and statistical analysis of neuronal action potential sequence information, achieving a systematic combination of the two. That is, firstly, it completes the high-speed parsing of neuronal action potential sequence extraction and classification, and then, based on the differences in visual signals, it can intelligently and accurately identify various visual stimulus types, start and end times, color information of randomized sequences, random motion direction information, visual stimulus pattern information of each frame, and other visual parameters, effectively parsing the neural encoding features of neurons under various types of visual stimuli.⑤ This invention provides comprehensive intelligent and precise visual function analysis software and database construction software. It can intelligently screen effective neurons from tens of thousands of neurons at high speed and perform statistical evaluation of functional characteristics. It can also automatically extract various visual feature curves and construct a visual electrophysiological database, thereby realizing research such as neuronal big data-based functional subtype classification and disease model evaluation. ⑥ This invention has high compatibility and a wide range of applications. In addition to its application in visual big data electrophysiological research, it is also perfectly compatible with other neurophysiological processes that generate action potential signals, such as low-throughput multi-electrode recording, patch-clamp recording, and in vivo brain region neural signal recording.

[0094] Example 3

[0095] More specifically, the big data visual function intelligent analysis system of this invention has been successfully applied in the State Key Laboratory of Ophthalmology. Combined with the ultra-high throughput retinal multi-electrode recording system in this laboratory, it has achieved the following: Figure 17 The high-speed, systematic, and comprehensive analysis of visual big data electrophysiological data shown is an indispensable intelligent analysis system for visual neuroscience research institutes. This system mainly includes an action potential extraction module (action potential detection module, effective channel rapid detection module), an action potential separation module (multi-data integration module, big data action potential high-speed classification module, integrated action potential separation module, action potential information recognition module), a functional analysis module (flicker stimulus analysis module, color stimulus analysis module, Bar stimulus analysis module, Chirp stimulus analysis module, spatiotemporal characteristic analysis module, ipRGCs characteristic analysis module), and an integrated statistical module (database construction module, statistical analysis module). It achieves a combination of action potential sequence encoding and intelligent analysis of visual stimuli, meeting the research needs of retinal ganglion cell visual encoding mechanisms, functional subtype classification, and disease and treatment model evaluation. Details are as follows:

[0096] Action Potential Extraction Module: This module mainly includes an action potential detection module and an effective channel rapid detection module. It is responsible for high-speed detection and extraction of action potential sequence information contained in each channel within a high-throughput recording channel (more than 4,000 or even tens of thousands of channels), and intelligently analyzes the various types of visual stimulus information contained therein (such as stimulus type, stimulus start and end time, color information of random sequence execution, random direction movement information, stimulus start time of each frame, etc.). The retinal multi-electrode recording system in our laboratory (BioCAM DupleX, HD-MEA Stimulo, Switzerland) has 4,096 recording channels, a sampling frequency of 20 kHz per channel, and an electrode spacing of 81 micrometers. It can acquire electrophysiological signals of retinal ganglion cells on a large scale, with a storage capacity of up to 10 Gb / min. This means that conventional electrophysiological analysis systems cannot analyze such large amounts of data. This invention adopts big data neuroscience computing technologies such as parallel computing, intelligent clustered reading of recording channels, and memory mapping. It accurately analyzes the action potential sequence information contained in each recording channel of the big data through steps such as high-speed reading, bandpass filtering, extraction, screening, and filtering. In addition, this module can intelligently identify various types of visual stimuli and their start and end times, random sequence information and other visual parameters, ensuring the accurate parsing of a large number of neuronal visual function codes.

[0097] Action Potential Separation Module: This module mainly includes a multi-data integration module, a high-speed action potential classification module for large-scale data, an integrated action potential separation module, and an action potential information recognition module. It is responsible for high-speed classification and accurate identification, deduplication, and removal of signals from non-long-processed cells within multiple electrophysiological big data recording channels, ultimately obtaining action potential sequence information from tens of thousands of retinal ganglion cells. Based on the fact that each recording channel in high-throughput electrophysiology simultaneously acquires action potential signals from multiple ganglion cells, this invention employs big data neuroscience computing technologies such as parallel computing, memory mapping, machine learning, principal component analysis, and cluster analysis. It can intelligently integrate multiple electrophysiological big data using the same detection threshold, different time strategies, and removal of invalid time series, thereby achieving rapid and accurate one-time classification of all data. It can also intelligently match and classify other data using already classified data as templates. This invention balances classification accuracy and efficiency, employing the accurate classification method Valley and the high-speed classification method K-means to simultaneously and rapidly analyze the classification information of action potential sequences within tens of thousands of recording channels. Manual classification can also be performed to improve the classification efficiency of neurons. A single neuron can be acquired by multiple recording electrodes simultaneously, and the retina also contains a very small number of amacrine cells with action potential firing characteristics. This invention uses parallel computing, correlation analysis and other strategies to intelligently identify action potential signals at different locations of the same neuron and automatically remove repetitive signals within the same neuron. It can also intelligently identify action potential information of amacrine cells and automatically remove it.

[0098] Functional Analysis Module: This module mainly includes flicker stimulus analysis, color stimulus analysis, bar stimulus analysis, chirp stimulus analysis, spatiotemporal characteristic analysis, and iPRGCs characteristic analysis modules. It is responsible for accurately and rapidly analyzing the visual functional coding features of tens of thousands of retinal ganglion cells under various visual stimulus types. Based on the inconsistency in the functional coding characteristics of retinal ganglion cells for different visual stimulus types, this invention uses algorithms such as parallel computing, curve fitting, pre- and post-stimulus response histogram plotting, and singular value decomposition to accurately and rapidly analyze the differences in color, frequency, contrast, orientation, spatiotemporal receptive field, and other features of tens of thousands of retinal ganglion cells and evaluate the main classifications of their functional subtypes.

[0099] The integrated statistics module primarily comprises a database construction module and a statistical analysis module. It constructs a database and classification atlas of functional characteristics of tens of thousands of retinal ganglion cells, and performs statistical evaluation of functional differences among various neuronal types. To accurately assess the classification of retinal ganglion cell functional subtypes, this invention utilizes algorithms such as parallel computing, memory mapping, feature extraction, singular value decomposition, and normalization to rapidly extract and visualize visual response feature curves for various types of retinal ganglion cells. Furthermore, hierarchical clustering analysis can be used to construct a functional classification atlas of ganglion cells. In addition, this invention can intelligently analyze the above visual function decoding results, automatically generating bar charts of statistical differences, thus fulfilling research needs such as the evaluation of visual function differences among ganglion cell subtypes and their associated disease models.

Claims

1. A method for intelligent analysis of visual functions in big data, characterized in that: The method includes the following steps: We acquire and filter effective recording channels from high-throughput electrophysiological big data, extract action potential sequence information, and parse visual stimulus information based on the duration information of simulated input signals. Action potential sequence information from multiple electrophysiological data is integrated, and non-visual stimulus time periods are identified and removed based on visual stimulus information to obtain a single integrated electrophysiological big data. The single integrated electrophysiological big data is classified through principal component analysis and multiple clustering methods to obtain action potential sequence information of a single neuron under different visual stimuli. Repeatedly collected and non-target neuronal signals are intelligently identified and automatically removed from the action potential sequence information of a single neuron to obtain non-repetitive action potential sequence information of a single type of neuron. Multiple clustering methods are used for classification. The specific steps include: using principal component analysis to reduce the dimensionality of the high-dimensional original action potential sequence information contained in each electrode, transforming the classification of action potential sequence information into a clustering problem of points in a two-dimensional plane; and using Valley clustering and K-means clustering methods in combination to intelligently cluster multiple action potential point sets of neurons in multiple recording channels based on the differences in amplitude and time characteristics of the integrated waveform of action potential sequence information through parallel computing and memory mapping. Based on the peak potential interval standard, noise waveforms deviating from the average waveform are automatically removed to obtain the action potential sequence information of the individual neuron. It also incorporates an action potential waveform classification and comparison function. Through progressively increasing clustering parameters, matching the maximum correlation, matching the shortest distance, and the lowest noise level threshold, the best matching result is selected from multiple clustering calculations. This allows for the analysis of the one-to-one correspondence between subclass neurons in each recording channel within multiple electrophysiological big data sets induced by different visual stimuli and pharmacological conditions in the same neural tissue, enabling synchronous comparison of the same neuron under different conditions. Based on the number of action potentials after clustering, noise waveforms are automatically identified and parameters are intelligently adjusted to obtain the best neuron classification result. The classification results of neurons in each recording channel are intelligently visualized. Visual features of non-repetitive action potential sequences of a single type of neuron are analyzed, and spatiotemporal visual function analysis is performed to obtain analytical data of that type of neuron, including visual function indicators and feature curves. Integrate and statistically analyze the parsed data of all types of neurons, and output the analysis results.

2. A big data visual function intelligent analysis system, characterized in that: To implement the method as described in claim 1, the system includes a cascaded action potential extraction module, an action potential separation module, a function analysis module, and an integrated statistics module; The action potential extraction module is used to screen effective recording channels from high-throughput electrophysiological data, extract action potential sequence information, and parse visual stimulus information based on the duration difference information of the simulated input signal, and input it into the action potential separation module. The action potential separation module is used to receive and integrate action potential sequence information from multiple electrophysiological data, identify and remove non-visual stimulus time periods based on visual stimulus information, and obtain a single integrated electrophysiological big data. The integrated electrophysiological big data is classified using principal component analysis and multiple clustering methods to obtain action potential sequence information of a single neuron under multiple visual stimuli or different experimental conditions. The integrated action potential sequence information is then automatically separated into action potential sequence information of a single visual stimulus type. The system intelligently identifies and automatically removes repetitive and non-target neuronal signals from the action potential sequence information of a single neuron, and outputs the non-repetitive action potential sequence information of a single type of neuron to the functional analysis module. The functional parsing module is used to receive and parse the visual features of the action potential sequence information of a single type of non-repetitive neuron, perform visual spatiotemporal functional analysis, and obtain and output the parsing data of this type of neuron, including visual function indicators and feature curves, to the integrated statistics module. The integrated statistics module is used to receive, integrate, and statistically analyze the parsed data of all types of neurons, establish a visual function database, and visualize the parsed results.

3. The big data visual function intelligent analysis system according to claim 2, characterized in that: The visual stimulus information includes visual stimulus type, stimulus start and end time, key time points, the start and end times of light application and removal in flashing stimuli and their stimulation time, the stimulation time of each frame of white noise, the direction and start time of each random run of bar stimuli, and the spectral information and order of each random sequence of color stimuli.

4. The big data visual function intelligent analysis system according to claim 2, characterized in that: The action potential extraction module includes an effective channel fast detection submodule and an action potential detection submodule, and the output of the effective channel fast detection submodule serves as the input of the action potential detection submodule; The effective channel fast detection submodule is used to access the electrophysiological data of the specified name recording channel on the hard disk through a memory mapping method; The system generates image format information for each recording channel to facilitate offline screening and calibration; it performs hierarchical integration and screening based on the magnitude of action potential amplitude, and combines the functions of selecting all channels, selecting time periods, and simulating visual stimulus signals to screen out recording channels with effective light responses, thereby reducing the computational resource consumption of invalid information in the subsequent analysis module. The action potential detection submodule is used to automatically eliminate invalid recording channels by parallel computing and clustered reading of the effective photoresponse recording channels, setting a detection standard of action potential emission frequency threshold, thereby obtaining action potential information in the effective recording channels within each original high-throughput data. This can eliminate invalid recording channels, thereby reducing the waste of subsequent computing and storage resources and parsing the action potential information contained in all recording channels within each data. It also extracts action potential sequence information from the original electrophysiological data at high speed by selecting bandpass filtering and threshold detection methods. Based on the duration information of the simulated input signal, it parses the individual visual stimulus information features and automatically outputs the electrophysiological data of a single visual stimulus from the integrated recording of multiple visual stimuli.

5. The big data visual function intelligent analysis system according to claim 4, characterized in that: The method of extracting action potential sequence information from the raw electrophysiological data by selecting bandpass filtering and threshold detection is as follows: First, bandpass filtering is performed within a certain frequency range by setting low-frequency and high-frequency thresholds to obtain filtered data containing action potential characteristics; then, a threshold standard is set to extract action potentials from the filtered data; next, outliers in the action potential amplitude distribution characteristics are analyzed, and noise waveforms contained in the action potential sequence information are filtered and removed based on the outliers to extract valid action potential sequence information; the data characteristics of the action potential sequence contained in a single recording channel after filtering, as well as the extracted action potential waveform and quantity characteristics, are intelligently visualized, and the operation of recalibrating and deleting action potential information in a single channel can be performed again.

6. The big data visual function intelligent analysis system according to claim 5, characterized in that: The action potential separation module specifically includes a multi-data integration submodule, a big data action potential high-speed classification submodule, an integrated action potential separation submodule, and an action potential information recognition submodule. The multi-data integration submodule is used to receive the action potential sequence information of a single visual stimulus output by the action potential extraction module. For multiple electrophysiological big data, it identifies and removes the action potential sequence information of non-stimulation time periods in the action potential sequence information. Then, according to the data time interval setting, it calibrates and adds the multiple electrophysiological big data. Through parallel computing, it obtains the re-integration of the action potential time series to get the single integrated electrophysiological big data, which contains the start time of each data and its recording information. The high-speed classification submodule for big data action potentials is used to classify the action potential sequences of multiple neurons in each electrode according to principal component analysis and clustering methods to obtain the action potential sequence of a single neuron; based on the information features of the same array of neurons under different conditions, it intelligently calculates and automatically matches the action potential sequence information of a single neuron in different data under different conditions. The integrated action potential separation submodule is used to separate and calibrate the action potential time series of the individual neurons according to the data name before integration, stimulation start time, stimulation end time, and sampling frequency of each data, thereby obtaining electrophysiological data with different sampling frequencies and action potential classification information, and finally realizing the synchronization and parsing of action potential separation in all recording channels of the multiple electrophysiological big data. The action potential information recognition submodule is used to identify the electrophysiological data output by the integrated action potential separation submodule, intelligently identify the source of neuronal cell body, axon, or dendrite signals and remove redundant signals repeatedly collected from the same neuron, intelligently identify and remove signals from cells without long processes that have action potential firing characteristics, and retain only non-repetitive action potential sequence information of a single type of neuron.

7. The big data visual function intelligent analysis system according to claim 2, characterized in that: The functional analysis module specifically includes a blinking stimulus analysis submodule, a color stimulus analysis submodule, a Bar stimulus analysis submodule, a Chirp stimulus analysis submodule, a spatiotemporal characteristic analysis submodule, and an ipRGCs characteristic analysis submodule. The flickering stimulus analysis submodule is used to perform white light flickering stimulus analysis on the non-repetitive action potential sequence information of the single type of neuron. Through parallel computation and pre- and post-stimulation response histogram plotting, it evaluates visual functional indicators such as maximum light response intensity, average light response intensity, delay time, and signal-to-noise ratio. Based on the characteristics of light-on and light-off stimuli, it calculates the dominant response index, automatically classifying neurons into light-on, light-on-off, and light-off types. Based on the response time-course characteristics, it calculates the response time-course index, automatically classifying neurons into persistent and transient types. Based on the light response intensity and signal-to-noise ratio, it automatically identifies neurons with light responses. It intelligently obtains light-evoked response curves through a fitting algorithm and extracts visual features for functional subtype classification. It is also used to analyze different light response stimulus intensities and quickly calculate the photosensitivity value of neurons through curve fitting. The color stimulus analysis submodule is used to perform color-coded feature analysis on the non-repetitive action potential sequence information of a single type of neuron output by the action potential separation module, identify the spectral information and order of each random sequence of color stimulus, analyze and visualize the histogram of neuron response before and after stimulation under each color spectrum stimulus through parallel computing, automatically analyze the visual functional response characteristics of neuron under each color, obtain the color response feature curve through curve fitting algorithm, and evaluate its color selection characteristics through the light response index of average light response intensity and maximum light response intensity. The Bar stimulation analysis submodule is used to analyze the direction selectivity and directional characteristics of the non-repetitive action potential sequence information of a single type of neuron output by the action potential separation module. It intelligently identifies the information of randomly moving direction sequences, duration segments, and blank stimuli, and analyzes the response histograms of multiple neurons before and after stimulation in each direction of motion through parallel computation. Then, it analyzes the response curves of neurons in each direction through curve fitting, and intelligently analyzes the optimal direction information of neurons through the maximum light response intensity and average light response intensity. Then, it analyzes the light response characteristic curves of neurons in each direction from 0 to 360 degrees through interpolation fitting algorithm, thereby calculating the direction selectivity and directional evaluation index of neurons. The average light response intensity is calculated by the total average value, half-saturation value, and automatically finding the area of ​​the response curve. The light response characteristic curves of neurons in each direction of motion are intelligently visualized, and the optimal stimulation curve and the response curve and Tunning curve under blank stimulation are automatically presented. The Chirp stimulation analysis submodule is used to analyze the non-repetitive action potential sequence information of the single type of neuron under white light flickering stimulation, sinusoidal frequency stimulation, and sinusoidal contrast stimulation, intelligently identify the onset time information of flickering, frequency and contrast stimulation, analyze visual function features through parallel computing and curve fitting, and intelligently visualize the light response feature curves of multi-neurons. The spatiotemporal characteristic analysis submodule is used to analyze the spatiotemporal receptive field characteristics of neurons from the non-repetitive action potential sequence information of the single type of neuron. It calculates the peak trigger average sequence by using the visual stimulus intensity and the number of action potentials fired in each frame, gradually increases the detection window size and adopts a strategy that does not directly locate the neuron cell body, so as to accurately and quickly analyze the spatial receptive field and temporal response characteristics. Through threshold evaluation, it intelligently screens out effective neurons with spatiotemporal characteristics and intelligently visualizes them as the spatial response characteristics of multiple neurons as a group and the temporal characteristic curve of a single neuron. The ipRGCs characteristic analysis submodule is used to analyze the light response characteristics of autonomous photosensitive ganglion cells from the non-repetitive action potential sequence information of the single type of neuron. It calculates the light response index at high speed and visualizes the light response characteristic curve through parallel computing and curve fitting. It automatically identifies autonomous photosensitive ganglion cells from multiple neurons rather than other neurons through threshold evaluation.

8. The big data visual function intelligent analysis system according to claim 7, characterized in that: The integrated statistics module includes a database construction submodule and a statistical analysis submodule. The database construction submodule is used to process the visual function analysis data output by the function analysis module. Through memory mapping, it reads and parses the visual feature curves in the big data neural decoding information of multiple neurons in all data. By filtering neurons with light response characteristics and merging them in a big data manner, hierarchical clustering information of neuron functional subtype classification is obtained through hierarchical clustering.

9. The big data visual function intelligent analysis system according to claim 8, characterized in that: The integrated statistics module also includes a statistical analysis submodule, and the output of the database construction submodule serves as the input of the statistical analysis submodule; The statistical analysis submodule is used to process the hierarchical clustering information output by the database construction submodule, import multiple visual function analysis data, intelligently select statistical analysis methods according to the number of data groups, and perform functional assessments of all neurons within the same neural tissue and all neurons between different tissues. Based on the cell function subtype, it intelligently assesses whether there are statistical differences between groups within each statistical indicator of each subtype of neurons, and draws a bar chart of the differences, thereby achieving accurate assessment of the functional differences of each subtype of neurons and their disease models.

Images