A method for classifying retinal cells

Abstract

The present invention discloses a method for classifying retinal cells, comprising the following steps: taking an animal retina, adhering a pre-treated biochip to the retina to obtain an adhered retina, performing various light stimulations on the adhered retina, collecting electrical signals, filtering the electrical signals to remove background noise and improve the signal-to-noise ratio, extracting spike potentials, performing statistical analysis to obtain pre-treated data, and classifying and labeling the pre-treated data. By applying light stimulation to the ex vivo retina, the present invention visualizes the responses of ganglion cells to different stimulations, and it can be clearly seen that different ganglion cells have different responses to different stimulations.

Description

Technical Field

[0001] The present invention relates to the field of biotechnology, and in particular to a method for classifying retinal cells. Background Art

[0002] For mammals, vision is the most important perceptual system, which provides 90% of sensory information. Image information is important for daily habitat, hunting and mating. The visual perception of mammals begins in the eyes. Light passes through the cornea, the lens and the vitreous body to reach the retina. The retina is a hierarchical tissue structure, which is divided into the outer nuclear layer, the inner nuclear layer and the ganglion cell layer. The outer nuclear layer is composed of photoreceptor cells, including cones and rods, the inner nuclear layer is composed of bipolar cells, and the ganglion cell layer contains only ganglion cells. The outer lamina is composed of horizontal cells, which separates the outer nuclear layer from the inner nuclear layer, and the inner lamina is composed of amacrine cells, which separates the inner nuclear layer from the ganglion cell layer.

[0003] Photoreceptors are cells in the retina that can convert light signals into neural signals. Rods distinguish between light and darkness and the outline of objects, and have a low threshold for receiving light stimulation. Cones distinguish between color and object details, and have a high threshold for receiving light stimulation. Photoreceptors transmit signals to horizontal cells and bipolar cells. Horizontal cells play a feedback regulatory role on photoreceptors. Bipolar cells continue to transmit signals downward to amacrine cells and ganglion cells. Amacrine cells also play a feedback regulatory role on bipolar cells. After receiving the signal, the ganglion cells continue to transmit it backward, cross-transmit it to the lateral geniculate body, and then transmit it to the visual processing cortex. The direction of information flow of signals in the retina is opposite to the direction of the physiological structure of the retina. Light must pass through ganglion cells and bipolar cells before reaching the photoreceptors.

[0004] The complex network structure in the retina is an essential foundation for supporting visual perception. In order to explore the image encoding method in the retina, Tom Baden's team used a light stimulation paradigm covering bright, color and checkerboard, recorded the response of ganglion cells by calcium imaging, and used clustering algorithms to explore the encoding characteristics of ganglion cells in the mouse retina, revealing that different cells have different responses to different stimuli; Klaudia P. Szatko's team used color stimulation to explore the visual pathway of the mouse retina, indicating that there is color opposition in the color pathway; Takeshi Yoshimatsu's team used color stimulation to explore the color pathway of the zebrafish retina, and used the principal component analysis algorithm to reveal that the retina divides color input into color axis and light and dark axis for processing.

[0005] Specific responses of the retina can be obtained by designing color, brightness, and checkerboard light stimulation paradigms, and they can be classified by simple machine learning algorithms. However, the involved stimulation paradigms are not comprehensive and cannot cover all features in daily images, and using simple machine learning algorithms cannot fully reveal the retinal coding mechanism. Summary of the Invention

[0006] To address the deficiencies of the above technical solutions, the purpose of the present invention is to provide a method for classifying retinal cells.

[0007] The purpose of the present invention is achieved through the following technical solutions.

[0008] A method for classifying retinal cells includes the following steps:

[0009] Step 1: Take the retina of an animal, adhere the preprocessed biochip to the retina to obtain the adhered retina.

[0010] Step 2: Apply light stimulation to the adhered retina, respectively applying local stimulation, global stimulation, flickering stimulation, color stimulation, moving bar stimulation, shape stimulation, and black-and-white checkerboard stimulation. After each application of one of the above stimulations, the retina needs to be kept in the dark for 10 s until the last stimulation is applied. The retina converts the light signal of the light stimulation into an electrical signal.

[0011] Step 3: Collect the electrical signal, filter the electrical signal to remove background noise, improve the signal-to-noise ratio, extract spike potentials, and perform statistical analysis to obtain the preprocessed data.

[0012] Step 4: Classify and label the preprocessed data. Due to different stimulations, the response of ganglion cells to each stimulation is different, and at the same time, the duration and interval of different stimulations are fixed. Therefore, the data contains both time information and spatial information. Classify the data based on this information to classify different types of retinal cells, including the following steps:

[0013] Step 4.1: Remove the 10 s rest time when the retina is kept in the dark, reduce the data dimension to 1 kHz, and perform sparse principal component analysis (sPCA) on the response signals of local stimulation, global stimulation, flickering stimulation, color stimulation, moving bar stimulation, and shape stimulation, and sequentially extract 20, 20, 4, 6, 16, and 18 feature vectors for the response signals of the above stimulations.

[0014] Step 4.2: Perform singular value decomposition on the response matrix for the moving bar stimulus, which is divided into a time-related component and a direction-related component. Use sPCA for feature extraction on the direction-related component to extract 6 eigenvectors. At the same time, perform exponential projection on the direction-related component, extract and calculate the direction index as the direction selectivity, and plot according to the selectivity. Different cells have different direction selectivities, and thus classify retinal ganglion cells accordingly.

[0015] Step 4.3: For the response signal of the black and white checkerboard stimulus, use the spike-triggered average method to calculate the receptive field of retinal ganglion cells. Use sPCA to extract 10 eigenvectors from the receptive field. At the same time, plot the receptive field size. Different retinal ganglion cells have different receptive fields, and thus classify retinal ganglion cells accordingly.

[0016] Step 4.4: Combine the eigenvectors extracted in Steps 4.1, 4.2, and 4.3, and use these 100 eigenvectors for clustering. Adopt GMM clustering and use the expectation-maximization algorithm for calculation. Constrain the covariance matrix of each component to be diagonal, and use the Bayesian Information Criterion (BIC) as an evaluation index. When the number of classes is 40, the BIC is the lowest, indicating the best clustering effect at this time.

[0017] In the above Step 2, the local stimulus is a light stimulus centered on the center of the retina with a diameter of 700 μm. The light intensity changes following a sine wave from 0 to SSS lux, the frequency gradually increases from 1 Hz to 5 Hz, and the stimulation time for each frequency is 10 s.

[0018] In the above Step 2, the global stimulus has a diameter of 2 mm. The light intensity changes following a sine wave from 0 to SSS lux, the frequency gradually increases from 1 Hz to 5 Hz, and the stimulation time for each frequency is 10 s.

[0019] In the above Step 2, the flickering stimulus has a diameter of 3 mm, the light intensity changes following a square wave from 0 to SSS lux, and the frequency is 1 Hz.

[0020] In the above Step 2, the color stimulus has a diameter of 3 mm. The retina is at rest in the dark for 10 s, then stimulated with blue light for 3 s, rested in the dark for 3 s, then stimulated with green light for 3 s, and this cycle repeats 3 times.

[0021] In the above Step 2, the moving bar stimulus has a length of 800 μm and a width of 100 μm. The stimulation directions are divided into 8, namely from top to bottom, from upper right to lower left, from right to left, from lower right to upper left, from bottom to top, from lower left to upper right, from left to right, and from upper left to lower right. The movement path length for each direction is 3 mm.

[0022] In step 2, the shapes of the shape stimuli are triangles, squares, and circles. First, the background is black, and the color of each shape is green. Each shape stimulus lasts for 3 s, followed by a 3-s dark rest. Then, the background is black, and the color of each shape is blue. Each shape stimulus lasts for 3 s, followed by a 3-s dark rest. The retina has a 10-s dark rest. When the background is white, the color of each shape is green, and each shape stimulus lasts for 3 s, followed by a 3-s dark rest. When the background is white and the shape is blue, each shape stimulus lasts for 3 s, and there is a 3-s dark rest between shapes. The retina has a 10-s dark rest. When the background is blue, the color of each shape is green, and each shape stimulus lasts for 3 s, followed by a 3-s dark rest. The retina has a 10-s dark rest. When the background is green, the color of each shape is blue, and each shape stimulus lasts for 3 s, followed by a 3-s dark rest.

[0023] In step 2, the black-and-white checkerboard stimulus consists of 50×50 squares, each square being 50×50 μm in size. The light and dark of each checkerboard change over time and are randomly generated by a Gaussian function, with a duration of 5 min.

[0024] In step 3, AxIS software is used to extract action potentials.

[0025] In step 3, a high-density multi-electrode array instrument is used to collect electrical signals.

[0026] The advantages and beneficial effects of the present invention are as follows:

[0027] 1. By applying light stimuli to the isolated retina, the present invention visualizes the responses of ganglion cells to different stimuli, and it can be clearly seen that different ganglion cells have different responses to different stimuli.

[0028] 2. For the isolated retina of the present invention, through the spike response signals in different stimuli, sparse principal component analysis is performed separately to extract features, and a mixed model classifier is used for clustering. Retinal nerve cell types with good clustering performance can be obtained, indicating that the isolated retina can be classified based on the signals obtained from the stimulation paradigm.

[0029] 3. The isolated retina of the present invention has different response patterns to different stimuli such as different directions, different colors, and different light intensities, indicating that the isolated retina can distinguish stimuli with different spatio-temporal characteristics and has different response signals. Based on the different response signals, retinal nerve cells can be classified. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of local stimulation in Embodiment 1 of the present invention.

[0031] Figure 2Schematic diagram of global stimulation for Embodiment 1 of the present invention.

[0032] Figure 3 Schematic diagram of flickering stimulation for Embodiment 1 of the present invention.

[0033] Figure 4 Schematic diagram of color stimulation for Embodiment 1 of the present invention.

[0034] Figure 5 Schematic diagram of moving bar stimulation for Embodiment 1 of the present invention.

[0035] Figure 6 Schematic diagram of shape stimulation for Embodiment 1 of the present invention.

[0036] Figure 7 Schematic diagram of black and white checkerboard stimulation for Embodiment 1 of the present invention.

[0037] Figure 8 Spike response raster plot for Embodiment 1 of the present invention.

[0038] Figure 9 Relationship diagram between retinal spike response and color for Embodiment 1 of the present invention.

[0039] Figure 10 Influence diagram of light intensity on spikes for Embodiment 1 of the present invention.

[0040] Figures 11 - 13 Relationship diagram between spikes and light exposure time for Embodiment 1 of the present invention.

[0041] Figure 14 Evaluation index diagram of clustering results for Embodiment 1 of the present invention.

[0042] Figure 15 Scatter plot of clustering results for Embodiment 1 of the present invention.

[0043] Figure 16 Response signal diagram of all cells for Embodiment 1 of the present invention.

[0044] Figure 17 Cell direction selectivity response diagram for Embodiment 1 of the present invention.

[0045] Figure 18 On and Off receptive field diagrams of different stimulated cells for Embodiment 1 of the present invention.

[0046] Figure 19 Overall structural schematic diagram of the light stimulation system capable of being combined with multiple electrophysiological devices for Embodiment 2 of the present invention. Detailed implementation manners

[0047] The technical solution of the present invention will be further described below in conjunction with specific embodiments.

[0048] Example 1

[0049] A method for classifying retinal cells, comprising the following steps:

[0050] Step 1, in this example, the retinas of C57 mice over 6 weeks old are taken, and the pre-treated biochip is adhered to the retinas to obtain adhered retinas. Among them, the biochip is pre-treated with polylysine (selecting a 6-well Axion biochip, each well consisting of 59 TiN / SiN planar circular electrodes with a diameter of 30 μm and an electrode center spacing of 200 μm) to promote the adhesion of the retinas to the biochip.

[0051] Step 2, the adhered retinas are subjected to light stimulation. Specifically, the retinas under each well of the Axion biochip are stimulated individually, and local stimulation, global stimulation, flicker stimulation, color stimulation, moving bar stimulation, shape stimulation, and black and white checkerboard stimulation are applied respectively. After each of the above stimulations is applied, the retinas need to be kept in the dark for 10 s until the last stimulation is applied. The retinas convert the optical signals of the light stimulation into electrical signals. Among them, as Figure 1 shown, the local stimulation is a light stimulation centered on the center of the retina with a diameter of 700 μm, the light intensity changes following a sine waveform from 0 to SSS lux, the frequency gradually increases from 1 Hz to 5 Hz, and the stimulation time for each frequency is 10 s; as Figure 2 shown, the global stimulation has a diameter of 2 mm, the light intensity changes following a sine waveform from 0 to SSS lux, the frequency gradually increases from 1 Hz to 5 Hz, and the stimulation time for each frequency is 10 s; as Figure 3 shown, the flicker stimulation has a diameter of 3 mm, the light intensity changes following a square wave from 0 to SSS lux, and the frequency is 1 Hz; as Figure 4 shown, the color stimulation has a diameter of 3 mm, the retinas rest quietly in the dark for 10 s, are stimulated with blue light for 3 s, rest in the dark for 3 s, are stimulated with green light for 3 s, and this cycle repeats 3 times. As Figure 5 shown, the moving bar stimulation has a length of 800 um and a width of 100 um, and the stimulation directions are divided into 8, namely from top to bottom, from top right to bottom left, from right to left, from bottom right to top left, from bottom to top, from bottom left to top right, from left to right, and from top left to bottom right. The movement path length in each direction is 3 mm; as Figure 6As shown, the shapes of the shape stimuli are triangles, squares, and circles. First, the background is black, and the color of each shape is green. Each shape stimulus lasts for 3 s, followed by a 3-s dark rest. Then, the background is black, and the color of each shape is blue. Each shape stimulus lasts for 3 s, followed by a 3-s dark rest. The retina has a 10-s dark rest. When the background is white, the color of each shape is green, and each shape stimulus lasts for 3 s, followed by a 3-s dark rest. When the background is white and the shape is blue, each shape stimulus lasts for 3 s, and there is a 3-s dark rest between shapes. The retina has a 10-s dark rest. When the background is blue, the color of each shape is green, and each shape stimulus lasts for 3 s, followed by a 3-s dark rest. The retina has a 10-s dark rest. When the background is green, the color of each shape is blue, and each shape stimulus lasts for 3 s, followed by a 3-s dark rest; as Figure 7 shown, the black-and-white checkerboard stimulus consists of 50×50 squares, each grid being 50×50 μm in size. The light and dark of each checkerboard change over time and are randomly generated by a Gaussian function, with a duration of 5 min.

[0052] Step 3, as Figure 8 shown, a high-density multi-electrode array instrument (the high-density multi-electrode array instrument is the MAESTRO MEA system produced by Axion Biosystems, USA) is used. This system collects electrical signals with an 8×8 electrode array arranged in a square grid. The Axion Integrated Studio (AxIS) Navigator software is used to monitor and record the electrical signals online, which can record and display the discharge state of retinal cells in real time. The data acquisition status of each electrode is monitored in the form of a window, and it is observed in real time whether the collected electrical signals contain action potentials and bursts, so as to confirm the quality of the recorded data to reflect the state of retinal cells. An adaptive threshold of 6 times the standard deviation is set (a lower threshold will increase false-positive errors, and small noise events will be misidentified as peaks, while a higher threshold may fail to detect small-amplitude action potentials). The electrical signals are filtered using a 300 - 3000 Hz bandpass filter to remove background noise and improve the signal-to-noise ratio. The AxIS software is used to extract action potentials (spikes) for statistical analysis to obtain preprocessed data.

[0053] As can be seen from Figure 8 this, for the raster plot of the action potential responses recorded by all channels of the high-density multi-electrode array instrument, the complete stimulus paradigm is applied to the ex vivo retina, and the responses of all electrode channels on the high-density multi-electrode array instrument are recorded. The stimulus paradigm includes all spatio-temporal characteristics of the stimulus paradigms described in Step 2. Different retinal ganglion cells in the retina have different coding patterns for different characteristics, so they can transmit different information for the brain to perceive vision.

[0054] Step 4. Use a GMM classifier to classify and label the preprocessed data. Due to different stimuli, the responses of ganglion cells to each stimulus vary. At the same time, the durations and intervals of different stimuli are fixed. Therefore, the data contains both time information and spatial information. Based on this information, the data is classified to distinguish different types of retinal cells.

[0055] Step 4.1. Remove the 10 - second rest time of the retina in the dark. Reduce the data dimension to 1 kHz. Perform sparse principal component analysis (sPCA) on the response signals for local stimulus, global stimulus, flash stimulus, color stimulus, moving bar stimulus, and shape stimulus respectively. Extract 20, 20, 4, 6, 16, and 18 feature vectors from the response signals of the above - mentioned stimuli in sequence.

[0056] Step 4.2. Perform singular value decomposition on the response matrix of the moving bar stimulus, which is divided into a time - related component and a direction - related component. Use sPCA for feature extraction on the direction - related component, extract 6 feature vectors. At the same time, perform exponential projection on the direction - related component, extract and calculate the direction index as the direction selectivity, and plot according to the selectivity, as Figure 14 shown. It can be seen from the figure that different cells have different direction selectivities.

[0057] Step 4.3. For the response signal of the black - and - white checkerboard stimulus, use the spike - triggered average method to calculate the receptive field of retinal ganglion cells. Use sPCA to extract 10 feature vectors from the receptive field. At the same time, plot the receptive field size, as Figure 15 shown, which reflects that the receptive fields of different retinal nerve cells are different, indicating that different types of retinal nerve cells can be distinguished by the response signal.

[0058] Step 4.4. Combine the feature vectors extracted in Steps 4.1, 4.2, and 4.3. Use these 100 feature vectors for clustering. Adopt GMM clustering and use the expectation - maximization algorithm for calculation. Constrain the covariance matrix of each component to be diagonal. Use the Bayesian information criterion (BIC) as an evaluation index, as Figure 16 shown. As can be seen from Figure 16 , when the number of classes is 40, the BIC is the lowest, indicating that the clustering effect is the best at this time. Figure 17 is the scatter plot of the clustering result. Figure 18 is the response signal diagram of one of the classes.

[0059] As Figure 9 shown, the relationship between the retinal spike response and color. Figure 9Blue indicates fewer action potential firings, and yellow indicates higher action potential firings. Among the two-color light stimuli of green and blue, the action potential firing is the highest during green light stimulation, indicating the highest sensitivity to green and insensitivity to blue, demonstrating color selectivity, which is consistent with the cones in the retina.

[0060] As Figure 10 shown, as the light intensity increases, the action potential firing rate shows an upward trend, which is consistent with the retinal mechanism. At low light intensities, the rods work to transmit signals to retinal nerve cells, and the retinal nerve cells fire spikes. As the light intensity increases, the cone cells also transmit signals to the retinal nerve cells, and at this time, the spike firing of the retinal nerve cells is higher than that at lower light intensities.

[0061] As Figure 11 、 12 and 13 shown, as the illumination time changes, the spike response is delayed. Regarding the relationship between the spike firing rate and the total number of firings, as the illumination time increases, the spike firing delay gradually increases, the total number of firings gradually increases, but the firing rate per unit time gradually decreases.

[0062] Example 2

[0063] As Figure 19 shown, the present invention uses the following system to perform light stimulation on the adhered retina of Example 1.

[0064] A light stimulation system that can be combined with multiple electrophysiological devices, including a computer 1, a projector 2, a bottom plate 3, a plurality of liftable struts 4, a convex lens assembly, and a beam splitting prism 6. The liftable struts 4 are fixedly installed on the bottom plate 3 by screws. The computer 1 is connected to the projector 2 and the electrophysiological device 7. The projector 2 is placed on one of the liftable struts 4. A bracket 9 is installed at the upper end of the remaining liftable struts 4. A convex lens assembly 5 (fixedly installed by screws) is installed on the bracket 9. There are 4 mounting holes on the convex lens assembly. A slide bar 8 is inserted into the mounting holes on the same horizontal line. A beam splitting prism 6 is inserted onto the slide bar 8.

[0065] Specifically, the bottom plate 3 is a perforated plate. The number of the liftable struts 4 is three. A support platform 11 is installed on one of the liftable struts 4, and the projector 2 is placed on the support platform 11. A bracket 9 (i.e., there are two brackets 9) is installed on each of the remaining two liftable struts 4. A convex lens assembly (i.e., there are two convex lenses 5, which are installed by screws) is installed on each bracket 9. The convex lens assembly includes a convex lens 5 and a convex lens bracket 10. The convex lens 5 is embedded in the convex lens bracket 10. The convex lens bracket 10 is rectangular. An installation hole is opened at each of the four right angles of the convex lens bracket 10. A slide bar 8 is inserted into the installation holes of the two convex lens brackets 10 on the same horizontal line. The two slide bars 8 below are fixed to the bracket 9 by a buckle. The number of the slide bars 8 is four, and the four slide bars 8 are horizontally parallel. Four through holes are opened on the beam splitting prism 6, and the four slide bars 8 are inserted into the four through holes.

[0066] Specifically, the liftable strut 4 includes a base 4.1 and a sliding rod 4.2. The sliding rod 4.2 is slidably installed in the base 4.1. An adjusting knob 4.3 is installed on the base 4.1. By rotating the adjusting knob 4.3, the sliding rod 4.2 is lifted and lowered in the base 4.1. The above-mentioned light stimulation system provided in this embodiment is used in combination with a high-density multi-electrode array instrument. Psychtoolbox is pre-installed on the computer 1. Through the EDITOR interface of the Matlab software, stimulation codes are written, and parameters such as the stimulation in step 2 of Embodiment 1 are set. The computer 1 and the programmable LED projector are connected by an HDMI data transmission line to present stimulation to the retina. At the same time, the USB port of the high-density multi-electrode array instrument is connected to the USB port of the computer 1 through a data line to synchronize stimulation and data recording.

[0067] The above has made an exemplary description of the present invention. It should be noted that without departing from the core of the present invention, any simple deformation, modification or equivalent replacement that can be made by those skilled in the art without creative labor falls within the protection scope of the present invention.

Claims

1. A method for classifying retinal cells, characterized in that: The following steps are involved: Step 1, taking an animal retina, and adhering a pretreated biochip to the retina to obtain an adhered retina, wherein the biochip is pretreated with poly-lysine to promote adhesion between the retina and the biochip; Step 2, light stimulation is applied to the retina after adhesion, and local stimulation, global stimulation, flickering stimulation, color stimulation, moving bar stimulation, shape stimulation and black and white checkerboard stimulation are applied respectively. After each application of the above stimulation, the retina needs to be kept in dark light for 10 seconds until the last stimulation is applied. The retina converts the light signal of the light stimulation into an electrical signal, wherein the local stimulation is centered on the center of the retina, with a diameter of 700 μm, and the light intensity changes from 0 to SSSlux following a sinusoidal waveform, and the frequency gradually increases from 1 Hz to 5 Hz, and the stimulation time of each frequency is 10 seconds; The diameter of the global stimulus is 2mm, the light intensity changes from 0 to SSSlux following a sine wave, the frequency gradually increases from 1Hz to 5Hz, and the stimulation time of each frequency is 10s; the diameter of the flicker stimulus is 3mm, the light intensity changes from 0 to SSSlux following a square wave, and the frequency is 1Hz; the diameter of the color stimulus is 3mm, the retina rests in dark light for 10s, blue stimulus for 3s, dark light rest for 3s, green stimulus for 3s, and the cycle is repeated 3 times, the length of the moving bar stimulus is 800um, the width is 100um, and the stimulation directions are divided into 8, namely from top to bottom, from upper right to bottom, from top to bottom to bottom, from top to bottom to bottom, from top to bottom to bottom, from top to bottom to bottom, from top to bottom to bottom To the lower left, from right to left, from the lower right to the top, from bottom to top, from the lower left to the upper right, from left to right, from the upper left to the lower right, the length of the movement path in each direction is 3mm; the shapes of the shape stimuli are triangles, squares and circles, first, the background is black, the color of each shape is green, each shape stimulation lasts for 3s, and rest in the dark for 3s, then the background is black, the color of each shape is blue, each shape stimulation lasts for 3s, rest in the dark for 3s, and rest in the dark for 10s in the retina; the background is white, the color of each shape is green, each shape stimulation lasts for 3s, and rest in the dark for 3s , the background is white, each shape is blue, each shape stimulation lasts for 3s, dark light rest for 3s between shapes, retinal dark light rest for 10s, the background is blue, the color of each shape is green, each shape stimulation lasts for 3s, dark rest for 3s, retinal dark light rest for 10s; the background is green, the color of each shape is blue, each shape stimulation lasts for 3s, dark rest for 3s; the black and white checkerboard stimulus consists of 50×50 squares, each grid size is 50×50um, the brightness and darkness of each checkerboard changes with time, and is randomly generated by a Gaussian function, and the duration is 5min; Step 3, collecting electrical signals, filtering the electrical signals, removing background noise, improving the signal-to-noise ratio, extracting spike potentials, performing statistical analysis, and obtaining preprocessed data; Step 4, classifying and labeling the preprocessed data, includes the following steps: Step 4.1, remove the 10-s rest time of the retina under dark light, reduce the data dimension to 1 kHz, perform sparse principal component analysis on the response signals of local stimulation, global stimulation, flicker stimulation, color stimulation, moving bar stimulation and shape stimulation, and extract 20, 20, 4, 6, 16 and 18 feature vectors of the response signals of the above stimulations in turn; Step 4.2, singular value decomposition is performed on the response matrix of the moving bar stimulus, which is divided into time-related components and direction-related components. sPCA is used to extract features of the direction-related components to extract 6 eigenvectors. At the same time, the direction-related components are exponentially projected, and the direction index is extracted and calculated as the direction selectivity. According to the selectivity, different cells have different direction selectivities, which is used to classify retinal neurons. Step 4.3, for the response signal of the black and white checkerboard stimulus, use the firing-triggering average method to calculate the receptive field of the retinal ganglion cells, use sPCA to extract 10 feature vectors for the receptive field, and plot the receptive field size. Different retinal neurons have different receptive fields, so the retinal neurons can be classified based on this. In step 4.4, the feature vectors extracted in steps 4.1, 4.2 and 4.3 are combined, and these 100 feature vectors are used for clustering. GMM clustering is adopted, and the maximum expectation algorithm is used for calculation. The covariance matrix of each component is constrained to be diagonal. The Bayesian Information Criterion (BIC) is used for evaluation. When the number of categories is 40, the BIC is the lowest, indicating that the clustering effect is best at this time.

2. The classification method according to claim 1, characterized in that: In step 3, AxIS software is used to extract the spike potential.

3. The classification method according to claim 1, characterized in that: In step 3, a high-density multi-electrode array instrument is used to collect electrical signals.

Images