CN114049964A - Modeling method of cross-brain-region artificial neural pathway - Google Patents
Modeling method of cross-brain-region artificial neural pathway Download PDFInfo
- Publication number
- CN114049964A CN114049964A CN202110559426.0A CN202110559426A CN114049964A CN 114049964 A CN114049964 A CN 114049964A CN 202110559426 A CN202110559426 A CN 202110559426A CN 114049964 A CN114049964 A CN 114049964A
- Authority
- CN
- China
- Prior art keywords
- neural
- pulse
- input
- nerve
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000010004 neural pathway Effects 0.000 title claims abstract description 47
- 210000000118 neural pathway Anatomy 0.000 title claims abstract description 47
- 238000000034 method Methods 0.000 title claims abstract description 38
- 230000001537 neural effect Effects 0.000 claims abstract description 119
- 210000005036 nerve Anatomy 0.000 claims abstract description 76
- 210000004556 brain Anatomy 0.000 claims abstract description 68
- 238000001208 nuclear magnetic resonance pulse sequence Methods 0.000 claims abstract description 17
- 230000009467 reduction Effects 0.000 claims abstract description 7
- 238000012800 visualization Methods 0.000 claims abstract description 3
- 210000002569 neuron Anatomy 0.000 claims description 48
- 210000002364 input neuron Anatomy 0.000 claims description 32
- 210000004205 output neuron Anatomy 0.000 claims description 31
- 230000006870 function Effects 0.000 claims description 27
- 238000010304 firing Methods 0.000 claims description 16
- 238000012216 screening Methods 0.000 claims description 16
- 230000008929 regeneration Effects 0.000 claims description 13
- 238000011069 regeneration method Methods 0.000 claims description 13
- 238000001914 filtration Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 9
- 238000009499 grossing Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 2
- 239000000126 substance Substances 0.000 claims description 2
- 230000002123 temporal effect Effects 0.000 claims 2
- 238000004364 calculation method Methods 0.000 abstract description 4
- 230000001149 cognitive effect Effects 0.000 abstract description 4
- 238000007781 pre-processing Methods 0.000 abstract 1
- 230000001172 regenerating effect Effects 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 6
- 230000001360 synchronised effect Effects 0.000 description 4
- 238000001276 Kolmogorov–Smirnov test Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000009827 uniform distribution Methods 0.000 description 3
- 241000282553 Macaca Species 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000003925 brain function Effects 0.000 description 1
- 208000029028 brain injury Diseases 0.000 description 1
- 210000003710 cerebral cortex Anatomy 0.000 description 1
- 230000003920 cognitive function Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000004907 flux Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000007433 nerve pathway Effects 0.000 description 1
- 230000007604 neuronal communication Effects 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012421 spiking Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000000946 synaptic effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4887—Locating particular structures in or on the body
- A61B5/4893—Nerves
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Neurology (AREA)
- Pathology (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Heart & Thoracic Surgery (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a method for predicting a brain region impulse neural signal, which comprises the following steps: 1) synchronously collecting pulse signals of the neural groups in the multiple brain areas; 2) calibrating the pulse signals of the nerve groups; 3) preprocessing a neural group pulse signal; 4) constructing a non-discretized neural pulse sequence kernel function; 5) dimension reduction of a regenerated nuclear Hilbert space; 6) solving an artificial neural pathway model in a regenerative nuclear hilbert space; 7) evaluating an artificial neural pathway model; 8) visualization of an artificial neural pathway model. The modeling method uses the non-discrete neural pulse sequence kernel function based on the time sequence neural pulse input, has higher output signal prediction precision, higher calculation efficiency and more stable performance, and is used for guiding the rehabilitation of the cognitive neural function.
Description
Technical Field
The invention belongs to the field of neural engineering, and particularly relates to a modeling method of a cross-brain-region artificial neural pathway.
Background
Neural pathways, consisting of synaptic connections of neurons, exist between major functional areas of the brain. Different brain areas communicate through nerve pathways to realize information transmission, so that normal cognitive functions (such as understanding, memory and the like) are realized. Upon brain injury, the neural pathways may be cut off, and cognitive nerve function may be reduced or even lost.
However, with the continuous development of microelectrode array technology and the continuous and deep research of modern medicine on brain functions, people can realize the synchronous acquisition of two groups of brain area nerve pulse signals, and can realize the functional connection between brain areas through analyzing and acquiring the correlation between the signals, thereby further constructing an artificial nerve path to supplement or replace the damaged original nerve path. For example, the patent application with publication number CN106529186A discloses a method for predicting impulse neural signals of brain region, and further, the patent application with publication number CN112101535A discloses a method for processing impulse neural signals and related apparatus.
Due to the complexity and nonlinearity of neural communication, the input signal flux is large, and artificial neural pathways need to be implanted into the human brain for work. Therefore, the mathematical model of the artificial neural pathway needs to have efficient and stable nonlinear expression capability, which provides a challenge for the traditional linear inefficient model.
Disclosure of Invention
In view of the above, the present invention provides a modeling method for a cross-brain-region artificial neural pathway, which is used for constructing an artificial neural pathway model capable of accurately and efficiently predicting and outputting a neural impulse signal of a brain region on the basis of multipoint synchronous recording of neural impulse signals, and the application of the artificial neural pathway model is used for guiding the rehabilitation of the cognitive neural function.
In order to achieve the above object, the embodiment of the present invention provides the following technical solutions:
a modeling method of a trans-brain region artificial neural pathway, comprising the steps of:
synchronously collecting nerve pulse signals of a plurality of nerve groups of an input brain area and an output brain area, and calibrating the time for distributing the nerve pulse signals and corresponding neurons according to the waveform characteristics of the nerve pulse signals;
carrying out time slot discretization processing on all nerve pulse signals; screening and filtering all neurons in an input brain area and an output brain area according to the neural pulse release rate;
aiming at each remaining output neuron screened and filtered in the output brain area, screening a plurality of input neurons from the input brain area as sample neurons according to the correlation between the neural pulse signals, and constructing a time sequence input neural pulse history of the sample neurons according to the time for issuing the neural pulse signals;
constructing a non-discretization neural pulse sequence function based on the time sequence input neural pulse history of the input neuron, projecting the time sequence input neural pulse history to a regeneration nuclear Hilbert space, and realizing the dimension reduction of the regeneration nuclear Hilbert space through clustering the time sequence input neural pulse history;
in a dimension-reduced regeneration kernel Hilbert space, a non-discretization nerve pulse sequence kernel function after linear weighting is used as a predicted value of a nerve pulse signal of an output neuron, a weight parameter of linear weighting is optimized through maximization of a likelihood function of the predicted value of the nerve pulse signal of the output neuron, and an artificial nerve path model is formed by a linear mapping relation consisting of the weight parameter.
The technical scheme provided by the embodiment has the beneficial effects that:
based on multi-region synchronous acquisition of multi-neuron neural pulse signals, time sequence input neural pulse history is adopted as an input variable, and an artificial neural pathway model capable of considering real-time prediction of the neural pulse signals is constructed on the basis of a non-discrete neural pulse sequence kernel function method. The model not only accepts the historical input of the time sequence input nerve pulse, improves the storage calculation efficiency, but also has high nonlinear expression capability and global optimal model stable solution due to the use of a kernel function method, so that the model has stable, efficient and accurate output nerve pulse signal prediction capability, and an artificial nerve channel consisting of the nerve pulse signal of the input neuron and the nerve pulse signal of the output neuron, and can be used for guiding the rehabilitation of the cognitive nerve function.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow diagram of a method for modeling a cross-brain region artificial neural pathway provided by an embodiment;
figure 2 is a schematic diagram of the artificial neural pathway provided in one embodiment operating in the brain.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is a flowchart of a modeling method of a cross-brain region artificial neural pathway provided in an embodiment. Figure 2 is a schematic diagram of the artificial neural pathway provided in one embodiment operating in the brain. As shown in fig. 1 and fig. 2, the modeling method of the trans-brain region artificial neural pathway provided by the embodiment includes the following steps:
step 1, synchronously collecting nerve pulse signals of multiple nerve groups of an input brain area and an output brain area.
In order to establish the artificial neural pathway, an input brain region corresponding to the input signal and an output brain region corresponding to the output signal need to be determined. The method comprises the steps of synchronously acquiring nerve pulse signals of multiple nerve groups of a brain area on the basis of determining an input brain area and an output brain area, specifically, burying more than two groups of channel electrode arrays at multiple positions (generally, the input brain area and the output brain area) on the surface of the cerebral cortex of a testee, and synchronously observing and recording nerve pulses issued by neurons in the brain area in real time when the testee performs related tasks, so as to realize the synchronous acquisition of the nerve pulse signals.
And 2, calibrating the time for sending the nerve pulse signal and the corresponding neuron according to the waveform characteristics of the nerve pulse signal.
Different neuron firing pulses are recorded by the same electrode, but the waveform characteristics of the firing pulses are different, and the time for firing the neural pulse signal and the neuron can be distinguished by analyzing the waveform characteristics. In an embodiment, the waveform features of the neural pulse signal include a peak value, a valley value, a peak-valley time interval, and the like, and the time for issuing the neural pulse signal and the corresponding neuron are calibrated according to the waveform features.
And 3, performing time slot discretization processing on all the nerve pulse signals.
In an embodiment, the time slot discretization processing of the neural pulse signal of the output brain region includes:
dividing the collected neural pulse signals according to the fixed time slot width, and recording the time slot with the pulse neural signals in the time slot as 1, otherwise recording as 0, thereby completing the discretization. For example, the fixed time slot width may be set to 10 milliseconds, and discretization of the neural pulse signal is performed by the time slot width of 10 milliseconds.
And 4, screening and filtering all neurons in the input brain area and the output brain area according to the nerve pulse release rate.
Because the distribution of the firing rate of the collected neural impulse signals is particularly wide, and some neural impulse firing rates are too high or too low, the part of neurons and the corresponding neural impulse signals need to be screened and filtered. In an embodiment, the screening and filtering of all neurons in the input brain region and the output brain region according to the neural pulse firing rate includes:
and filtering out neurons which are not in the range of the neural pulse release rate threshold value according to the set range of the neural pulse release rate threshold value aiming at all neurons in the input brain area and the output brain area. In an embodiment, the threshold range of the neural impulse firing rate may be set as [2Hz,40Hz ], and neurons with neural impulse firing rates outside the range of [2Hz,40Hz ] are screened and filtered out.
And 5, screening and filtering each remaining output neuron in the output brain area, and screening a plurality of input neurons from the input brain area as sample neurons according to the correlation between the nerve pulse signals.
On the basis of filtering out neurons through a neural impulse firing rate threshold range, further screening of the neurons is needed to obtain the neurons capable of establishing an artificial neural pathway model as sample data, and the sample neurons are named as sample neurons. In an embodiment, for each remaining output neuron in the output brain region, screening a plurality of input neurons from the input brain region as sample neurons according to a correlation between neural pulse signals, includes:
and aiming at each output neuron which is filtered and left in each output brain area, calculating mutual information between the output neuron and each input neuron in the input brain area, and screening a plurality of input neurons with the top n of the mutual information as sample neurons.
In an embodiment, n may be 10, after pairwise mutual information values of each path of pulse neural signals of the input brain area and the output brain area are calculated, for each path of pulse neural signals of the output brain area, the input neurons of the first 10 input brain areas with the highest mutual information values are selected as sample neurons, and the neural pulse signals corresponding to the sample neurons are used as input data of the artificial neural pathway model and used for predicting the neural pulse signals of the output neurons.
The mutual information calculation mode is as follows:
wherein x isiFor the input of the neural impulse signal of the ith input neuron of the brain region, yjFor outputting the neural pulse signal of the j output neuron in the brain region, p (x)t,yt) Is an event xtAnd ytJoint probability of simultaneous occurrence, p (x)t) And p (y)t) Respectively represent an event xtAnd ytProbability of occurrence of each event, event x, since the original brain region spiking neural signal has been discretizedtAnd event ytAll the values of (1) are in the range of {0,1 }.
And 6, constructing a time sequence of the sample neuron according to the time of the issuing of the neural pulse signal and inputting the neural pulse history.
On the basis of obtaining the sample neuron, correspondingly listing the latest occurring historical input nerve pulse time in the related input neuron at each discrete time interval to construct a time sequence input nerve pulse history, wherein the time sequence input nerve pulse history is an event sequence formed by the time of input neuron nerve pulse emission. In an embodiment, constructing a time-series input neural pulse history of a sample neuron as a function of time at which the neural pulse signal was fired comprises:
the time sequence formed by the time when the sample neuron sends out the neural pulse signal is taken as the time sequence input neural pulse history and is expressed as
Wherein the content of the first and second substances,
denotes the nth sample neuron at tkThe time at which the mth time before the time of day the nerve impulse signal was delivered.
And 7, constructing a non-discretized neural pulse sequence function based on the time sequence input neural pulse history of the input neuron, and projecting the time sequence input neural pulse history to a regeneration nuclear Hilbert space.
On the basis of obtaining the time sequence input nerve pulse history of the input neuron, constructing a nerve pulse sequence function by using an inner product function corresponding to the two input nerve pulse histories, and projecting an input signal to a regeneration nuclear Hilbert space. In an embodiment, constructing a non-discretized neural pulse sequence function from a history of input neuron-based time-series input neural pulses comprises:
the non-discretized neural pulse sequence kernel function κ (·) is expressed as:
dist(χi,χj)2＝κc(χi,χi)-2κc(χi,χj)+κc(χj,χj)
wherein, dist (χ)i,χj) Representing the ith time series input neural pulse history ×iAnd the jth time series input nerve pulse history chijThe distance between them is used for measuring χiHexix-jThe degree of dissimilarity between them is represented by κc(. cndot.) represents the cross firing strength kernel function between the histories of inputting two time-series input nerve impulses, M1 and M2 are indexes of the firing times of the nerve impulse signals, M is the total firing times of the nerve impulse signals, and N is the total number of neurons.
And 8, realizing the dimension reduction of the regeneration nuclear Hilbert space by clustering the history of the time-series input nerve pulse.
In order to limit the dimension of the regenerated kernel hilbert space from increasing linearly with input sample data and avoid the over-fitting problem of the input-output model, the dimension of the regenerated kernel hilbert space needs to be reduced. In an embodiment, the dimension reduction of the regenerated kernel hilbert space is realized by clustering the history of the sequential input neural pulses, and includes:
and clustering the time sequence input neural pulse histories according to the distance between the two time sequence input neural pulse histories, determining a clustering center, and forming a dimension-reduced regeneration kernel Hilbert space by using the time sequence input neural pulse histories corresponding to the clustering center.
In the embodiment, in the dimension reduction processing, the distance dist (χ) between the histories of nerve pulses is input at any two input timingsi,χj) And obtaining a plurality of violent results and corresponding clustering center points by adopting a K center point algorithm, and then forming a regenerated kernel Hilbert space after dimensionality reduction by taking the clustering center points as characteristic points representing the regenerated kernel Hilbert space.
And 9, taking the non-discretized neural pulse sequence kernel function after linear weighting as a predicted value of the neural pulse signal of the output neuron in the dimension-reduced regeneration kernel Hilbert space.
In the embodiment, in a dimension-reduced regeneration kernel Hilbert space, a neural pulse sequence kernel function kappa (·) constructed between time sequence input neural pulse histories corresponding to each clustering center is used as a basis, and a predicted value of a neural pulse signal of an output neuron based on the time sequence input neural pulse histories is obtained through linear weighting.
And step 10, optimizing linearly weighted weight parameters by maximizing the likelihood function of the predicted value of the neural pulse signal of the output neuron, and forming an artificial neural pathway model by linear mapping relation formed by the weight parameters.
In the embodiment, a neural pulse sequence kernel function constructed by two time sequence input neural pulse histories is used as a variable, and an artificial neural pathway model is formed by a weighted linear mapping relation of the variable. When the output neural pulse prediction model is solved, a solving framework of maximum likelihood estimation is applied, and an iteration reweighting least square method is adopted to obtain a global optimal solution of the weight of the neural pathway model. Specifically, the method optimizes the linearly weighted weight parameters by taking the maximization of the likelihood function of the predicted value of the neural pulse signal of the output neuron as a target, and maximizes the likelihood function of the neural pulse signal of the output neuron until reaching a global optimal value by continuously changing the weight parameters so as to obtain the artificial neural pathway model.
And 11, obtaining the artificial neural pathway by applying the artificial neural pathway model.
In the embodiment, when the artificial neural pathway model is applied to obtain the artificial neural pathway, the neural pulse signals input into the brain area are collected and discretized, after screening and filtering processing, time sequence input neural pulse history of input neurons is constructed, based on the constructed time sequence input neural pulse history, the neural pulse signals of output neurons are predicted by using the artificial neural pathway model, and the artificial neural pathway is formed according to the input neurons and the output neurons.
And step 12, visualizing the constructed artificial neural pathway model.
In an embodiment, the constructed artificial neural pathway model is further visualized, and the visualization process is as follows: the method comprises the steps of taking a nerve pulse signal contained in a time sequence input nerve pulse history corresponding to a cluster center as a representative nerve pulse signal of an input neuron, smoothing the nerve pulse signal, combining corresponding weight parameters to obtain a corresponding relation between the pulse nerve signal of the input neuron and the pulse nerve signal of an output neuron, smoothing any two nerve pulse signals in a plurality of nerve pulse signals of the input neuron corresponding to a plurality of cluster centers, and combining corresponding weight parameters to display the interaction of the pulse nerve signals of the two input neurons in the determination of the nerve pulse signal of the output neuron.
In the examples, evaluation of an artificial neural pathway model was also performed. And calculating the goodness of fit of the impulse neural signals of the output neurons predicted by the artificial neural pathway model by using the discrete-time rescaled Kolmogonov-Schmilrnov statistic as an evaluation standard for measuring the effect of the model. The specific calculation method of the Kermonov-Scironov statistic of discrete time rescaling comprises the following steps:
(1) in order to eliminate the influence caused by discrete time sampling, a new sequence q (t) is generated on the basis of the predicted output neuron firing probability lambda (t), and the specific form of the sequence q (t) is as follows:
q(t)＝-log(1-λ(t)Δ)
(2) cutting a sequence q (t) according to a pulse sequence generated by a neuron to be predicted, integrating q (t) values between two adjacent pulse signals, and correcting deviation generated by discrete time processing, wherein the specific form is as follows:
wherein, tiRefers to the time of occurrence of the ith pulse signal of the neuron to be predicted, ti+1The time of the (i + 1) th pulse signal of the neuron to be predicted, delta (i) is one orderThe resulting random variables were collected as follows:
where r (i) is a random variable uniformly distributed between [0,1 ].
(3) Scaling ζ (i) to yield:
z(i)＝1-e-ζ(i)
according to the Kolmonov-Schmilnoff test theory, the distribution of the set { z (i) } should follow a uniform distribution between [0,1 ];
(4) and (3) rearranging the set { z (i) } according to the numerical values from small to large, simultaneously drawing a graph with the uniform distribution between the standards [0,1] in a coordinate system for comparison, wherein the horizontal axis is the set { z (i) } and the vertical axis is the standard uniform distribution, and the maximum value difference of the two values on the vertical axis is taken as the range of the effect metric value and the 95% confidence interval to obtain a ratio value, which is also called DTR-KS metric value.
The modeling method of the artificial neural pathway of the trans-brain area can accept non-discretization neural pulse emitting time input, and the input mode can efficiently store neural pulse signals; the method has better nonlinear expression capability and better output prediction precision; the model has a global optimal solution, so the solution of the model is stable, and the prediction result and the performance of the model are relatively stable.
Examples of the experiments
Experiments are carried out on PMd brain areas and M1 brain areas of the macaque, the background of the signals is acquired when the macaque executes a four-way center-out task, the effective duration of the signals is 636.4 seconds, the signals comprise 127 paths of PMd brain areas and 102 paths of M1 brain areas, and 100 paths of effective PMd signal paths and 54 paths of effective M1 signal paths are obtained after screening.
The experimental effect is compared with the effect obtained by other pulse signal models, including a neural network model (NN), a second-order Laguerre-Volterra model (LVM), a Generalized Linear Model (GLM) and the like. Table 1 shows a lateral comparison of the pulse signal prediction effect of the present method with the remaining methods, where the DTR-KS statistic is an abbreviation of Discrete Time rescaled Kolmogorov Smirnov test (Discrete Time recalling Kolmogorov Smirnov test) statistic, and if the value is less than 1, the output prediction passes the Discrete Time rescaled Kolmogorov Smirnov test, and the value closer to 0 indicates the better prediction effect of the model.
TABLE 1 comparison of the effectiveness of the present method with several other types of prediction methods
The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.
Claims (10)
1. A modeling method of a trans-brain region artificial neural pathway is characterized by comprising the following steps:
synchronously collecting nerve pulse signals of a plurality of nerve groups of an input brain area and an output brain area, and calibrating the time for distributing the nerve pulse signals and corresponding neurons according to the waveform characteristics of the nerve pulse signals;
carrying out time slot discretization processing on all nerve pulse signals; screening and filtering all neurons in an input brain area and an output brain area according to the neural pulse release rate;
aiming at each remaining output neuron screened and filtered in the output brain area, screening a plurality of input neurons from the input brain area as sample neurons according to the correlation between the neural pulse signals, and constructing a time sequence input neural pulse history of the sample neurons according to the time for issuing the neural pulse signals;
constructing a non-discretization neural pulse sequence function based on the time sequence input neural pulse history of the input neuron, projecting the time sequence input neural pulse history to a regeneration nuclear Hilbert space, and realizing the dimension reduction of the regeneration nuclear Hilbert space through clustering the time sequence input neural pulse history;
in a dimension-reduced regeneration kernel Hilbert space, a non-discretization nerve pulse sequence kernel function after linear weighting is used as a predicted value of a nerve pulse signal of an output neuron, a weight parameter of linear weighting is optimized through maximization of a likelihood function of the predicted value of the nerve pulse signal of the output neuron, and an artificial nerve path model is formed by a linear mapping relation consisting of the weight parameter.
2. The method of claim 1, wherein the waveform features of the neural impulse signal include peak value, valley value, peak-valley interval, and the time for firing the neural impulse signal and the corresponding neuron are calibrated according to the waveform features.
3. The method for modeling a cross-brain region artificial neural pathway of claim 1, wherein said performing a time-slot discretization process on the neural impulse signal of the output brain region comprises:
dividing the collected neural pulse signals according to the fixed time slot width, and recording the time slot with the pulse neural signals in the time slot as 1, otherwise recording as 0, thereby completing the discretization.
4. The method of claim 1, wherein the screening and filtering of all neurons in the input brain region and the output brain region according to the neural impulse firing rate comprises:
and filtering out neurons which are not in the range of the neural pulse release rate threshold value according to the set range of the neural pulse release rate threshold value aiming at all neurons in the input brain area and the output brain area.
5. The method of claim 1, wherein the selecting each output neuron remaining for the output brain region, and selecting a plurality of input neurons from the input brain region as sample neurons according to the correlation between the neural impulse signals comprises:
and aiming at each output neuron which is filtered and left in each output brain area, calculating mutual information between the output neuron and each input neuron in the input brain area, and screening a plurality of input neurons with the top n of the mutual information as sample neurons.
6. The method of claim 1, wherein constructing a time-series input neural pulse history of a sample neuron as a function of time to fire a neural pulse signal comprises:
the time sequence formed by the time when the sample neuron sends out the neural pulse signal is taken as the time sequence input neural pulse history and is expressed as
Wherein the content of the first and second substances,
denotes the nth sample neuron at tkThe time at which the mth time before the time of day the nerve impulse signal was delivered.
7. The method of modeling a trans-brain region artificial neural pathway of claim 1, wherein said constructing a non-discretized neural pulse sequence function from a history of input neuron-based temporal input neural pulses comprises:
the non-discretized neural pulse sequence kernel function κ (·) is expressed as:
dist(χi,χj)2＝κc(χi,χi)-2χc(χi,χj)+κc(χj,χj)
wherein, dist (χ)i,χj) Representing the ith time series input neural pulse history ×iAnd the jth time series input nerve pulse history chijThe distance between them is used for measuring χiHexix-jThe degree of dissimilarity between them is represented by κc(. cndot.) represents the cross-firing strength kernel between the input two time-series input neural pulse histories.
8. The method of modeling a trans-brain region artificial neural pathway of claim 1, wherein said reducing dimensions of the regenerated nuclear hilbert space by clustering the temporal input neural pulse history comprises:
and clustering the time sequence input neural pulse histories according to the distance between the two time sequence input neural pulse histories, determining a clustering center, and forming a dimension-reduced regeneration kernel Hilbert space by using the time sequence input neural pulse histories corresponding to the clustering center.
9. The method of modeling a trans-brain region artificial neural pathway of claim 1, further comprising: collecting the neural pulse signals input into the brain area for discretization, constructing a time sequence input neural pulse history of the input neurons after screening, filtering and processing, predicting the neural pulse signals of the output neurons by using an artificial neural pathway model based on the constructed time sequence input neural pulse history, and forming an artificial neural pathway according to the input neurons and the output neurons.
10. A method of modelling a trans-brain region artificial neural pathway as claimed in any one of claims 1 to 9, further comprising: the constructed artificial neural pathway model is visualized, and the visualization process comprises the following steps: the method comprises the steps of taking a nerve pulse signal contained in a time sequence input nerve pulse history corresponding to a cluster center as a representative nerve pulse signal of an input neuron, smoothing the nerve pulse signal, combining corresponding weight parameters to obtain a corresponding relation between the pulse nerve signal of the input neuron and the pulse nerve signal of an output neuron, smoothing any two nerve pulse signals in a plurality of nerve pulse signals of the input neuron corresponding to a plurality of cluster centers, and combining corresponding weight parameters to display the interaction of the pulse nerve signals of the two input neurons in the determination of the nerve pulse signal of the output neuron.
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110559426.0A CN114049964A (en) | 2021-05-21 | 2021-05-21 | Modeling method of cross-brain-region artificial neural pathway |
| PCT/CN2021/095360 WO2021238825A1 (en) | 2020-05-25 | 2021-05-23 | Modeling method for artificial neural pathway across encephalic regions |
| US17/927,824 US20230206038A1 (en) | 2020-05-25 | 2021-05-23 | A modeling method for artificial neural pathway across encephalic regions |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110559426.0A CN114049964A (en) | 2021-05-21 | 2021-05-21 | Modeling method of cross-brain-region artificial neural pathway |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114049964A true CN114049964A (en) | 2022-02-15 |
Family
ID=80204346
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110559426.0A Pending CN114049964A (en) | 2020-05-25 | 2021-05-21 | Modeling method of cross-brain-region artificial neural pathway |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114049964A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023178737A1 (en) * | 2022-03-24 | 2023-09-28 | 中国科学院深圳先进技术研究院 | Spiking neural network-based data enhancement method and apparatus |
-
2021
- 2021-05-21 CN CN202110559426.0A patent/CN114049964A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023178737A1 (en) * | 2022-03-24 | 2023-09-28 | 中国科学院深圳先进技术研究院 | Spiking neural network-based data enhancement method and apparatus |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shoham et al. | Statistical encoding model for a primary motor cortical brain-machine interface | |
| Wang et al. | Online seizure prediction using an adaptive learning approach | |
| EP2797034B1 (en) | Event analyzer and computer-readable storage medium | |
| CN113842118B (en) | Epileptic seizure real-time detection monitoring system for epileptic video electroencephalogram examination | |
| CN117503057B (en) | Epileptic seizure detection device and medium for constructing brain network based on high-order tensor decomposition | |
| Thill et al. | Anomaly Detection in Electrocardiogram Readings with Stacked LSTM Networks. | |
| JPWO2019216378A1 (en) | Arithmetic logic unit, detection device, arithmetic method, and computer program | |
| CN114049964A (en) | Modeling method of cross-brain-region artificial neural pathway | |
| Soleimani-B et al. | Adaptive prediction of epileptic seizures from intracranial recordings | |
| WO2021238825A1 (en) | Modeling method for artificial neural pathway across encephalic regions | |
| US11712193B2 (en) | Reliable seizure detection with a parallelizable, multi-trajectory estimate of lyapunov exponents | |
| GB2522752A (en) | A system and method of detecting or predicting the onset of a neurological episode | |
| Çelebi et al. | An emotion recognition method based on EWT-3D–CNN–BiLSTM-GRU-AT model | |
| Wang et al. | A supervised learning algorithm to binary classification problem for spiking neural networks | |
| CN111523662A (en) | Dynamic process control limit determination method and device based on ADALINE neural network | |
| Biswas | Extraction of network information-quality and quantity-from nodes of neuronal network | |
| Okatan et al. | High performance decoding of behavioral information from mean background activity in extracellular neural recordings | |
| Barnathan | The Role and Predictability of Rich-Club Architecture in Mesoscale Cortical Networks | |
| Tverdokhlib et al. | Complex parameters evaluation of wavelet transformation | |
| CN118214502A (en) | Digital broadcast signal quality real-time monitoring method and system | |
| Subbotin | Neural network modeling of medications impact on the pressure of a patient with arterial hypertension | |
| Nowacki et al. | Deep learning for spike detection in deep brain stimulation surgery | |
| El Dawlatly et al. | Clustering synaptically-coupled neuronal populations under systematic variations in temporal dependence | |
| Boltshauser | Anomaly detection for prediction of failures in manufacturing environments: Machine learning based semi-supervised anomaly detection for multivariate time series to predict failures in a CNC-machine | |
| Vachkov | Monitoring System for Anomaly Detection in Plants with Continuous Operation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |