Neuron signal extraction self-supervised learning method and system based on sparse decomposition
By employing a self-supervised learning method based on sparse decomposition and utilizing the U-Net network for self-supervised and self-constrained training, the problem of background noise and impurities in neuroimaging is solved, enabling fast and accurate extraction of neuronal signals and reducing the cost of manual annotation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-30
- Publication Date
- 2026-04-10
AI Technical Summary
In existing technologies, background noise and impurities introduced by the optical properties of microscopes during neural imaging affect the extraction of neural signals. Traditional algorithms are computationally time-consuming and have poor noise resistance, while supervised learning methods require manual annotation, which is labor-intensive and time-consuming.
A self-supervised learning method based on sparse decomposition is adopted. Two U-Net networks are used to extract sparse signals and background information respectively. The network is trained by a loss function of self-supervised and self-constrained learning, and the neuron signals are extracted by combining a threshold segmentation method.
It achieves rapid and accurate extraction of neuronal signals, meets the requirements of real-time imaging, reduces the cost of manual annotation, and improves the accuracy and efficiency of signal extraction.
Smart Images
Figure CN121837069A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of deep learning computer image processing, and particularly relates to a neuron signal extraction self-supervised learning method based on sparse decomposition. BACKGROUND
[0002] In the field of neural imaging, the fluorescence signal triggered by the calcium ion indicator is used to reflect the action potential discharge mode of neurons, which provides key data support for understanding the brain information processing mechanism. However, in the actual imaging process, the optical characteristics of the microscope system itself will introduce significant background noise, and there are background fluorescence, blood vessels and other impurity information in the image, which affect the extraction of neural signals.
[0003] The traditional neural signal extraction algorithm is time-consuming in calculation and cannot meet the requirements of real-time processing, and has the problem of poor noise resistance. The supervised learning method for extracting neural signals needs experts to manually label neurons, which consumes more human and time costs. SUMMARY
[0004] The application provides a neuron signal extraction self-supervised learning method and system based on sparse decomposition, which aims to solve the problems of long calculation time in the existing traditional neural signal extraction method and the need for manual signal labeling and low extraction accuracy in the existing deep learning neural signal extraction method.
[0005] In a first aspect, the purpose of the application is to provide a neuron signal extraction self-supervised learning method based on sparse decomposition, comprising the following steps: S1: collecting neural microscopic imaging and denoising to obtain denoised neural microscopic imaging as a data set; S2: constructing a sparse decomposition network and training the sparse decomposition network using a loss function for performing a sparse decomposition task; S3: inputting the denoised neural microscopic imaging into the trained sparse decomposition network for inference to obtain a sparse part image representing neuron information; S4: drawing a ∆F / F image of the sparse part image representing neuron information and performing threshold segmentation processing to extract the spatial position of neurons; S5: extracting the ∆F / F time sequence signal of each neuron according to the spatial position of the neuron, and recording the spatial position and start and end time of each neuron event.
[0006] Further, a preferred scheme is provided: the sparse decomposition network comprises two U-Net networks.
[0007] Further, a preferred scheme is provided: the two U-Net networks are respectively used for extracting sparse signals and extracting background information.
[0008] Further, the preferred solution is provided: in S2, the first U-Net network is used to extract background information, and its loss function is represented as: , wherein, denotes the denoised neural microscopic imaging, denotes the output of the first U-Net network, denotes the nuclear norm of x, denotes the sum of each element of the matrix x, denotes the parameter balancing the low-rankness and sparsity, and its specific value is , wherein d is the side length of the training set image.[1] Further, the preferred solution is provided: in S2, the second U-Net network is used to extract sparse signals, and its loss function is represented as: .
[0009] Further, the preferred solution is provided: the ∆F / F image is a normalized fluorescence intensity change image, represented as: = , wherein, is the fluorescence intensity at this position, is the baseline fluorescence intensity.
[0010] In the second aspect, the purpose of the present application is to propose a neuron signal extraction self-supervised learning system based on sparse decomposition, characterized in that the system is realized based on the neuron signal extraction self-supervised learning method based on sparse decomposition according to any one or more of the above solutions, and the system comprises: a data acquisition module for acquiring neural microscopic imaging and performing denoising to obtain denoised neural microscopic imaging as a data set; a network construction module for constructing a sparse decomposition network and training the sparse decomposition network using a loss function for performing a sparse decomposition task; an image inference module for inputting the denoised neural microscopic imaging into the trained sparse decomposition network for inference to obtain a sparse part image representing neuron information; a threshold segmentation module for drawing a ∆F / F image of the sparse part image representing neuron information and performing threshold segmentation processing to extract the spatial position of neurons; an information recording module for extracting the ∆F / F time series signal of each neuron according to the spatial position of the neuron and recording the spatial position and start and end time of each neuron event.
[0011] In a third aspect, the present application is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program stored in the memory, the processor executes the neuron signal extraction self-supervised learning method based on sparse decomposition according to any one or more of the above-mentioned schemes.
[0012] In a fourth aspect, the present application is to provide a computer readable storage medium for storing a computer program, wherein the computer program executes the neuron signal extraction self-supervised learning method based on sparse decomposition according to any one or more of the above-mentioned schemes.
[0013] Compared with the prior art, the present application has the following advantages: The sparse decomposition used in the method of the present application takes advantage of the sparsity of neuron signals in time and space, thereby separating the signals from the background and obtaining clean neuron images for signal extraction. The sparse decomposition method of self-supervised learning has fast reasoning speed, can meet the requirements of real-time imaging, and has high extraction accuracy.
[0014] The present application is applicable to the field of neural imaging technology. BRIEF DESCRIPTION OF DRAWINGS
[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to the structures shown in these drawings without creative labor.
[0016] Figure 1 The flow chart of the neuron signal extraction self-supervised learning method based on sparse decomposition described in the specific embodiments of the present application. DETAILED DESCRIPTION
[0017] In the following description, specific details are set forth in order to provide a thorough understanding of the present application. However, persons skilled in the art will understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary details.
[0018] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.
[0019] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be practiced in other ways different from those described herein without departing from the scope of the present application, and those skilled in the art can make similar extensions without departing from the connotation of the present application, so the present application is not limited to the specific embodiments disclosed below.
[0020] Embodiment one The present embodiment proposes a neuron signal extraction self-supervised learning method based on sparse decomposition, which removes the non-uniform fluorescent background and other impurity information by using a trained self-supervised deep learning network to perform sparse decomposition on neural microscopic imaging, obtains a sparse part image that retains neural signals, and finally uses a threshold segmentation method to quickly and accurately extract the temporal variation and spatial position information of neuron signals, thereby realizing real-time observation and recording of neural activity.
[0021] The method comprises the following steps: Step 1: Obtain the denoised neural microscopic imaging as a data set to avoid that the high gray value noise points are incorrectly divided into sparse neural signals.
[0022] Step 2: Construct a sparse decomposition network, wherein the sparse decomposition network comprises two U-Net networks, and the two networks are trained in each round in turn to extract sparse signals and extract background information, respectively.
[0023] Step 3: Train the two U-Net networks by using a loss function for performing a sparse decomposition task, and the outputs of the two networks in each round are also used for self-supervised and self-constrained learning to strengthen the convergence speed and stability of the model.
[0024] Specifically, the first U-Net network ( ) is used to extract background information, and the loss function thereof is represented as: , wherein, denotes the denoised neural microscopic imaging, denotes the output of the first U-Net network, denotes the nuclear norm of x, denotes the sum of each element of the matrix x, denotes a parameter for balancing the low rank and sparsity, and the specific value is wherein d is the side length of the training set image.
[0025] The second U-Net network (U-Net2) is used to extract the sparse signal, and the loss function thereof is represented as: .
[0026] The self-supervised self-constrained learning method only needs the denoised neural imaging to realize the neural signal extraction, without the need for artificial labeling of the training model. The self-constrained term is introduced into the loss function to improve the model stability and convergence speed.
[0027] Step 4: input the denoised neural microscopic imaging into the trained sparse decomposition network for inference to obtain the sparse partial image representing the neuron information.
[0028] Step 5: draw the ∆F / F image of the sparse partial image and perform threshold segmentation processing to extract the spatial position of the neuron.
[0029] Step 6: according to the spatial position of the neuron recorded in step e, extract the ∆F / F time sequence signal of each neuron, and record the spatial position and start and end time of each neuron event.
[0030] The method described in the embodiment inputs the denoised neural microscopic imaging into two U-Net networks in sequence for self-supervised self-constrained training, so that the two networks respectively output the sparse partial image representing the neural signal and the low-rank partial image representing the background. The sparse partial image removes the fluorescence background, artifacts, impurities and other information in the original image, and the threshold segmentation method can be used to accurately and quickly extract the neuron signal.
[0031] Embodiment Two The embodiment proposes a self-supervised learning system for neuron signal extraction based on sparse decomposition, characterized in that the system is realized based on a self-supervised learning method for neuron signal extraction based on sparse decomposition as described in Embodiment One, and the system comprises: a data acquisition module for acquiring neural microscopic imaging and performing denoising to obtain denoised neural microscopic imaging as a data set; a network construction module for constructing a sparse decomposition network and training the sparse decomposition network using a loss function for performing a sparse decomposition task; an image inference module for inputting the denoised neural microscopic imaging into the trained sparse decomposition network for inference to obtain a sparse partial image representing neuron information; Threshold segmentation module: used to draw the ∆F / F image representing the sparse part of the neuron information, and perform threshold segmentation processing to extract the spatial position of the neuron; Information recording module: used to extract the ∆F / F time sequence signal of each neuron according to the spatial position of the neuron, and record the spatial position and start and end time of each neuron event.
[0032] It can be understood that the present application is described through some embodiments, and those skilled in the art know that various changes or equivalent replacements can be made to the features and embodiments without departing from the spirit and scope of the present application. In addition, under the guidance of the present application, the features and embodiments can be modified to adapt to specific conditions and materials without departing from the spirit and scope of the present application. Therefore, the present application is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of the present application are within the scope of the present application.
[0033] [1] represents the parameter of balance low rank and sparsity.
Claims
1. A self-supervised learning method for extracting neuronal signals based on sparse decomposition, characterized in that, Includes the following steps: S1: Acquire neural microscopic images and denoise them to obtain the denoised neural microscopic images as a dataset. S2: Construct a sparse decomposition network and train the sparse decomposition network using the loss function for performing the sparse decomposition task; S3: Input the denoised neural microscopic image into the trained sparse decomposition network for inference to obtain the sparse part image representing neuron information. S4: Draw the ∆F / F image representing the sparse part of the neuron information, and perform threshold segmentation to extract the spatial location of the neuron; S5: Based on the spatial location of the neurons, extract the ∆F / F time sequence signal of each neuron and record the spatial location and start and end time of each neuron event.
2. The self-supervised learning method for extracting neuronal signals based on sparse decomposition according to claim 1, characterized in that, The sparse decomposition network comprises two U-Net networks.
3. The self-supervised learning method for extracting neuronal signals based on sparse decomposition according to claim 2, characterized in that, The two U-Net networks are used to extract sparse signals and extract background information, respectively.
4. The self-supervised learning method for extracting neuronal signals based on sparse decomposition according to claim 2, characterized in that, In S2, the first U-Net network is used to extract background information, and its loss function is expressed as: ， in, This represents a denoised neural microscopic image. This represents the output of the first U-Net network. This indicates the search for the nuclear norm of x. This represents finding the sum of all elements in matrix x. The parameter represents the balance between low rank and sparsity, with a specific value of ,in d The side length of the training set images.
5. The self-supervised learning method for extracting neuronal signals based on sparse decomposition according to claim 4, characterized in that, In S2, the second U-Net network is used to extract sparse signals, and its loss function is expressed as: 。 6. The self-supervised learning method for extracting neuronal signals based on sparse decomposition according to claim 1, characterized in that, The ∆F / F image is a normalized fluorescence intensity change image, represented as follows: = ， in, The fluorescence intensity at this location, The baseline fluorescence intensity is given.
7. A supervised learning system for extracting neuronal signals based on sparse decomposition, characterized in that, The system is implemented based on a self-supervised learning method for extracting neuronal signals based on sparse decomposition as described in any one of claims 1-6, and the system includes: Data acquisition module: used to acquire neural microscopic images and perform denoising to obtain denoised neural microscopic images as a dataset; Network construction module: used to construct the sparse decomposition network and train the sparse decomposition network using the loss function of performing the sparse decomposition task; Image reasoning module: used to input the denoised neural microscopic image into the trained sparse decomposition network for reasoning, and obtain the sparse part image representing neuronal information; Thresholding segmentation module: used to draw the ∆F / F image representing the sparse part of the image that represents neuron information, and perform thresholding segmentation to extract the spatial location of neurons; Information recording module: used to extract the ∆F / F time sequence signal of each neuron according to the spatial location of the neuron, and record the spatial location and start and end time of each neuron event.
8. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes a self-supervised learning method for extracting neuronal signals based on sparse decomposition according to any one of claims 1-6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program that executes a self-supervised learning method for extracting neuronal signals based on sparse decomposition, as described in any one of claims 1-6.