A method and system for wide-field single-photon calcium imaging of brain neurons

Abstract

The application discloses a kind of brain neuron wide-field single-photon calcium imaging method and system, wherein brain neuron wide-field single-photon calcium imaging method includes the following steps: from the reference frame image of continuous multiple frames of brain nerve image, reference frame image is divided into multiple overlapping subareas;From the brain nerve image of continuous multiple frames, one frame image is selected as current frame image one by one, and current frame image is divided into multiple overlapping subareas, each subarea corresponds with the size and position of the subarea of reference frame image;The displacement vector of the subarea of corresponding position on current frame image and reference frame image is calculated respectively using optical flow method;Whether all displacement vectors are consistent is judged, if yes, brain blood vessel features are used for image registration to remove motion artifact, if not, NorMCorre algorithm is used to remove motion artifact.Displacement state is analyzed first before image registration, and appropriate registration method is selected according to displacement state, which improves registration accuracy.

Description

Technical Field

[0001] This invention relates to the field of neuroimaging technology, specifically to a method and system for wide-field single-photon calcium imaging of brain neurons. Background Technology

[0002] Calcium imaging is an important tool for observing the activity of the brain's nervous system. During long-term dynamic calcium imaging of the brains of conscious, living animals, complex motion artifacts, including translation, rotation, and non-rigid deformations, can occur between consecutively acquired image sequences due to unavoidable biological movements (e.g., respiration, heartbeat, minor head movements), instrument drift (e.g., thermal expansion or mechanical instability of the microscope platform causing image position shifts), or non-rigid deformations of the brain tissue itself (e.g., blood flow, edema). Therefore, precise image registration (artifact correction) is essential to ensure that the spatial location (usually defined as the region of interest) of the same neuron corresponds accurately in images acquired at different time points, thereby accurately extracting the neuronal calcium signal and enabling long-term dynamic tracking.

[0003] The NorMCorre algorithm is a commonly used non-rigid motion correction method. Based on optical flow, it first divides the image into multiple overlapping small regions and estimates the displacement field of each small region. By comparing the pixel intensity distribution of the reference frame and the current frame, the displacement vector of each small region is calculated. Smoothing constraints are applied to the displacement fields of adjacent small regions to ensure the continuity of overall deformation. Then, through multiple rounds of iterative optimization of the displacement field, the image residual (the difference between the reference frame and the corrected image) is gradually reduced, thereby correcting non-rigid motion in the image and making the acquired calcium imaging more accurate.

[0004] However, the NorMCorre algorithm focuses on modeling and correcting non-rigid deformations. Its core algorithm mechanism (small-block translation) cannot effectively handle rigid motions (such as overall translation or rotation of the field of view) or large-scale displacements. When rigid or large-scale displacements occur between images (e.g., displacements exceeding approximately 50 pixels due to slight loosening of the animal's head or shift in its initial position), the algorithm's correction effectiveness decreases significantly, and the risk of failure increases dramatically. A 50-pixel displacement only accounts for 1.25% of the image width in high-resolution imaging (e.g., 4000×5500 pixels), and the absolute displacement value is very small from a macroscopic perspective. This tiny displacement threshold severely limits the applicability of NorMCorre, forcing experiments to meet extremely stringent, near-ideal mouse head fixation stability requirements. This necessitates highly skilled operators or significantly more expensive and powerful fixation equipment. Furthermore, the system has low tolerance for environmental factors such as vibration and temperature, leading to difficult, time-consuming, high-failure-rate, and costly experimental preparations, becoming a bottleneck for large-scale applications and long-term stable observations. Therefore, there is an urgent need for a novel calcium imaging method and system that can intelligently predict displacement patterns and automatically switch to the optimal registration strategy under wide-field single-photon imaging conditions, in order to solve the problem of correction failure of existing technologies (such as the single NorMCorre algorithm) when faced with large-scale rigid displacement or compound motion (rigid translation / rotation superimposed local deformation). Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for wide-field single-photon calcium imaging of brain neurons, aiming to solve the technical problem in existing technologies where large overall rigid displacement of the image causes traditional registration algorithms (such as NorMCorre) to fail, thus failing to effectively correct motion artifacts. The method for wide-field single-photon calcium imaging of brain neurons includes the following steps:

[0006] A fixed reference frame image is created from a sequence of multiple consecutive brain neural images, and the reference frame image is divided into a preset number of sub-regions with preset positions and preset sizes that overlap with each other.

[0007] From the sequence of consecutive brain neural images, one frame is selected one by one as the current frame image to be processed. The current frame image is divided into sub-regions that have the same number, the same position, and the same size as the reference frame image and also overlap with each other.

[0008] The displacement vector of each sub-region of the current frame image relative to the corresponding sub-region on the reference frame image is calculated using the optical flow method.

[0009] If all displacement vectors are consistent, it is determined that the image has undergone global rigid displacement, and an image registration algorithm based on cerebral blood vessel feature point matching is used to correct the motion artifacts of the current frame image. If not, it is determined that there is local non-rigid deformation, and the NorMCorre algorithm based on non-rigid motion model is used to correct the motion artifacts of the current frame image.

[0010] The basic idea of ​​the above method is to calculate the displacement vector of each sub-region using the NorMCorre algorithm, and then determine whether there is rigid displacement in the overall image. If so, image registration is performed using cerebral vascular features to remove motion artifacts. Since the brain contains many cerebral blood vessels, which have relatively obvious features and are usually in relatively fixed positions, even if displacement occurs, the amount of displacement is small and continuous. Therefore, image registration using cerebral vascular features can effectively remove motion artifacts.

[0011] Specifically, the following steps are used to determine whether all the displacement vectors are consistent:

[0012] Construct a displacement vector field matrix by arranging all the displacement vectors according to their spatial positional relationships;

[0013] Calculate the average displacement vector of all the aforementioned displacement vectors;

[0014] Subtract the average displacement vector from each displacement vector in the displacement vector field matrix to obtain the displacement difference field matrix, and calculate the Frobenius norm of the displacement difference field matrix.

[0015] If the Frobenius norm of the displacement difference field matrix is ​​less than the first threshold, then all the displacement vectors are consistent.

[0016] Specifically, image registration is performed using cerebral vascular features through the following steps:

[0017] Select one frame from the sequence of multiple consecutive brain neural images as a fixed image;

[0018] One frame is selected from the continuous multi-frame brain neural image sequence as a floating image;

[0019] Extract cerebral vascular features from the fixed image and construct a SIFT descriptor for the fixed image; extract cerebral vascular features from the floating image and construct a SIFT descriptor for the floating image.

[0020] Based on the SIFT descriptor of the fixed image and the SIFT descriptor of the floating image, coarse registration is performed on the fixed image and the floating image;

[0021] The fixed image and the floating image are precisely registered using a B-spline free deformation model. Specifically, the cerebral vascular features of the fixed image and the floating image are extracted through the following steps:

[0022] A vascular network topology map of the fixed image or the floating image is generated using a Frangi filter combined with a skeletonization algorithm.

[0023] Extract the vascular network bifurcation points from the vascular network topology diagram.

[0024] Further, the SIFT descriptor of the fixed image is constructed through the following steps:

[0025] Determine whether there are vascular network bifurcation points in the fixed image. If so, set the vascular network bifurcation points as key feature points of the fixed image and construct the SIFT descriptor of the fixed image. If not, divide the blood vessels in the fixed image into at least two segments, and use the endpoints of each segment as key feature points to construct the SIFT descriptor of the fixed image.

[0026] Specifically, the SIFT descriptor of the floating image is constructed through the following steps:

[0027] If there are vascular network bifurcation points in the fixed image, then the vascular network bifurcation points in the floating image are set as key feature points of the floating image, and a SIFT descriptor of the floating image is constructed. If there are no vascular network bifurcation points in the fixed image, then the blood vessels in the floating image are divided into at least two segments, and the endpoints of each segment are used as key feature points to construct a SIFT descriptor of the floating image.

[0028] Specifically, the coarse registration of the fixed image and the floating image further includes the following steps:

[0029] Select a pair of matching key feature points of a fixed image and key feature points of a floating image, and calculate the forward transformation matrix from the key feature points of the floating image to the matching key feature points of the fixed image;

[0030] Calculate the inverse transformation matrix from the key feature points of the fixed image to the matching key feature points of the floating image;

[0031] Determine whether the forward transformation matrix and the inverse transformation matrix are consistent. If they are, then the key feature points of the fixed image and the key feature points of the floating image that are matched participate in the objective function of constructing the B-spline free deformation model.

[0032] Specifically, the fine registration of the fixed image and the floating image includes the following steps:

[0033] The SIFT descriptors of the fixed image and the floating image are matched to initially align the fixed image and the floating image.

[0034] The objective function is constructed by combining the confidence weight coefficient and the regularization term weight coefficient with the rigid transformation function and the second-order partial displacement field.

[0035] The B-spline free deformation model is constructed based on the objective function, and then fine registration is performed.

[0036] The present invention also provides a wide-field single-photon calcium imaging system for brain neurons, comprising: a fluorescence microscope module, a calcium imaging data analysis module, and an experimental platform. The fluorescence microscope module is used to acquire a sequence of consecutive multi-frame brain neural images of a live animal brain. The calcium imaging data analysis module uses the method described in any one of claims 1 to 8 to process the consecutive multi-frame brain neural images. The experimental platform is used to hold samples and provide an experimental environment. The experimental platform includes a sample fixation device and an environmental parameter control unit. The head fixation device is used to fix the brain of a live animal, and the environmental parameter control unit is used to control various parameters of the experimental environment.

[0037] Specifically, the calcium imaging data analysis module includes an image segmentation unit, a displacement vector calculation unit, and an image registration unit. The image segmentation unit is used to create a fixed reference frame image from a sequence of multiple consecutive brain nerve images, select one frame image from the sequence of multiple consecutive brain nerve images as the current frame image to be processed, and divide the reference frame image and the current frame image into sub-regions with the same preset number, the same preset position, and the same preset size, and which overlap with each other. The displacement vector calculation unit is used to calculate the displacement vector of each sub-region and determine whether all the displacement vectors are consistent. The image registration unit is used to call different registration algorithms to register the multiple consecutive brain nerve images according to the output result of the displacement vector calculation unit. If the output result of the displacement vector calculation unit is yes, it is determined that the image has a global rigid displacement, and the motion artifact of the current frame image is corrected by an image registration algorithm based on cerebral blood vessel feature point matching. If the output result of the displacement vector calculation unit is no, it is determined that there is local non-rigid deformation, and the motion artifact of the current frame image is corrected by the NorMCorre algorithm based on a non-rigid motion model.

[0038] The technical effects and advantages of this invention are as follows: By registering brain nerve images using the above method, an appropriate registration method can be automatically selected based on the displacement state of the brain nerve images, resulting in clearer brain nerve images and facilitating further processing. Simultaneously, it improves the automation level of the program, reduces the operational skills required of experimental personnel, and lowers the difficulty of system deployment. Attached Figure Description

[0039] Figure 1 This is an overall flowchart of the wide-field single-photon calcium imaging method for brain neurons of the present invention.

[0040] Figure 2 This is a flowchart of image registration using cerebral blood vessel feature point matching in this invention.

[0041] Figure 3 This is a flowchart illustrating the process of filtering key feature points for mismatches in this invention.

[0042] Figure 4 This is a full-cortical wide-field calcium image acquired by the calcium imaging system of the present invention.

[0043] Figure 5 This is a single-cell wide-field calcium image acquired by the calcium imaging system of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1

[0046] refer to Figure 1 Embodiment 1 of the present invention provides a method for wide-field calcium imaging of brain neurons, comprising the following steps:

[0047] S1. Create a fixed reference frame image from a sequence of multiple consecutive brain neural images, and divide the reference frame image into sub-regions with a preset number, preset position, and preset size that overlap with each other.

[0048] S2. Select one frame image from a sequence of multiple consecutive brain neural images as the current frame image to be processed, and divide the current frame image into sub-regions that have the same number, the same position, the same size, and the same overlap with each other as the reference frame image.

[0049] S3. Calculate the displacement vector of each sub-region of the current frame image relative to the corresponding sub-region on the reference frame image using the optical flow method.

[0050] S4. Determine whether all displacement vectors are consistent. If yes, determine that the image has undergone global rigid displacement, and use an image registration algorithm based on cerebral blood vessel feature point matching to correct the motion artifacts of the current frame image. If no, determine that it contains local non-rigid deformation, and use the NorMCorre algorithm based on a non-rigid motion model to correct the motion artifacts of the current frame image.

[0051] The basic idea of ​​the above method is to calculate the displacement vector of each sub-region using the NorMCorre algorithm, and then determine whether there is rigid displacement in the overall image. If so, image registration is performed using cerebral vascular features to remove motion artifacts. Since the brain contains many cerebral blood vessels, which have relatively obvious features and are usually in relatively fixed positions, even if displacement occurs, the amount of displacement is small and continuous. Therefore, image registration using cerebral vascular features can effectively remove motion artifacts.

[0052] Specifically, in step S1, the reference frame image can be either the first frame of the video or a time-averaged frame. A time-averaged frame is a static image generated by averaging multiple consecutive frames of a calcium imaging video over time. It integrates frame sequence information from different times, eliminating random noise and instantaneous motion interference to form a stable spatial feature template. Its calculation formula is as follows:

[0053]

[0054] In the formula, N is the total number of frames in the video, x, y are the coordinates of pixels, and Frame is the number of frames. t (x,y) is the fluorescence intensity value at pixel position (x,y) in the t-th frame of the image.

[0055] There are other methods for calculating the time-averaged frame, all of which are existing technologies and are documented in various research papers on the NoRMCorre algorithm, so they will not be elaborated here.

[0056] In step S3, the optical flow method calculates the displacement vectors of sub-regions of the reference frame image and the current frame image by analyzing the changes in pixel intensity (fluorescence intensity) in the image.

[0057] In step S4, the following steps are used to determine whether all displacement vectors are consistent:

[0058] S41. Construct a displacement vector field matrix by arranging all displacement vectors according to their spatial positional relationships;

[0059] For example, n displacement vectors d i =(x i ,y i The constructed displacement vector field matrix is:

[0060]

[0061] S42. Calculate the average displacement vector of all displacement vectors;

[0062] The components of the average displacement vector are the average of the corresponding components of all displacement vectors.

[0063]

[0064] S43. Subtract the average displacement vector from each displacement vector in the displacement vector field matrix to obtain the displacement difference field matrix, and calculate the Frobenius norm of the displacement difference field matrix.

[0065] S44. If the Frobenius norm of the difference matrix is ​​less than the first threshold, then all displacement vectors are consistent.

[0066] In step S4, when all displacement vectors are consistent, it can be determined that rigid body displacement has occurred in the image. Image registration requires an image registration algorithm based on cerebral blood vessel feature point matching. (Refer to...) Figure 2 This includes the following steps:

[0067] S41. Select one frame from a sequence of multiple consecutive brain neural images as a fixed image;

[0068] S42. Select one frame from a series of consecutive brain neural image sequences as a floating image.

[0069] S43. Extract cerebral vascular features from the fixed image and construct the SIFT descriptor for the fixed image; extract cerebral vascular features from the floating image and construct the SIFT descriptor for the floating image.

[0070] S44. Based on the SIFT descriptors of the fixed image and the floating image, perform coarse registration of the fixed image and the floating image;

[0071] S45. Use the B-spline free deformation model to perform fine registration of the fixed and floating images.

[0072] B-spline free deformation model is a parametric, non-rigid registration method that generates a smooth and locally controllable deformation field by adjusting the positions of control points. Its core idea is to leverage the local support and continuity of B-splines to flexibly align local structural differences in an image while maintaining deformation smoothness. It can effectively align images with complex local deformations, and its key advantage lies in the balance between local controllability and global smoothness.

[0073] Before registering the B-spline free deformation model, coarse registration of the image using SIFT descriptors can reduce the complexity of the objective function of the B-spline free deformation model, thereby reducing the computational load of registration and improving the computational speed.

[0074] Specifically, there are various methods for extracting blood vessel features from images, including traditional morphology-based analysis methods and deep learning methods based on segmentation models. This invention employs a traditional morphology-based analysis method, which has low computational cost and can effectively reduce time costs when processing large amounts of video or image data.

[0075] Specifically, cerebral vascular features are extracted from fixed and floating images through the following steps:

[0076] S431. Generate vascular network topology maps of fixed or floating images using the Frangi filter combined with skeletonization algorithm.

[0077] S432. Extract the vascular network bifurcation points from the fixed or floating image of the vascular network topology.

[0078] The Frangi filter is a classic image processing algorithm proposed by Alejandro Frangi in 1998 for enhancing tubular structures (such as blood vessels) in images. It detects local linear or tubular structures by analyzing the eigenvalues ​​of the Hessian matrix of an image, while suppressing other shaped regions (such as planes or spots).

[0079] Skeletonization is an image processing technique that simplifies two-dimensional shapes or three-dimensional structures into a "skeleton," aiming to preserve the topological structure and key geometric features of an object while removing redundant details. The skeleton is typically represented as a single pixel (or minimum width) representation of the object's center line, thus effectively eliminating the influence of blood vessel expansion or small displacements during feature recognition, and accurately extracting key features of the blood vessel (such as bifurcation points).

[0080] After extracting vascular features, SIFT descriptors for both fixed and floating images are constructed using the vascular network bifurcation points as key feature points:

[0081] S433. Determine whether there are bifurcation points of the vascular network in the fixed image. If so, set the bifurcation points of the vascular network as key feature points of the fixed image and construct the SIFT descriptor of the fixed image. If not, divide the blood vessels in the fixed image into at least two segments, take the endpoints of each segment as key feature points, and construct the SIFT descriptor of the fixed image.

[0082] Considering that the brain features displayed in the floating image are not significantly different from those in the fixed image, the method for creating the SIFT descriptor for the floating image can be determined based on the fixed image. For example, when there are vascular network bifurcation points in the fixed image, the floating image also uses these bifurcation points as key feature points to construct the SIFT descriptor. When there are no vascular network bifurcation points in the fixed image, the blood vessels in the floating image are similarly divided into at least two segments, and the endpoints of each segment are used as key feature points to construct the SIFT descriptor.

[0083] SIFT (Scale-Invariant Feature Transform Descriptor) is a high-dimensional vector used in computer vision to represent local features of an image. Its core advantage lies in its robustness to scale changes, rotations, and illumination variations.

[0084] Specifically, the main steps for constructing a SIFT descriptor are as follows:

[0085] S434. Assign a principal direction to each key feature point to achieve rotation invariance;

[0086] S435. Construct a 128-dimensional feature vector, i.e., a SIFT descriptor, based on the location, scale, and principal direction of key feature points.

[0087] Based on SIFT descriptors, coarse registration of fixed and floating images can be achieved. The registration method can use kd-tree neighbor search or FLANN matcher to match key feature points of fixed and floating images based on the similarity of descriptors.

[0088] In the coarse registration process, to improve registration accuracy and reduce the computational burden of subsequent fine registration steps, bidirectional matching verification can be implemented to filter out key feature points of mismatches, referencing... Figure 3 Specifically, it includes the following steps:

[0089] S441. Select a pair of matching key feature points of the fixed image and key feature points of the floating image, and calculate the forward transformation matrix from the key feature points of the floating image to the aligned key feature points of the fixed image.

[0090] S442. Calculate the inverse transformation matrix from the key feature points of the fixed image to the corresponding key feature points of the floating image;

[0091] S443. Determine whether the forward transformation matrix and the inverse transformation matrix are consistent. If they are, then the key feature points of the fixed image and the key feature points of the floating image in this pair of matching are used to construct the objective function of the B-spline free deformation model.

[0092] In step S443, the method for determining whether the forward transformation matrix and the inverse transformation matrix are the same is the same as the method for comparing the overall displacement matrix and the average displacement matrix. That is, the forward transformation matrix and the inverse transformation matrix are subtracted, and the F-norm of the second difference matrix is ​​calculated. If the F-norm of the second difference matrix is ​​less than the matching accuracy threshold, then the forward transformation matrix and the inverse transformation matrix are considered to be the same.

[0093] The RANSAC algorithm can also be used to remove key feature points that are mismatched. Through random sampling and iterative verification, “interior points” that support the same geometric transformation model can be selected from the matching point pairs.

[0094] After coarse registration of the fixed and floating images, fine registration is performed using a B-spline curve free deformation model to align local structural differences in the images. This includes the following steps:

[0095] S444. Match the SIFT descriptors of the fixed image and the SIFT descriptors of the floating image to perform preliminary alignment of the fixed image and the floating image.

[0096] S445. The objective function is constructed by combining the confidence weight coefficient and the regularization term weight coefficient with the rigid transformation function and the second-order partial displacement field.

[0097] The objective function can be expressed in the following form:

[0098]

[0099] In the formula, W k The confidence weighting coefficient is... It is a rigid transformation function. Here are the original coordinates of the k-th key feature point in the floating image. To fix the image with The original coordinates of the kth key feature point matched, where α is the weight coefficient of the regularization term.

[0100] S446. Construct a B-spline free deformation model based on the objective function and perform fine registration.

[0101] Using regularization weight coefficients can ensure the smoothness and continuity of the displacement field and prevent local overfitting. At the same time, confidence weight coefficients are used to adjust the transformation function weights of key feature points to control the accuracy of the registration results, thereby balancing accuracy and performance.

[0102] By registering brain nerve images using the above method, an appropriate registration method can be automatically selected based on the displacement state of the brain nerve images, resulting in clearer brain nerve images and facilitating further processing. This also improves the automation level of the program, reduces the operational skills required of experimental personnel, and lowers the difficulty of system deployment.

[0103] Example 2

[0104] The present invention also provides a wide-field single-photon calcium imaging system for brain neurons, including a fluorescence microscope system, a calcium imaging data analysis system, and an experimental platform.

[0105] A fluorescence microscopy system is used to acquire calcium imaging data of brain neurons. The fluorescence microscopy system disclosed in this invention employs a dual-light source configuration, including a 450nm LED light source for exciting GCaMP fluorescence signals and a 405nm LED light source for recording hemodynamic signals and other calcium-independent artifacts. Both LED light sources are precisely time-controlled and intensity-adjusted via an Arduino controller. This dual-light source design can effectively remove interference signals in subsequent data analysis, improving the accuracy of calcium signal recording and analysis. The objective lens configuration can be quickly changed according to different experimental needs. The 50mm f1.2 camera objective lens is suitable for macroscopic calcium imaging of the entire cerebral cortex in vivo or the entire slice ex vivo. Its ultra-large field of view allows for observation of overall brain region activity, but the spatial resolution cannot reach the single-cell level. The 1×0.125NA objective lens maintains a large field of view (8×11mm) while possessing single-cell resolution capability, making it particularly suitable for simultaneous recording of large-scale cross-brain region neuronal population activity. The 5×0.5NA objective lens achieves the best balance between field of view (5×5mm) and spatial resolution, making it suitable for single-cell resolution synchronous recording of large-scale neuronal populations in single or adjacent brain regions.

[0106] The calcium imaging data analysis system uses the above method to process multiple consecutive frames of brain nerve images, including an image segmentation unit, a displacement vector calculation unit, and an image registration unit.

[0107] The image segmentation unit is used to create a fixed reference frame image from a sequence of multiple consecutive brain neural image frames, and select one frame image from the sequence of multiple consecutive brain neural image frames one by one as the current frame image to be processed, dividing the reference frame image and the current frame image into sub-regions with the same preset number, the same preset position and the same preset size and overlapping each other.

[0108] The displacement vector calculation unit is used to calculate the displacement vector of each sub-region and determine whether all displacement vectors are consistent.

[0109] The image registration unit is used to call different registration algorithms to register brain nerve images of multiple consecutive frames according to the output result of the displacement vector calculation unit. If the output result of the displacement vector calculation unit is yes, it is determined that the image has a global rigid displacement, and the image registration algorithm based on cerebral blood vessel feature point matching is used to correct the motion artifacts of the current frame image. If the output result of the displacement vector calculation unit is no, it is determined that there is a local non-rigid deformation, and the NorMCorre algorithm based on the non-rigid motion model is used to correct the motion artifacts of the previous frame image.

[0110] The experimental platform is used to hold samples and provide the experimental environment. It generally includes a head fixation device and an environmental parameter control unit. The head fixation device is used to fix the brain of a live animal, and the environmental parameter control unit is used to control various parameters of the experimental environment, such as temperature, humidity, and lighting conditions. Both the head fixation device and the environmental parameter control unit are existing technologies and will not be described in detail here.

[0111] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for wide-field single-photon calcium imaging of brain neurons, characterized in that, Includes the following steps: A fixed reference frame image is created from a sequence of multiple consecutive brain neural images, and the reference frame image is divided into a preset number of sub-regions with preset positions and preset sizes that overlap with each other. From the sequence of consecutive brain neural images, one frame is selected one by one as the current frame image to be processed. The current frame image is divided into sub-regions that have the same number, the same position, and the same size as the reference frame image and also overlap with each other. The displacement vector of each sub-region of the current frame image relative to the corresponding sub-region on the reference frame image is calculated using the optical flow method. If all displacement vectors are consistent, it is determined that the image has undergone global rigid displacement, and an image registration algorithm based on cerebral blood vessel feature point matching is used to correct the motion artifacts of the current frame image. If not, it is determined that there is local non-rigid deformation, and the NorMCorre algorithm based on non-rigid motion model is used to correct the motion artifacts of the current frame image.

2. The method according to claim 1, characterized in that, The following steps are used to determine whether all the displacement vectors are consistent: Construct a displacement vector field matrix by arranging all the displacement vectors according to their spatial positional relationships; Calculate the average displacement vector of all the aforementioned displacement vectors; Subtract the average displacement vector from each displacement vector in the displacement vector field matrix to obtain the displacement difference field matrix, and calculate the Frobenius norm of the displacement difference field matrix. If the Frobenius norm of the displacement difference field matrix is ​​less than the first threshold, then all the displacement vectors are consistent.

3. The method according to claim 1, characterized in that, Image registration using cerebral vascular features is performed through the following steps: Select one frame from the sequence of multiple consecutive brain neural images as a fixed image; One frame is selected from the continuous multi-frame brain neural image sequence as a floating image; Extract cerebral vascular features from the fixed image and construct a SIFT descriptor for the fixed image; extract cerebral vascular features from the floating image and construct a SIFT descriptor for the floating image. Based on the SIFT descriptor of the fixed image and the SIFT descriptor of the floating image, coarse registration is performed on the fixed image and the floating image; The fixed image and the floating image are precisely registered using a B-spline free deformation model.

4. The method according to claim 3, characterized in that, The cerebral vascular features of the fixed image and the floating image are extracted through the following steps: A vascular network topology map of the fixed image or the floating image is generated using a Frangi filter combined with a skeletonization algorithm. Extract the vascular network bifurcation points from the vascular network topology diagram.

5. The method according to claim 4, characterized in that, The SIFT descriptor for the fixed image is constructed using the following steps: Determine whether there are vascular network bifurcation points in the fixed image. If so, set the vascular network bifurcation points as key feature points of the fixed image and construct the SIFT descriptor of the fixed image. If not, divide the blood vessels in the fixed image into at least two segments, and use the endpoints of each segment as key feature points to construct the SIFT descriptor of the fixed image.

6. The method according to claim 5, characterized in that, The SIFT descriptor of the floating image is constructed using the following steps: If there are vascular network bifurcation points in the fixed image, then the vascular network bifurcation points in the floating image are set as key feature points of the floating image, and a SIFT descriptor of the floating image is constructed. If there are no vascular network bifurcation points in the fixed image, then the blood vessels in the floating image are divided into at least two segments, and the endpoints of each segment are used as key feature points to construct a SIFT descriptor of the floating image.

7. The method according to claim 6, characterized in that, The coarse registration of the fixed image and the floating image also includes the following steps: Select a pair of matching key feature points of a fixed image and key feature points of a floating image, and calculate the forward transformation matrix from the key feature points of the floating image to the matching key feature points of the fixed image; Calculate the inverse transformation matrix from the key feature points of the fixed image to the matching key feature points of the floating image; Determine whether the forward transformation matrix and the inverse transformation matrix are consistent. If they are, then the key feature points of the fixed image and the key feature points of the floating image that are matched participate in the objective function of constructing the B-spline free deformation model.

8. The method according to claim 7, characterized in that, The fine registration of the fixed image and the floating image includes the following steps: The SIFT descriptors of the fixed image and the floating image are matched to initially align the fixed image and the floating image. The objective function is constructed by combining the confidence weight coefficient and the regularization term weight coefficient with the rigid transformation function and the second-order partial displacement field. The B-spline free deformation model is constructed based on the objective function, and then fine registration is performed.

9. A wide-field single-photon calcium imaging system for brain neurons, characterized in that, include: Fluorescence microscope module, calcium imaging data analysis module, and experimental platform; The fluorescence microscope module is used to acquire a sequence of consecutive multi-frame brain nerve images of the brain of a live animal. The calcium imaging data analysis module employs the method described in any one of claims 1 to 8 to process the continuous multi-frame brain neural images; The experimental platform is used to hold samples and provide an experimental environment. The experimental platform includes a head fixation device and an environmental parameter control unit. The head fixation device is used to fix the brain of a live animal, and the environmental parameter control unit is used to control various parameters of the experimental environment.

10. The system according to claim 9, characterized in that, The calcium imaging data analysis module includes an image segmentation unit, a displacement vector calculation unit, and an image registration unit; The image segmentation unit is used to create a fixed reference frame image from a sequence of multiple consecutive brain neural images, and select one frame image from the sequence of multiple consecutive brain neural images as the current frame image to be processed, and divide the reference frame image and the current frame image into sub-regions with the same preset number, the same preset position and the same preset size and which overlap with each other. The displacement vector calculation unit is used to calculate the displacement vector of each sub-region and determine whether all the displacement vectors are consistent. The image registration unit is used to call different registration algorithms to register the brain nerve images of the consecutive multiple frames according to the output result of the displacement vector calculation unit. If the output result of the displacement vector calculation unit is yes, it is determined that the image has a global rigid displacement, and the motion artifact of the current frame image is corrected by the image registration algorithm based on cerebral blood vessel feature point matching. If the output result of the displacement vector calculation unit is no, it is determined that there is a local non-rigid deformation, and the motion artifact of the current frame image is corrected by the NorMCorre algorithm based on the non-rigid motion model.

Images