Method and device for identifying neuronal cell bodies

Abstract

The present application provides a method and device for identifying neuronal cell bodies, comprising: obtaining a fluorescent video having multiple neuronal cell bodies; identifying the multiple neuronal cell bodies based on a trained second neuronal cell body identification and segmentation network to obtain a set of candidate cell body regions for the multiple neuronal cell bodies, wherein the candidate cell body region set includes multiple candidate cell body regions, and the candidate cell body regions include position information and boundary information of the neuronal cell bodies; performing bidirectional tracking and matching on the candidate cell body region set based on the order of the frames of the fluorescent video to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, wherein the independent mask sequence includes the mask of the neuronal cell body in any frame of the fluorescent video. The neuronal cell body identification method and device of the present application adopts a serial workflow of coarse positioning and fine identification, and can accurately and comprehensively identify neuronal cell bodies in the fluorescent video.

Description

Technical Field

[0001] The present application relates to the field of video processing, and more specifically, to a method and device for identifying neuron cell bodies, a computer-readable storage medium, and an electronic device. Background Art

[0002] Fluorescence imaging is currently one of the primary techniques used in neuroscience to study the dynamics of neural networks. In recent years, fluorescence imaging microscopes have been able to simultaneously record the activity of hundreds of neuronal cell bodies as mice move freely within their field of view. This has provided a crucial imaging tool for studying the encoding and decoding of signals from neuronal populations during behaviors such as learning, memory, social interaction, and fear.

[0003] However, during the processing and analysis of fluorescence videos, existing methods for identifying neuronal cell bodies have low accuracy. Summary of the Invention

[0004] In view of this, embodiments of the present application provide a method and device for identifying neuronal cell bodies, which can accurately and comprehensively identify neuronal cell bodies within a field of view.

[0005] In a first aspect, an embodiment of the present application provides a method for identifying neuronal cell bodies, comprising: obtaining a fluorescent video, the fluorescent video having multiple neuronal cell bodies; identifying the multiple neuronal cell bodies based on a trained second neuronal cell body identification and segmentation network, and obtaining a set of candidate cell body regions of the multiple neuronal cell bodies, the candidate cell body region set including multiple candidate cell body regions, the candidate cell body regions including position information and boundary information of the neuronal cell bodies; performing bidirectional tracking and matching on the candidate cell body region set according to the order of frames of the fluorescent video, and obtaining an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, wherein the independent mask sequence includes the mask of the neuronal cell body in any frame in the fluorescent video.

[0006] In certain embodiments of the present application, the duration of the fluorescence video is T, and according to the order of the frames of the fluorescence video, a set of candidate cell body regions is bidirectionally tracked and matched to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, including: S131: calculating the first matching distance between the first candidate cell body region in the candidate cell body region set and the first template, the first candidate cell body region includes the candidate cell body region of the t-th frame in the fluorescence video, and the first template includes the candidate cell body region corresponding to the t-1-th frame of the fluorescence video, where t≥2; S132: if the first matching distance is not greater than the second preset threshold, it is marked as a successful match, and the first template is updated and the first cumulative error is calculated. If the first matching distance is greater than the second preset threshold, it is marked as a failed match, and the first candidate cell body region is added to the first template; S133: forward traverse all frames of the fluorescence video, iteratively execute S131 and S132, and obtain a forward mask sequence and multiple first cumulative errors, the forward mask sequence includes the first candidate cell body region that is successfully matched.

[0007] In certain embodiments of the present application, bidirectional tracking and matching is performed on a set of candidate cell body regions according to the order of frames of the fluorescent video to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, and also includes: S134: calculating a second matching distance between a second candidate cell body region in the candidate cell body region set and a second template, the second candidate cell body region including the candidate cell body region corresponding to the t-th frame in the fluorescent video, and the second template including the candidate cell body region corresponding to the t+1-th frame of the fluorescent video, where t≤T-1; S135: if the second matching distance is not greater than a second preset threshold, it is marked as a successful match, and the second template is updated and a second cumulative error is calculated. If the second matching distance is greater than the second preset threshold, it is marked as a failed match, and the second candidate cell body region is added to the second template; S136: reversely traverse all frames of the fluorescent video, iteratively execute S134 and S135, and obtain a reverse mask sequence and multiple second cumulative errors, and the reverse mask sequence includes the second candidate cell body region that is successfully matched.

[0008] In certain embodiments of the present application, bidirectional tracking and matching is performed on a set of candidate cell body regions according to the order of frames of the fluorescence video to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, which also includes: traversing the forward mask sequence and the reverse mask sequence, and retaining the mask sequence in which the forward mask sequence and the reverse mask sequence are the same as each other as the independent mask sequence; if the forward mask sequence and the reverse mask sequence are different, retaining the mask sequence with the smaller cumulative error between the forward mask sequence and the reverse mask sequence as the independent mask sequence.

[0009] In certain embodiments of the present application, the first matching distance is obtained based on the centroid distance and intersection-over-union ratio between the first candidate cell region and the first template, and the second matching distance is obtained based on the centroid distance and intersection-over-union ratio between the second candidate cell region and the second template.

[0010] In certain embodiments of the present application, before identifying multiple neuronal cell bodies according to a trained second neuronal cell body recognition and segmentation network to obtain a set of candidate cell body regions of multiple neuronal cell bodies, the recognition method of the first aspect also includes: S115: constructing a training set and obtaining manual annotation results of the training set, the training set includes images in the fluorescent video, and the manual annotation results include information of manually annotated neuronal cell bodies; S116: inputting the training set into the first neuronal cell body recognition and segmentation network to be trained to obtain a first recognition result of the neuronal cell body, and fitting the first recognition result through a super ellipse model to obtain a second recognition result of a super ellipse shape; S117: matching the second recognition result with the manual annotation result to obtain a third recognition result of the neuronal cell body after matching and a loss function, the loss function is used to update the parameters of the first neuronal cell body recognition and segmentation network, and the loss function includes a super ellipse shape loss function; S118: repeating S116 to S118 to obtain a second neuronal cell body recognition and segmentation network.

[0011] In certain embodiments of the present application, before identifying multiple neuron cell bodies based on a trained second neuron cell body identification and segmentation network to obtain a set of candidate cell body regions of the multiple neuron cell bodies, the identification method of the first aspect also includes: using a chaotic double oscillator model to perform desynchronization processing on the fluorescence video to obtain a desynchronized time series and obtaining a first nonlinear denoised video of the fluorescence video based on the desynchronized time series, and obtaining a baseline of the fluorescence video; obtaining an enhanced denoised video of the fluorescence video based on the first nonlinear denoised video of the fluorescence video and the baseline.

[0012] In the second aspect, an embodiment of the present application provides a device for identifying neuronal cell bodies, including: an acquisition module for acquiring a fluorescent video, the fluorescent video having multiple neuronal cell bodies; an identification module for identifying multiple neuronal cell bodies based on a trained second neuronal cell body identification and segmentation network, and obtaining a set of candidate cell body regions of multiple neuronal cell bodies, the candidate cell body region set including multiple candidate cell body regions, the candidate cell body regions including position information and boundary information of the neuronal cell bodies; a matching module for performing bidirectional tracking and matching on the candidate cell body region set according to the order of frames of the fluorescent video, and obtaining an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, wherein the independent mask sequence includes the mask of the neuronal cell body in any frame in the fluorescent video.

[0013] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute the neuron cell body identification method according to any one of the first aspects.

[0014] In a fourth aspect, an embodiment of the present application provides an electronic device comprising: a processor; and a memory for storing processor-executable instructions, wherein the processor is configured to execute the neuron cell body identification method of any one of the first aspects.

[0015] The neuron cell body identification method and device of the present application adopt a serial workflow of coarse positioning and fine identification, that is, the neuron cell body is first coarsely identified by using a neuron cell body identification and segmentation network to obtain a set of candidate cell body regions, and then the candidate cell body region set is finely identified by using two-way tracking and matching. It can accurately and comprehensively identify neuron cell bodies in fluorescent videos, which is of great significance for subsequent related analysis of neurons. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of a method for identifying neuronal cell bodies provided in one embodiment of the present application.

[0017] Figure 2 A schematic diagram of a non-overlapping forward mask sequence provided in accordance with an embodiment of the present application.

[0018] Figure 3 A schematic diagram of a non-overlapping reverse mask sequence provided in accordance with an embodiment of the present application.

[0019] Figure 4 A schematic diagram of an overlapping forward mask sequence provided in accordance with another embodiment of the present application.

[0020] Figure 5 A schematic diagram of an overlapping reverse mask sequence provided in accordance with another embodiment of the present application.

[0021] Figure 6 A flowchart of a forward tracking and matching method provided in one embodiment of the present application.

[0022] Figure 7 A flowchart of a reverse tracking and matching method provided in one embodiment of the present application.

[0023] Figure 8 A flowchart of a method for obtaining an independent mask sequence provided in one embodiment of the present application.

[0024] Figure 9 A flowchart of a training method for a neuron cell body recognition and segmentation network provided in one embodiment of the present application.

[0025] Figure 10A flowchart of a method for enhancing fluorescence video provided in one embodiment of the present application is shown.

[0026] Figure 11 A schematic diagram of the structure of a neuron cell body identification device provided in one embodiment of the present application.

[0027] Figure 12 A block diagram of an electronic device for executing a method for identifying neuron cell bodies provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0028] The following describes embodiments of the present application in more detail with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are for illustrative purposes only and are not intended to limit the scope of protection of the present application.

[0029] As used herein, the term "including" and its variations are open-ended, i.e., "including but not limited to." The term "according to" means "at least in part according to." The term "one embodiment" means "at least one embodiment," and the term "another embodiment" means "at least one additional embodiment." Definitions of other terms are provided in the following description.

[0030] Fluorescence imaging is currently one of the primary techniques used in neuroscience to study the dynamics of neuronal networks. After capturing fluorescent video of neurons using a fluorescence imaging microscope, it is possible to identify neuronal cell bodies and extract their signal curves, thereby analyzing the dynamics of individual and group neurons.

[0031] In recent years, fluorescence imaging microscopes have been able to simultaneously record the activity signals of hundreds of neuronal cell bodies as mice move freely within their field of view, providing an important imaging tool for studying the neuronal population encoding and decoding involved in mouse behaviors such as learning, memory, social interaction, and fear. During the processing and analysis of fluorescence videos, accurately and comprehensively identifying neuronal cell bodies within the field of view is a key step in calcium imaging video processing, which is crucial for subsequent extraction of neuronal calcium signals, classification of neuronal subpopulations, and analysis of neuronal network dynamics.

[0032] However, existing methods for identifying neuronal somata during fluorescence video processing and analysis are suboptimal, leading to significant errors in subsequent neuron-related analyses. For example, the morphology of identified neuronal somata can differ significantly from their actual morphology, and it can be difficult to identify overlapping or adherent neuronal somata.

[0033] In order to solve the above problems, the present application provides a method for identifying neuronal cell bodies.

[0034] Figure 1 The flowchart of the neuron cell body identification method provided by one embodiment of the present invention is schematically shown. Figure 1 The method is performed by a computing device (e.g., a server), but the present invention is not limited thereto. The server may be a single server, or may be composed of several servers, or may be a virtualization platform, or may be a cloud computing service center, and the present invention is not limited thereto. Figure 1 As shown, the method includes the following contents.

[0035] S110: Acquire a fluorescent video, wherein the fluorescent video has multiple neuronal cell bodies.

[0036] Specifically, a fluorescence imaging microscope can be used to capture fluorescence videos, which can include activity signals of hundreds of neuronal cell bodies when a living organism (eg, a mouse) moves freely within the field of view.

[0037] S120: Identify multiple neuron cell bodies according to the trained second neuron cell body identification and segmentation network to obtain a set of candidate cell body regions for the multiple neuron cell bodies, where the set of candidate cell body regions includes multiple candidate cell body regions, and the candidate cell body regions include position information and boundary information of the neuron cell bodies.

[0038] Specifically, the second neuron cell body recognition and segmentation network can be a trained neuron cell body recognition model, such as a U-Net network or a Mask RCNN model. The fluorescent video is input into the second neuron cell body recognition and segmentation network to roughly locate the neuron cell body, and a set of candidate cell body regions is obtained. The candidate cell body region set includes multiple candidate cell body regions. For example, it can be set that a candidate cell body region contains only one neuron cell body. The candidate cell body region can include the position information of the neuron cell body. The position information of the neuron cell body can be represented by a rectangular box, and the morphology of the neuron cell body can be represented by a polygonal boundary within the rectangular box.

[0039] S130: performing bidirectional tracking and matching on the candidate cell body region set according to the sequence of the frames of the fluorescence video to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, wherein the independent mask sequence includes the mask of the neuron cell body in any frame of the fluorescence video.

[0040] Specifically, the order of the frames of the fluorescence video may include forward and reverse (or sequential and reverse order). Bidirectional tracking matching may include forward tracking matching and reverse tracking matching. Forward tracking matching may start from the first frame of the fluorescence video and track and match the neuronal cell body until the end of the last frame. Reverse tracking matching may start from the last frame of the fluorescence video and track and match the neuronal cell body until the end of the first frame. Through forward tracking matching and reverse tracking matching, the neuronal cell body is accurately identified to obtain an independent mask sequence of the candidate cell body region set. The independent mask sequence may include the mask of the neuronal cell body in any frame in the fluorescence video.

[0041] The neuron cell body identification method of the present application adopts a serial workflow of coarse positioning and fine identification, that is, the neuron cell body is first coarsely identified by using a neuron cell body identification and segmentation network to obtain a set of candidate cell body regions, and then the candidate cell body region set is finely identified by using two-way tracking matching. It can accurately and comprehensively identify neuron cell bodies in fluorescent videos, which is of great significance for subsequent neuron calcium signal extraction, neuron subpopulation division, and neuron network dynamics analysis.

[0042] In one embodiment, if Figure 6 As shown, the duration of the fluorescence video is T. According to the order of the frames of the fluorescence video, the candidate cell body region set is bidirectionally tracked and matched to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, including the following contents.

[0043] S131: Calculate a first matching distance between a first candidate cell body region in the candidate cell body region set and the first template, where the first candidate cell body region includes the candidate cell body region of the t-th frame in the fluorescence video, and the first template includes the candidate cell body region corresponding to the t-1-th frame of the fluorescence video, where t≥2.

[0044] Specifically, at the beginning, the first candidate cell body region identified in the t-1 frame (e.g., t=2, i.e., the 1st frame) of the fluorescence video can be used as the first template. For each frame in the video, the first matching distance L between each first candidate cell body region identified in the tth frame (e.g., t=2, i.e., the 2nd frame) and each first template is calculated. p1 For example, the first matching distance L p1 It can be obtained based on the centroid distance and intersection-union ratio between the first candidate cell region and the first template, for example, it can be obtained by the following formula: p1 =L C1 +λL IoU1 , where L C1 is the centroid distance between the two cell body regions (i.e., the first candidate cell body region and the first template), L IoU1 is the intersection-over-union ratio of the two cell body regions, and λ is a hyperparameter.

[0045] S132: If the first matching distance is not greater than the second preset threshold, it is marked as a successful match, and the first template is updated and the first cumulative error is calculated. If the first matching distance is greater than the second preset threshold, it is marked as a failed match, and the first candidate cell body region is added to the first template.

[0046] Specifically, the first matching distance L can be p1 The cell body pair (i.e., the first candidate cell body region and the first template) that is less than or equal to the preset threshold is considered to be tracked and matched successfully, the template is updated and the first cumulative error of the cell body pair is calculated. p1 Cell body pairs greater than a preset threshold are considered tracking matches failed, and the first candidate cell body region that failed tracking matches in frame t is added as a new template to the first template. For example, the template can be updated by replacing the first template that successfully matched in frame t-1 with the first candidate cell body region in frame t, or by updating the first template based on the average mask value of the first candidate cell body region and the first template.

[0047] S133: forwardly traverse all frames of the fluorescence video, iteratively execute S131 and S132, and obtain a forward mask sequence and a plurality of first accumulated errors, wherein the forward mask sequence includes the first candidate cell body region that is successfully matched.

[0048] Specifically, all time frames of the fluorescence video can be traversed forward, S131 and S132 can be iteratively executed, and the first candidate cell body region that is successfully matched is used as the mask of the neuron cell body and arranged in the order of the frames to obtain the forward mask sequence of the neuron cell body. and the first cumulative error E1 of each independent neuron cell body = {E1 m ,m=1,…,n m1}, where n m1 is the number of independent neuron somata that are positively matched.

[0049] This embodiment forwardly traverses the fluorescence video and performs forward tracking and matching on the candidate cell body region set, thereby obtaining a forward mask sequence and a first cumulative error of the candidate cell body region set.

[0050] In one embodiment, if Figure 7 As shown, according to the order of the frames of the fluorescence video, the candidate cell body region set is bidirectionally tracked and matched to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, and also includes the following contents.

[0051] S134: Calculate a second matching distance between a second candidate cell body region in the candidate cell body region set and the second template, where the second candidate cell body region includes the candidate cell body region corresponding to the t-th frame in the fluorescence video, and the second template includes the candidate cell body region corresponding to the t+1-th frame of the fluorescence video, where t≤T-1.

[0052] Specifically, at the beginning, the second candidate cell body region identified in the t+1th frame (e.g., t=T-1, i.e., the Tth frame) of the fluorescence video can be used as the second template. For each frame in the video, the second matching distance L between each second candidate cell body region identified in the tth frame (e.g., t=T-1, i.e., the T-1th frame) and each second template is calculated. p2 For example, the second matching distance L p2 It can be obtained based on the centroid distance and intersection-union ratio between the second candidate cell region and the second template, for example, it can be obtained by the following formula: p2 =L C2 +λL IoU2 , where L C2 is the centroid distance between the two cell body regions (i.e., the second candidate cell body region and the second template), L IoU2 is the intersection-over-union ratio of the two cell body regions, and λ is a hyperparameter.

[0053] S135: If the second matching distance is not greater than the second preset threshold, it is marked as a successful match, and the second template is updated and the second cumulative error is calculated. If the second matching distance is greater than the second preset threshold, it is marked as a failed match, and the second candidate cell body region is added to the second template.

[0054] Specifically, the second matching distance L can be p2 The cell body pair (i.e., the second candidate cell body region and the second template) that is less than or equal to the preset threshold is considered to be tracked and matched successfully, the template is updated and the second cumulative error of the cell body pair is calculated. p2 Cell body pairs with a value greater than a preset threshold are considered to have failed tracking and matching, and the second candidate cell body region that failed tracking and matching in frame t is added as a new template to the second template. For example, the template can be updated by replacing the second template that successfully matched in frame t+1 with the second candidate cell body region in frame t, or by updating the second template based on the average mask value of the second candidate cell body region and the second template.

[0055] S136: traverse all frames of the fluorescence video in reverse order, iteratively execute S134 and S135, and obtain a reverse mask sequence and a plurality of second accumulated errors. The reverse mask sequence includes the second candidate cell body region that has been successfully matched.

[0056] Specifically, all time frames of the fluorescence video can be traversed in reverse, S134 and S135 can be iteratively executed, and the second candidate cell body region that is successfully matched is used as the mask of the neuron cell body and arranged in the order of the frames to obtain the reverse mask sequence of the neuron cell body. and the second cumulative error E2 of each independent neuron cell body = {E2 m ,m=1,…,n m2}, where n m2 is the number of independent neuronal cell bodies that are matched in reverse order.

[0057] This embodiment reversely traverses the fluorescence video and performs reverse tracking and matching on the candidate cell body region set, thereby obtaining a reverse mask sequence and a second cumulative error of the candidate cell body region set.

[0058] In one embodiment, if Figure 8 As shown, according to the order of the frames of the fluorescence video, the candidate cell body region set is bidirectionally tracked and matched to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, and also includes the following contents.

[0059] S137: Traverse the forward mask sequence and the reverse mask sequence, and retain the mask sequence with the same forward mask sequence and reverse mask sequence as an independent mask sequence.

[0060] Specifically, for neuronal cell bodies without overlap or adhesion in the fluorescence video, the mask sequence with the same forward mask sequence and reverse mask sequence can be retained as an independent mask sequence. Figure 2 is a schematic diagram of a forward mask sequence without overlap. Figure 2 As shown, in a certain field of view of the fluorescent video, there is no neuron soma in the field in the t-1 frame, and two neuron soma appear in the field in the t frame. Mask 211 and mask 221 are the candidate soma regions identified in the t frame, and mask 212 and mask 222 are the candidate soma regions identified in the t+1 frame. After forward tracking and matching, a forward mask sequence of the neuron soma is obtained. The forward mask sequence of the neuron soma includes the forward mask of the neuron soma in any frame. For example, one forward mask sequence includes mask 211 and mask 212, and the other forward mask sequence includes mask 221 and mask 222.

[0061] Figure 3 For Figure 2 Schematic diagram of a non-overlapping reverse mask sequence for the same field of view. Figure 3As shown, in this field of view, there is no neuron soma in the t-1th frame, and two neuron soma appear in the tth frame. After reverse tracking and matching, a reverse mask sequence of the neuron soma is obtained. For example, one reverse mask sequence includes mask 311 and mask 312, and the other reverse mask sequence includes mask 321 and mask 322.

[0062] like Figure 2 and Figure 3 As shown in Figure 2, when there is no overlap between the neuron cell bodies, the positions of the forward mask sequence and the reverse mask sequence are the same at every moment. Therefore, for neuron cell bodies that do not overlap or adhere to each other, the mask sequence with the same forward mask sequence and reverse mask sequence can be retained as an independent mask sequence.

[0063] S138: If the forward mask sequence and the reverse mask sequence are different, retain the mask sequence with the smaller cumulative error between the forward mask sequence and the reverse mask sequence as an independent mask sequence.

[0064] Specifically, for overlapping or adhered neuronal cell bodies in the fluorescence video, the mask sequence with the smaller cumulative error between the forward mask sequence and the reverse mask sequence can be retained as an independent mask sequence. Figure 4 is a schematic diagram of a forward mask sequence with overlap. Figure 4 As shown, in another field of view of the fluorescent video, there are no neuron somas in the t-1 frame. In frame t, due to the partial overlap of neuron regions, it is very likely that the incorrectly identified mask 411 will be used as a candidate soma region for frame t identification and as a new template during the recognition process. Mask 412 and mask 422 are candidate soma regions for frame t+1 identification. In this case where there is overlap of neuron somas, after forward tracking and matching, the forward mask sequence of the incorrect neuron soma mask 411 is mask 412 or mask 422, resulting in both the recording of incorrect soma identification mask information in the template and the generation of an incorrect matching sequence.

[0065] Figure 5 For Figure 4 Schematic diagram of overlapping reverse mask sequences for the same field of view. Figure 5As shown, in this field of view, there are no neuronal cell bodies in the t-1 frame. In the t frame, the incorrectly identified mask 502 is obtained, and the correctly identified masks 511 and 521 are obtained in the t+1 frame. During reverse tracking and matching, the masks 511 and 521 of the t+1 frame are first used as new templates. When matching the t frame, because the shape of mask 502 is similar to the shape of the combined masks 511 and 521 in the template, when updating the template in the t frame, the incorrectly identified mask 502 is divided into masks 512 and 522 based on masks 511 and 521, and the reverse matching sequence is updated. The t frame can be matched based on the intersection-and-union (IOU) between the masks.

[0066] like Figure 2 and Figure 3 As shown in the figure, when neuronal cell bodies overlap, the positions of the forward and reverse mask sequences at a certain moment are different. Therefore, for overlapping or adherent neuronal cell bodies, the mask sequence with the smaller cumulative error between the forward and reverse mask sequences can be retained as the independent mask sequence.

[0067] This embodiment can identify neuronal cell bodies with overlap and adhesion in the fluorescent video by comparing the difference between the forward mask sequence and the reverse mask sequence, thereby improving the accuracy of neuronal cell body identification.

[0068] In one embodiment, if Figure 9 As shown, before identifying multiple neuron cell bodies based on the trained second neuron cell body identification and segmentation network to obtain a set of candidate cell body regions of multiple neuron cell bodies, the following contents are also included.

[0069] S116: Construct a training set and obtain manual annotation results of the training set, where the training set includes images in the fluorescent video, and the manual annotation results include information of manually annotated neuronal cell bodies.

[0070] Specifically, the training set may include a plurality of images randomly selected from a plurality of fluorescence videos and manual annotation results of corresponding neuronal cell bodies, and the manual annotation results may include a segmentation result R1 of the neuronal cell body.

[0071] S117: Input the training set into the first neuron cell body recognition and segmentation network to be trained to obtain a first recognition result of the neuron cell body, and fit the first recognition result through a super ellipse model to obtain a second recognition result of the super ellipse shape.

[0072] Specifically, N images can be randomly extracted from the training set and input into the first neuron cell body recognition and segmentation network to be trained to obtain a first recognition result. The first neuron cell body recognition and segmentation network can be a U-Net network or a MaskRCNN model. The first recognition result can include a first bounding box recognition result of the neuron cell body. and the first instance segmentation result Where ni is the number of neuron somata identified in the i-th image.

[0073] The first bounding box recognition result P B Multiple bounding boxes can be included, and a bounding box can contain only one neuron cell body. The shape of the bounding box can be rectangular, and the coordinates of the four vertices can be used to represent the bounding box.

[0074] The first instance segmentation result P S The first bounding box recognition result P can be B The binary mask image contains the boundary information of the neuron cell body. The first instance segmentation result P S The second recognition result of the super-ellipse shape of the neuron cell body is obtained by fitting the super-ellipse model. The second recognition result can be the segmentation result of the first super-ellipse shape instance.

[0075] The superelliptical model can be composed of the following six equations:

[0076] Standard ellipse parametric equation:

[0077]

[0078] Translation equation:

[0079]

[0080] Rotation equation:

[0081]

[0082] Convex equation:

[0083]

[0084] Concave equation:

[0085]

[0086] in

[0087] Asymmetric equation:

[0088]

[0089] The equations contain 8 unknown parameters, namely: the semi-major axis parameter a of the ellipse x , the semi-minor axis parameter a of the ellipse y , the shape parameter ε of the ellipse, the translation parameter l on the x-axis x , translation parameter l on the y-axis y , rotation angle r, convex parameter t, concave parameter b and asymmetric parameter ω, the first instance segmentation result P is obtained by the least squares method. S Fitting is performed to obtain the second recognition result of the super-elliptical morphology of the neuron cell body.

[0090] The manual annotation results may also include the first segmentation results of the manually annotated neuron cell bodies corresponding to N images and the first segmentation result R S The corresponding first bounding box annotation result where n′ i is the number of manually annotated neuron cell bodies in the i-th image.

[0091] S118: Match the second recognition result with the manual labeling result to obtain a third recognition result of the matched neuron cell body and a loss function, the loss function is used to update the parameters of the first neuron cell body recognition and segmentation network, and the loss function includes a superellipse morphology loss function.

[0092] Specifically, the first superellipse instance segmentation result P can be segmented based on the non-maximum suppression and the intersection-over-union ratio index of the first preset threshold. O With the first segmentation result R S Matching is performed, and the first bounding box recognition result P B And the first bounding box annotation result R B Perform matching to obtain the third recognition result and loss function L.

[0093] The third recognition result may include the second bounding box recognition result of the matched neuron cell body Second bounding box annotation result The corresponding second instance segmentation result PS*=PS*i,j,i=1,…,N,j=1,…ni*, the second segmentation result of the manually labeled neuron cell body And the second superellipse instance segmentation result in is the number of bounding boxes of matched neuron somata.

[0094] The loss function L is used to update the parameters of the first neuron cell body recognition segmentation network, which can include the bounding box number loss function Bounding box location loss function Mask loss function And the superelliptical morphological loss function Loss function L = αL BN +L B +L S +βL o Used to update the parameters of the first neuron cell body recognition and segmentation network. The update method can be batch gradient descent, where α and β are hyperparameters.

[0095] S119: Repeat S116 to S118 to obtain a second neuron cell body recognition and segmentation network.

[0096] After setting the loss function L, steps S116 to S118 are repeatedly executed, and the first neuron cell body recognition and segmentation network is updated to obtain a second neuron cell body recognition and segmentation network.

[0097] In this embodiment, the first instance segmentation result P S The first superellipse morphology instance segmentation result P of the neuron cell body is obtained by fitting the superellipse model. O , which smoothes the boundaries of neuron cell bodies and makes the identified neuron cell bodies closer to the actual morphology of neurons.

[0098] In one embodiment, the first matching distance is obtained based on the centroid distance and the intersection-over-union ratio between the first candidate cell region and the first template, and the second matching distance is obtained based on the centroid distance and the intersection-over-union ratio between the second candidate cell region and the second template.

[0099] The calculation method of the first matching distance and the second matching distance can be found in the description of the above embodiment and will not be repeated here.

[0100] In one embodiment, before identifying multiple neuron somas based on the trained second neuron soma identification and segmentation network to obtain a set of candidate soma regions of the multiple neuron somas, the following contents are also included.

[0101] S111: performing desynchronization processing on the fluorescence video using a chaotic double oscillator model to obtain a desynchronization time series, obtaining a first nonlinear denoising video of the fluorescence video according to the desynchronization time series, and obtaining a baseline of the fluorescence video.

[0102] Specifically, the time series of each pixel position in the fluorescence video can be used as an external perturbation term and applied to one of the oscillators in the double oscillator model to obtain the first desynchronized time series of the fluorescence video. The chaotic double oscillator model can be a Duffing chaotic double oscillator model, which is expressed as the following formula.

[0103]

[0104] ω is the frequency of the periodic driving force, ξ is the damping coefficient, k is the coupling coefficient, F is the amplitude of the periodic driving force, and s(t) is the input perturbation force. The periodic driving force is also called the periodic driving force, and the perturbation force is also called the external perturbation term. The time series of each pixel position in the fluorescence video is used as the external perturbation term and applied to one oscillator in the two-oscillator Duffing chaos model. No external perturbation term is applied to the other oscillator, thereby obtaining the first desynchronization time series. The first desynchronization time series can detect fast transients and slow decaying signals in the video, such as scintillation events (also called "calcium transient events" in calcium imaging fluorescence videos). Desynchronization refers to the loss of synchronization between two oscillators when they are in the same potential well in a period-doubling state. In a broader sense, desynchronization refers to the loss of synchronization between two oscillators in various states, such as period-doubling, chaos, and large-scale periodicity. The driving force periodic term is introduced during the time scale transformation of the Duffing chaos model to facilitate the detection of periodic signals of various frequencies. The time series of each pixel position refers to a sequence in which the pixel values ​​of the pixel on each video frame are arranged in chronological order. For example, it can be a plurality of pixel values ​​represented in the form of a vector.

[0105] After obtaining the first out-of-sync time series, the first out-of-sync time series is adjusted within a dynamic range to obtain a first nonlinear denoised video. The "dynamic range adjustment" may be performed by linearly amplifying the amplitude range of the out-of-sync time series to the amplitude range of the time series at the corresponding pixel position. The adjusted first out-of-sync time series is the first nonlinear denoised video.

[0106] S112: Obtain an enhanced denoised video of the fluorescence video according to the first nonlinear denoised video of the fluorescence video and the baseline.

[0107] Specifically, the baseline can be estimated by fitting a model to the signal, for example using the MLspike algorithm. MLspike is an algorithm that reconstructs neuronal spike activity from noisy calcium recordings and can be used in MATLAB to estimate the baseline.

[0108] Optionally, the fluorescence video can be randomly shuffled in time to obtain a disordered video of the fluorescence video, and then the disordered video can be desynchronized using a chaotic double oscillator model to obtain a second desynchronized time series, and a second nonlinear denoised video of the disordered video can be obtained based on the second desynchronized time series, and the envelope of the second nonlinear denoised video can be calculated to obtain the baseline of the fluorescence video.

[0109] Optionally, the fluorescence video may be rearranged in a manner of generating non-repeating frame numbers using a uniform pseudo-random number generator, thereby achieving random shuffling by time.

[0110] Optionally, the time series of each pixel position in the out-of-order video can be applied as an external perturbation term to one of the oscillators in the dual-oscillator model to obtain a second out-of-synchronization time series of the out-of-order video. The chaotic dual-oscillator model can be a Duffing chaotic dual-oscillator model. The time series of each pixel position refers to a sequence of pixel values ​​of that pixel on each video frame arranged in chronological order. For example, it can be a plurality of pixel values ​​represented in the form of a vector.

[0111] After dynamic range adjustment of the second out-of-sync time series, the average of its lower envelope serves as the baseline of the fluorescence video signal when a flicker event occurs. When no flicker event occurs, the average of the upper and lower envelopes of the second out-of-sync time series after dynamic range adjustment can be used as the baseline of the fluorescence video signal when no flicker event occurs. Dynamic range adjustment can be performed by linearly amplifying the amplitude range of the second out-of-sync time series to the amplitude range of the time series at the corresponding pixel location.

[0112] This embodiment performs desynchronization processing on the fluorescence video based on the chaotic double oscillator model and combines the baseline of the video to effectively enhance the fluorescence video and improve the signal-to-noise ratio of the video.

[0113] Figure 11 A schematic structural diagram of a neuron cell body identification device provided in one embodiment of the present application includes the following contents.

[0114] The acquisition module 1110 is configured to acquire a fluorescent video having a plurality of neuronal cell bodies.

[0115] The identification module 1120 is used to identify multiple neuronal cell bodies based on the trained second neuronal cell body identification and segmentation network to obtain a set of candidate cell body regions of the multiple neuronal cell bodies. The candidate cell body region set includes multiple candidate cell body regions, and the candidate cell body regions include position information and boundary information of the neuronal cell bodies.

[0116] The matching module 1130 is used to perform bidirectional tracking matching on the candidate cell body region set according to the order of the frames of the fluorescence video to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking matching, wherein the independent mask sequence includes the mask of the neuronal cell body in any frame of the fluorescence video.

[0117] The neuron cell body identification device of the present application adopts a serial workflow of coarse positioning and fine identification, that is, the neuron cell body is first coarsely identified by using a neuron cell body identification and segmentation network to obtain a set of candidate cell body regions, and then the candidate cell body region set is finely identified by using two-way tracking matching. It can accurately and comprehensively identify neuron cell bodies in fluorescent videos, which is of great significance for subsequent neuron calcium signal extraction, neuron subpopulation division, and neuron network dynamics analysis.

[0118] According to an embodiment of the present application, the duration of the fluorescence video is T, and the matching module 1130 executes S131: calculating the first matching distance between the first candidate cell body region in the candidate cell body region set and the first template, the first candidate cell body region includes the candidate cell body region of the tth frame in the fluorescence video, and the first template includes the candidate cell body region corresponding to the t-1th frame of the fluorescence video, where t≥2; S132: if the first matching distance is not greater than the second preset threshold, it is marked as a successful match, and the first template is updated and the first cumulative error is calculated. If the first matching distance is greater than the second preset threshold, it is marked as a failed match, and the first candidate cell body region is added to the first template; S133: forward traversing all frames of the fluorescence video, iteratively executing S131 and S132, and obtaining a forward mask sequence and multiple first cumulative errors. The forward mask sequence includes the first candidate cell body region that is successfully matched.

[0119] According to an embodiment of the present application, the matching module 1130 also executes: S134: calculating the second matching distance between the second candidate cell body region in the candidate cell body region set and the second template, the second candidate cell body region includes the candidate cell body region corresponding to the t-th frame in the fluorescence video, and the second template includes the candidate cell body region corresponding to the t+1-th frame of the fluorescence video, where t≤T-1; S135: if the second matching distance is not greater than the second preset threshold, it is marked as a successful match, and the second template is updated and the second cumulative error is calculated. If the second matching distance is greater than the second preset threshold, it is marked as a failed match, and the second candidate cell body region is added to the second template; S136: reversely traverse all frames of the fluorescence video, iteratively execute S134 and S135, and obtain a reverse mask sequence and multiple second cumulative errors. The reverse mask sequence includes the second candidate cell body region that is successfully matched.

[0120] According to an embodiment of the present application, the matching module 1130 traverses the forward mask sequence and the reverse mask sequence, and retains the mask sequence with the same forward mask sequence and the reverse mask sequence as an independent mask sequence; if the forward mask sequence and the reverse mask sequence are different, the mask sequence with the smaller cumulative error between the forward mask sequence and the reverse mask sequence is retained as an independent mask sequence.

[0121] According to an embodiment of the present application, the first matching distance is obtained based on the centroid distance and intersection-over-union ratio between the first candidate cell body region and the first template, and the second matching distance is obtained based on the centroid distance and intersection-over-union ratio between the second candidate cell body region and the second template.

[0122] According to an embodiment of the present application, the recognition module 1120 also executes: S116: constructing a training set and obtaining the manual annotation results of the training set, the training set includes images in the fluorescent video, and the manual annotation results include information of manually annotated neuronal cell bodies; S117: inputting the training set into the first neuronal cell body recognition and segmentation network to be trained to obtain a first recognition result of the neuronal cell body, and fitting the first recognition result through a super ellipse model to obtain a second recognition result of a super ellipse shape; S118: matching the second recognition result with the manual annotation result to obtain a third recognition result of the neuronal cell body after matching and a loss function, the loss function is used to update the parameters of the first neuronal cell body recognition and segmentation network, and the loss function includes a super ellipse shape loss function; S119: repeating S116 to S118 to obtain a second neuronal cell body recognition and segmentation network.

[0123] According to an embodiment of the present application, the neuron cell body identification device also includes an enhancement module, which uses a chaotic double oscillator model to perform desynchronization processing on the fluorescence video to obtain a desynchronized time series and obtain a first nonlinear denoised video of the fluorescence video based on the desynchronized time series, and obtains the baseline of the fluorescence video; based on the first nonlinear denoised video of the fluorescence video and the baseline, an enhanced denoised video of the fluorescence video is obtained.

[0124] The specific definition of the neuron cell body identification device can be found in the above-mentioned definition of the neuron cell body identification method, which will not be repeated here.

[0125] Figure 12 FIG2 is a block diagram of an electronic device 1200 for executing a method for identifying neuronal cell bodies provided by an exemplary embodiment of the present application, including a processor 1210 and a memory 1220 .

[0126] The memory 1220 is configured to store the processor executable instructions. The processor 1210 is configured to run the executable instructions to execute the neuron cell body identification method according to any one of the above embodiments.

[0127] The present application also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute the neuron cell body identification method according to any one of the above embodiments.

[0128] The neuron cell body identification method and device of the present application adopt a serial workflow of coarse positioning and fine identification, that is, the neuron cell body is first coarsely identified by using a neuron cell body identification and segmentation network to obtain a set of candidate cell body regions, and then the candidate cell body region set is finely identified by using two-way tracking and matching. It can accurately and comprehensively identify neuron cell bodies in fluorescent videos, which is of great significance for subsequent neuron calcium signal extraction, neuron subpopulation division, and neuron network dynamics analysis.

[0129] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any other combination. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated therein. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a digital video disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0130] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0131] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0132] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0133] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0134] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for identifying neuronal cell bodies, characterized in that: include: Acquiring a fluorescent video, wherein the fluorescent video has a plurality of neuronal cell bodies; Identifying the plurality of neuron somas according to a trained second neuron soma identification and segmentation network to obtain a set of candidate soma regions for the plurality of neuron somas, wherein the set of candidate soma regions includes a plurality of candidate soma regions, and the candidate soma regions include position information and boundary information of the neuron somas; performing bidirectional tracking and matching on the set of candidate cell body regions according to the sequence of frames of the fluorescence video to obtain an independent mask sequence of the set of candidate cell body regions after bidirectional tracking and matching, wherein the independent mask sequence includes a mask of the neuronal cell body in any frame of the fluorescence video, and the bidirectional tracking and matching includes forward tracking matching and backward tracking matching, wherein the forward tracking matching is used to represent tracking and matching of candidate cell body regions in the set of candidate cell body regions in a direction from the first frame to the last frame of the fluorescence video; The reverse tracking and matching is used to represent tracking and matching of candidate cell body regions in the candidate cell body region set in a direction from the last frame of the fluorescence video to the first frame.

2. The identification method according to claim 1, characterized in that The duration of the fluorescence video is T. The bidirectional tracking and matching is performed on the candidate cell body region set according to the order of the frames of the fluorescence video to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching, including: S131: calculating a first matching distance between a first candidate cell body region in the set of candidate cell body regions and a first template, where the first candidate cell body region includes the candidate cell body region of the t-th frame in the fluorescence video, and the first template includes the candidate cell body region corresponding to the t-1-th frame of the fluorescence video, where t≥2; S132: If the first matching distance is not greater than a second preset threshold, marking the match as successful, updating the first template, and calculating a first cumulative error; if the first matching distance is greater than the second preset threshold, marking the match as failed, and adding the first candidate cell body region to the first template; S133: forwardly traverse all frames of the fluorescence video, iteratively execute S131 and S132, and obtain a forward mask sequence and a plurality of the first accumulated errors, wherein the forward mask sequence includes the first candidate cell body region that is successfully matched.

3. The identification method according to claim 2, characterized in that The step of performing bidirectional tracking and matching on the candidate cell body region set according to the sequence of frames of the fluorescence video to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching further includes: S134: calculating a second matching distance between a second candidate cell body region in the set of candidate cell body regions and a second template, where the second candidate cell body region includes the candidate cell body region of the t-th frame in the fluorescence video, and the second template includes the candidate cell body region corresponding to the t+1-th frame of the fluorescence video, where t≤T-1; S135: If the second matching distance is not greater than a second preset threshold, marking it as a successful match, updating the second template, and calculating a second cumulative error; if the second matching distance is greater than the second preset threshold, marking it as a failed match, and adding the second candidate cell body region to the second template; S136: traverse all frames of the fluorescence video in reverse order, iteratively execute S134 and S135, and obtain a reverse mask sequence and a plurality of second accumulated errors, wherein the reverse mask sequence includes the second candidate cell body regions that are successfully matched.

4. The identification method according to claim 3, characterized in that The step of performing bidirectional tracking and matching on the candidate cell body region set according to the sequence of frames of the fluorescence video to obtain an independent mask sequence of the candidate cell body region set after bidirectional tracking and matching further includes: Traversing the forward mask sequence and the reverse mask sequence, and retaining the mask sequence that is the same as the forward mask sequence and the reverse mask sequence as the independent mask sequence; If the forward mask sequence and the reverse mask sequence are different, the mask sequence with the smaller cumulative error between the forward mask sequence and the reverse mask sequence is retained as the independent mask sequence.

5. The identification method according to claim 3, characterized in that: The first matching distance is obtained based on the centroid distance and the intersection-and-union ratio between the first candidate cell body region and the first template, and the second matching distance is obtained based on the centroid distance and the intersection-and-union ratio between the second candidate cell body region and the second template.

6. The identification method according to claim 1, characterized in that Before identifying the plurality of neuron cell bodies according to the trained second neuron cell body identification and segmentation network to obtain a set of candidate cell body regions of the plurality of neuron cell bodies, the method further includes: S116: constructing a training set and obtaining manual annotation results of the training set, wherein the training set includes images in the fluorescence video, and the manual annotation results include information of manually annotated neuronal cell bodies; S117: Inputting the training set into a first neuron cell body recognition and segmentation network to be trained to obtain a first recognition result of the neuron cell body, and fitting the first recognition result through a superellipse model to obtain a second recognition result of a superellipse shape; S118: Matching the second recognition result with the manual labeling result to obtain a third recognition result of the neuron cell body after matching and a loss function, wherein the loss function is used to update the parameters of the first neuron cell body recognition and segmentation network, and the loss function includes a superellipse morphology loss function; S119: Repeat S116 to S118 to obtain the second neuron cell body recognition and segmentation network.

7. The identification method according to claim 1, characterized in that: Before identifying the plurality of neuron cell bodies according to the trained second neuron cell body identification and segmentation network to obtain a set of candidate cell body regions of the plurality of neuron cell bodies, the method further includes: Performing desynchronization processing on the fluorescence video using a chaotic double oscillator model to obtain a desynchronization time series, obtaining a first nonlinear denoised video of the fluorescence video based on the desynchronization time series, and acquiring a baseline of the fluorescence video; An enhanced denoised video of the fluorescence video is obtained according to the first nonlinear denoised video of the fluorescence video and the baseline.

8. A neuron cell body identification device, characterized in that: include: an acquisition module, configured to acquire a fluorescent video having a plurality of neuronal cell bodies; an identification module, configured to identify the plurality of neuronal cell bodies according to a trained second neuronal cell body identification and segmentation network, and obtain a set of candidate cell body regions for the plurality of neuronal cell bodies, wherein the set of candidate cell body regions includes a plurality of candidate cell body regions, and the candidate cell body regions include position information and boundary information of the neuronal cell bodies; a matching module, configured to perform bidirectional tracking matching on the set of candidate cell body regions according to the order of frames of the fluorescence video, to obtain an independent mask sequence of the set of candidate cell body regions after bidirectional tracking matching, wherein the independent mask sequence includes a mask of the neuronal cell body in any frame of the fluorescence video, the bidirectional tracking matching includes forward tracking matching and backward tracking matching, the forward tracking matching is used to represent tracking matching of candidate cell body regions in the set of candidate cell body regions in a direction from the first frame to the last frame of the fluorescence video; The reverse tracking and matching is used to represent tracking and matching of candidate cell body regions in the candidate cell body region set in a direction from the last frame of the fluorescence video to the first frame. 9 . A computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the neuron cell body identification method according to claim 1 .

10. An electronic device comprising: processor; a memory for storing instructions executable by the processor, The processor is configured to execute the neuron cell body identification method according to any one of claims 1 to 7.

Images