A neuron morphology image recognition system and method based on AI

Abstract

The present invention belongs to the field of neuron recognition technology, and specifically relates to an AI-based neuron morphology image recognition system and method. By constructing a multi-dimensional classification feature vector, the invention fully considers the morphological parameters and color feature information of neurons, thereby more comprehensively reflecting the characteristics of neurons. Based on this, neurons are accurately classified and identified. Through statistical analysis and offset processing, the most representative baseline feature differences can be determined, providing multi-dimensional data support for the training of machine learning models. This ensures that the recognition accuracy of subsequent machine learning models is correspondingly improved. The invention also demonstrates high stability and reliability when processing large-scale, complex neuron morphology images, thereby meeting the current demand for high-precision and high-efficiency neuron morphology image recognition.

Description

Technical Field

[0001] The present invention belongs to the field of neuron recognition technology, and specifically relates to an AI-based neuron morphology image recognition system and method. Background Art

[0002] With the deepening of neuroscience research and the development of microscopic imaging technology, it has become increasingly easier to obtain neuronal morphological images. Traditional neuron identification methods (which rely on manual observation and analysis) have gradually failed to meet the needs of large-scale, high-efficiency neuronal morphological image recognition. However, how to quickly and accurately identify different types of neurons from massive neuronal morphological images has become an urgent problem to be solved in the field of neuroscience research.

[0003] The existing neuron recognition methods mostly rely on the design of artificial features for recognition, or identify by specifying specific markers or areas in neurons. This is not only time-consuming and labor-intensive, but also difficult to guarantee recognition efficiency and accuracy. Especially when processing large-scale and complex neuron morphological images, the artificial feature design method often cannot meet the user's recognition needs. Based on this, the present invention proposes an AI-based neuron morphological image recognition method to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based neuron morphology image recognition system and method, which can automatically and quickly and accurately identify different types of neurons from massive neuron morphology images, thereby improving the efficiency and accuracy of neuron recognition.

[0005] The technical solutions adopted by the present invention are as follows:

[0006] An AI-based neuron morphology image recognition method, comprising:

[0007] Acquiring microscopic image data containing neuronal morphology, wherein the microscopic image data includes structural information of a plurality of neurons and their interconnections;

[0008] Performing edge extraction on the microscopic image data to identify the boundaries of each neuron and the morphological parameters of each neuron;

[0009] Classifying the neurons according to the morphological parameters, determining the category of each neuron, and obtaining multiple classification subsets;

[0010] Extracting features of neurons in each of the classification subsets and performing statistical analysis to obtain a baseline feature vector for each of the classification subsets;

[0011] The preset machine learning model is trained and verified using the benchmark feature vector to obtain a neuron recognition model;

[0012] Unidentified microscopic image data is acquired and processed using a trained neuron recognition model to automatically identify and label neurons in the unidentified microscopic image data.

[0013] In a preferred embodiment, after obtaining the microscopic image data containing neuronal morphology, preprocessing is performed simultaneously, and the preprocessing process includes:

[0014] Performing denoising on the microscopic image data to improve image clarity and signal-to-noise ratio;

[0015] Perform contrast enhancement on the denoised microscopic image data to enhance the connection structure between adjacent neurons;

[0016] The contrast-enhanced microscopic image data are scale-normalized to unify the scales of the microscopic image data.

[0017] In a preferred embodiment, the step of performing edge extraction on the microscopic image data to identify the boundaries of each neuron and the morphological parameters of each neuron includes:

[0018] Acquire pre-processed microscopic image data and record it as an image to be identified;

[0019] Segmenting the image to be identified layer by layer to obtain the edge contour of the neuron and simultaneously constructing the edge curve of the neuron;

[0020] The area, perimeter, and aspect ratio of the neuron are calculated based on the edge curve and summarized as the morphological parameters of the neuron.

[0021] In a preferred embodiment, the step of calculating the area, perimeter, and aspect ratio of the neuron based on the edge curve includes:

[0022] Constructing a virtual coordinate system covering the edge curve, and counting the position coordinates of the inflection point of the edge curve in the virtual coordinate system, and synchronously recording them as the inflection point coordinates;

[0023] Obtaining an area measurement function, inputting the inflection point coordinates into the area measurement function, and recording an output result of the area measurement function as the area of ​​the neuron;

[0024] Calculate the distance between the coordinates of adjacent inflection points and sum them up to get the perimeter of the neuron;

[0025] The distances between adjacent inflection point coordinates in the longitudinal state are summed to obtain the long axis length of the neuron. The distances between adjacent inflection point coordinates in the transverse state are summed to obtain the short axis length of the neuron. The ratio between the long axis length and the short axis length is determined as the aspect ratio of the neuron.

[0026] In a preferred embodiment, the step of classifying neurons according to the morphological parameters to determine the category of each neuron includes:

[0027] Obtain the morphological parameters and color feature information of neurons and jointly construct a multi-dimensional classification feature vector;

[0028] Obtaining a classification function, and inputting the classification feature vectors into the classification function in pairs, and recording the output result of the classification function as a classification condition parameter;

[0029] Obtaining a classification threshold, and comparing the classification threshold with a classification condition parameter;

[0030] When the classification condition parameter is greater than or equal to the classification threshold, it indicates that the neurons corresponding to the classification condition parameter are neurons of the same category and are aggregated into the same classification subset;

[0031] When the classification condition parameter is less than the classification threshold, it indicates that the neurons corresponding to the classification condition parameter are neurons of different categories and are aggregated into different classification subsets.

[0032] In a preferred embodiment, the step of extracting features of neurons in each of the classification subsets and performing statistical analysis to obtain a baseline feature vector for each of the classification subsets includes:

[0033] Acquiring neurons within the same classification subset, as well as morphological characteristics and color characteristics of each neuron;

[0034] Counting the differences between the morphological features and color features of each neuron, and recording them as the first feature difference and the second feature difference respectively;

[0035] Performing offset processing on the first characteristic difference and the second characteristic difference respectively to obtain a first evaluation interval corresponding to the first characteristic difference and a second evaluation interval corresponding to the second characteristic difference;

[0036] Counting the number of neurons corresponding to the first characteristic difference in the first evaluation interval and recording it as a first condition parameter;

[0037] Counting the number of neurons corresponding to the second characteristic difference in the second evaluation interval and recording it as a second condition parameter;

[0038] comparing the first condition parameter with the second condition parameter;

[0039] If the first condition parameter is greater than the second condition parameter, recording the first evaluation interval as a standard evaluation interval;

[0040] If the first condition parameter is less than or equal to the second condition parameter, recording the second evaluation interval as the standard evaluation interval;

[0041] The morphological features and color features within the standard evaluation interval are summarized and output as a baseline feature difference.

[0042] In a preferred embodiment, the step of training and verifying a preset machine learning model using the benchmark feature vector to obtain a neuron recognition model includes:

[0043] Obtaining a preset machine learning model and inputting the baseline feature vector into the machine learning model for preliminary training;

[0044] During the initial training process, the recognition accuracy of the machine learning model is monitored in real time and recorded as the recognition accuracy parameter;

[0045] When the recognition accuracy parameter is greater than or equal to a preset accuracy threshold, it indicates that the training effect of the machine learning model meets the standard, and the machine learning model at this time is determined as a neuron recognition model;

[0046] When the recognition accuracy parameter is less than the preset accuracy threshold, it indicates that the training effect of the machine learning model has not met the standard, and the machine learning model will continue to be optimized and trained until the recognition accuracy parameter of the machine learning model is greater than or equal to the preset accuracy threshold.

[0047] In a preferred embodiment, after the unidentified microscopic image data is output, it is synchronously input into a neuron recognition model;

[0048] The neuron recognition model analyzes the input microscopic image data and identifies neurons in the unidentified microscopic image data based on the learned reference feature vectors;

[0049] In the process of identifying neurons in the unidentified microscopic image data, the neuron recognition model will compare the similarity between the neuron features in the unidentified microscopic image data and the benchmark feature vector one by one to determine the category of the neurons, and simultaneously highlight neurons of different categories, and add corresponding identification labels to neurons of different categories.

[0050] The present invention also provides an AI-based neuron morphology image recognition system, using the above-mentioned AI-based neuron morphology image recognition method, comprising:

[0051] An image acquisition module, the image acquisition module being used to acquire microscopic image data containing neuronal morphology, wherein the microscopic image data includes structural information of multiple neurons and their interconnections;

[0052] An edge extraction module, configured to extract edges from the microscopic image data and identify the boundaries of each neuron and the morphological parameters of each neuron;

[0053] A classification module, configured to classify neurons according to the morphological parameters, determine the category of each neuron, and obtain multiple classification subsets;

[0054] A feature extraction module is used to extract features of neurons in each of the classification subsets and perform statistical analysis to obtain a baseline feature vector for each of the classification subsets;

[0055] A model training module, the model training module is used to train and verify a preset machine learning model using the benchmark feature vector to obtain a neuron recognition model;

[0056] The recognition module is used to obtain unrecognized microscopic image data, apply the trained neuron recognition model to process it, and automatically recognize and label neurons in the unrecognized microscopic image data.

[0057] And, an electronic device, comprising:

[0058] at least one processor;

[0059] and a memory communicatively coupled to the at least one processor;

[0060] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned AI-based neuronal morphological image recognition method.

[0061] The technical effects achieved by the present invention are:

[0062] The present invention constructs a multi-dimensional classification feature vector, fully considering the morphological parameters and color feature information of neurons, thereby being able to more comprehensively reflect the characteristics of neurons, and then accurately classify and identify neurons based on this. Through statistical analysis and offset processing, the most representative baseline feature difference can be determined, providing multi-dimensional data support for the training of machine learning models, thereby ensuring that the recognition accuracy of subsequent trained machine learning models is correspondingly improved. It can also demonstrate high stability and reliability when processing large-scale and complex neuronal morphological images, thereby meeting the current demand for high precision and efficiency in neuronal morphological image recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a schematic flow chart of the method of the present invention;

[0064] Figure 2 It is a schematic diagram of the system modules of the present invention;

[0065] Figure 3 It is a schematic structural diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0066] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0067] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0068] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it constitute a separate or selective embodiment that is mutually exclusive of other embodiments.

[0069] See also Figure 1 As shown, the present invention provides an AI-based neuron morphology image recognition method, comprising:

[0070] S1. Acquiring microscopic image data containing neuronal morphology, wherein the microscopic image data includes structural information of multiple neurons and their interconnections;

[0071] In step S1, before performing neuron morphology image recognition, it is first necessary to obtain the collected microscopic image data. The microscopic image data is usually obtained by professional microscopic imaging equipment. The collected microscopic image data not only contains detailed information of multiple neurons, but also covers the structural information of the interconnections between these neurons. After obtaining the microscopic image data containing neuron morphology, preprocessing is performed simultaneously. The preprocessing process includes:

[0072] De-noising the microscopic image data to improve the image clarity and signal-to-noise ratio;

[0073] Perform contrast enhancement on the denoised microscopic image data to enhance the connection structure between adjacent neurons;

[0074] Performing scale normalization on the contrast-enhanced microscopic image data to unify the scales of each microscopic image data;

[0075] Specifically, after successfully acquiring microscopic image data containing neuronal morphology, corresponding preprocessing operations need to be performed immediately to ensure the accuracy and efficiency of subsequent analysis. First, denoising is performed on the original microscopic image data. The purpose is to remove various noise interferences in the microscopic image data, thereby significantly improving the clarity of the image and optimizing the signal-to-noise ratio of the image, making the details of the neuronal structure more prominent and easier to identify. Secondly, after completing the denoising process, the denoised microscopic image data is subjected to contrast enhancement processing, which can effectively enhance the connection structure between adjacent neurons, making the boundaries and connection relationships between neurons clearer, which is convenient for subsequent analysis and research. Finally, the microscopic image data that has undergone contrast enhancement processing is scale normalized. The purpose is to unify the scale of each microscopic image data, eliminate the scale inconsistency problem caused by shooting conditions or equipment differences, and ensure that all image data are consistent and comparable in subsequent analysis.

[0076] S2. Perform edge extraction on the microscopic image data to identify the boundaries of each neuron and the morphological parameters of each neuron;

[0077] In step S2, after the microscopic image data has been preprocessed, corresponding edge extraction processing can be performed on it, so that the boundaries of each neuron can be accurately identified, and the morphological parameters of each neuron, such as size and shape, can be further extracted. The steps of performing edge extraction on the microscopic image data and identifying the boundaries of each neuron and the morphological parameters of each neuron include:

[0078] Acquire pre-processed microscopic image data and record it as an image to be identified;

[0079] The image to be identified is segmented layer by layer to obtain the edge contour of the neuron and simultaneously construct the edge curve of the neuron;

[0080] The area, perimeter, and aspect ratio of neurons are calculated based on the edge curve and summarized as the morphological parameters of neurons;

[0081] Specifically, when performing precise edge extraction on microscopic image data, the pre-processed microscopic image data is first obtained and recorded as the image to be identified for subsequent processing. The image to be identified is then segmented layer by layer in detail. The corresponding image segmentation algorithm (such as the level set method, threshold segmentation method, or edge detection algorithm) is used to gradually outline the edge contours of the neurons. In this process, the edge curve of the neurons is simultaneously constructed to ensure the accuracy and integrity of the edge curve. Finally, based on the constructed edge curve, the key morphological parameters of each neuron, such as the area, perimeter, and aspect ratio, are calculated using mathematical calculation methods. These parameters are then summarized to form a neuron morphological parameter dataset for subsequent analysis and research.

[0082] In addition, the steps of calculating the area, perimeter, and aspect ratio of the neuron based on the edge curve include:

[0083] Construct a virtual coordinate system covering the edge curve, and count the position coordinates of the inflection point of the edge curve in the virtual coordinate system, and simultaneously record them as the inflection point coordinates;

[0084] Obtain an area measurement function, input the inflection point coordinates into the area measurement function, and record the output of the area measurement function as the area of ​​the neuron;

[0085] Calculate the distance between the coordinates of adjacent inflection points and sum them up to get the perimeter of the neuron;

[0086] The distances between adjacent inflection point coordinates in the longitudinal state are summed to obtain the long axis length of the neuron. The distances between adjacent inflection point coordinates in the transverse state are summed to obtain the short axis length of the neuron. The ratio between the long axis length and the short axis length is determined as the aspect ratio of the neuron.

[0087] After the edge curve of the neuron is extracted from the microscopic image data, the area, perimeter, and aspect ratio of the neuron are calculated synchronously based on the edge curve. First, a virtual coordinate system that can completely cover the edge curve needs to be constructed, and the positions of the inflection points on the edge curve are counted in the virtual coordinate system. At the same time, the specific position coordinates of these inflection points are recorded in detail to form a list of inflection point coordinate data. After that, an area measurement function is obtained for measuring the area of ​​the neuron. The recorded inflection point coordinate data are input into the area measurement function one by one. The area of ​​the neuron is obtained through the calculation of the area measurement function. The expression of the area measurement function is: Where S represents the area of ​​the neuron, n represents the number of inflection points, i represents the number of inflection points, x and y represent the horizontal and vertical coordinates of the inflection points, respectively. Subsequently, the physical distances between adjacent inflection point coordinates are calculated one by one and accumulated and summed, and finally the circumference of the neuron can be obtained. Furthermore, in the longitudinal state, the distances between adjacent inflection point coordinates are accumulated and summed to obtain the long axis length of the neuron. In the transverse state, the distances between adjacent inflection point coordinates are also accumulated and summed to obtain the short axis length of the neuron. Finally, the ratio of the obtained long axis length to the short axis length is calculated to determine the aspect ratio of the neuron.

[0088] S3. Classify the neurons according to the morphological parameters, determine the category of each neuron, and obtain multiple classification subsets;

[0089] In step S3, after the morphological parameters of the neurons are determined, the neurons can be classified based on the obtained morphological parameters to determine the category of each neuron, and all neurons can be divided into multiple different classification subsets. The step of classifying the neurons based on the morphological parameters and determining the category of each neuron includes:

[0090] Obtain the morphological parameters and color feature information of neurons and jointly construct a multi-dimensional classification feature vector;

[0091] Obtain a classification function, and input the classification feature vectors into the classification function in pairs, and record the output of the classification function as a classification condition parameter;

[0092] Obtaining a classification threshold, and comparing the classification threshold with a classification condition parameter;

[0093] When the classification condition parameter is greater than or equal to the classification threshold, it indicates that the neurons corresponding to the classification condition parameter are neurons of the same category and are aggregated into the same classification subset;

[0094] When the classification condition parameter is less than the classification threshold, it indicates that the neurons corresponding to the classification condition parameter are neurons of different categories and are aggregated into different classification subsets;

[0095] Specifically, when classifying the morphological parameters of neurons, it is necessary not only to clarify the morphological parameters of each neuron, but also to collect its color feature information, such as color depth, distribution pattern, etc., to construct a multi-dimensional classification feature vector containing rich information. Then, a preset classification function is introduced, and the classification feature vectors are paired and input into the classification function for processing. The expression of the classification function is: In the formula, R represents the classification condition parameter, n represents the number of feature points in the classification feature vector, and a i and b i They respectively represent the different classification feature vectors input simultaneously, and then a pre-set classification threshold is introduced, and the classification threshold is compared and analyzed with the classification condition parameter. During the comparison process, if the classification condition parameter is found to be greater than or equal to the set classification threshold, it means that the neurons corresponding to the classification condition parameter have high consistency in morphology and characteristics, and can be determined as neurons of the same category, and they are summarized and classified into the corresponding same classification subset. On the contrary, if the classification condition parameter is less than the set classification threshold, it means that the neurons corresponding to the classification condition parameter have obvious differences in morphology or characteristics, and should be determined as neurons of different categories, and they should be classified into different classification subsets for subsequent further analysis and research.

[0096] S4. Extract features of neurons in each classification subset and perform statistical analysis to obtain the baseline feature vectors under each classification subset;

[0097] In step S4, based on the determination of the classification subsets, feature extraction is performed on the neurons in each classification subset, and corresponding statistical analysis is performed to ultimately obtain the benchmark feature vectors under each classification subset. The steps of extracting features from the neurons in each classification subset and performing statistical analysis to obtain the benchmark feature vectors under each classification subset include:

[0098] Obtain neurons within the same classification subset, as well as the morphological and color characteristics of each neuron;

[0099] Counting the differences between the morphological features and color features of each neuron, and recording them as the first feature difference and the second feature difference respectively;

[0100] Performing offset processing on the first characteristic difference and the second characteristic difference respectively to obtain a first evaluation interval corresponding to the first characteristic difference and a second evaluation interval corresponding to the second characteristic difference;

[0101] Counting the number of neurons corresponding to the first characteristic difference in the first evaluation interval and recording it as the first conditional parameter;

[0102] Counting the number of neurons corresponding to the second characteristic difference in the second evaluation interval and recording it as the second conditional parameter;

[0103] comparing the first condition parameter and the second condition parameter;

[0104] If the first condition parameter is greater than the second condition parameter, the first evaluation interval is recorded as the standard evaluation interval;

[0105] If the first condition parameter is less than or equal to the second condition parameter, the second evaluation interval is recorded as the standard evaluation interval;

[0106] Summarize the morphological and color features within the standard evaluation interval and output them as the baseline feature difference;

[0107] Specifically, detailed and systematic feature extraction is performed on the neurons in each classification subset, and on this basis, in-depth and comprehensive statistical analysis is carried out to gradually derive the specific steps of the baseline feature vector under each classification subset, which mainly includes the following key links: First, all neuron samples in the same classification subset are fully acquired, and the morphological features and color features of each neuron sample are carefully collected to ensure the integrity and accuracy of the data; secondly, the morphological features and color features of each collected neuron are compared one by one, and the differences between them are counted and recorded as the first feature difference and the second feature difference respectively for subsequent analysis; after that, the first feature difference and the second feature difference will be subjected to necessary offset processing (bidirectional equidistant offset, the specific offset can be adjusted according to actual conditions), so as to obtain the first evaluation interval corresponding to the first feature difference and the second evaluation interval corresponding to the second feature difference. The corresponding second evaluation interval, then, in the first evaluation interval, the number of neurons corresponding to the first characteristic difference is counted and recorded as the first conditional parameter. Similarly, in the second evaluation interval, the number of neurons corresponding to the second characteristic difference is counted and recorded as the second conditional parameter. Subsequently, the first conditional parameter and the second conditional parameter are compared and analyzed to determine the more representative parameter. If the value of the first conditional parameter is greater than the value of the second conditional parameter, the first evaluation interval is identified as the standard evaluation interval and used as the basis for subsequent analysis. Conversely, if the value of the first conditional parameter is less than or equal to the value of the second conditional parameter, the second evaluation interval is identified as the standard evaluation interval. Finally, the morphological characteristics and color characteristics of all neurons in the standard evaluation interval are systematically summarized and integrated, and the final result is output as the benchmark characteristic difference, providing reliable data support for subsequent classification and recognition work.

[0108] S5. Train and verify the preset machine learning model using the benchmark feature vector to obtain a neuron recognition model;

[0109] In step S5, after the baseline feature vector is determined, the preset machine learning model is trained and verified using the baseline feature vector to construct the required neuron recognition model. The steps of training and verifying the preset machine learning model using the baseline feature vector to obtain the neuron recognition model include:

[0110] Obtain a preset machine learning model and input the baseline feature vector into the machine learning model for preliminary training;

[0111] During the initial training process, the recognition accuracy of the machine learning model is monitored in real time and recorded as the recognition accuracy parameter;

[0112] When the recognition accuracy parameter is greater than or equal to the preset accuracy threshold, it indicates that the training effect of the machine learning model meets the standard, and the machine learning model at this time is determined to be a neuron recognition model;

[0113] When the recognition accuracy parameter is less than the preset accuracy threshold, it indicates that the training effect of the machine learning model has not met the standard, and the machine learning model will continue to be optimized and trained until the recognition accuracy parameter of the machine learning model is greater than or equal to the preset accuracy threshold;

[0114] Specifically, after the baseline feature vector is determined, it is first necessary to obtain a pre-set machine learning model and input the baseline feature vector into the machine learning model to start the preliminary training process. At this stage, the machine learning model begins to learn and adapt to the morphological features and color features corresponding to the baseline feature vector. During the preliminary training process, the recognition accuracy of the machine learning model is monitored in real time. This process can use various technical means such as cross-validation and accuracy curve analysis to ensure the objectivity and accuracy of the monitoring results, and record the monitored accuracy data in detail as the recognition accuracy parameter. When the monitored recognition accuracy parameter is greater than or equal to the pre-set accuracy threshold, it can be determined that the machine learning model is good. The training effect of the learning model has reached the expected standard. At this time, the machine learning model of this training stage is officially determined as a neuron recognition model. However, if the recognition accuracy parameter is less than the preset accuracy threshold, it indicates that the training effect of the current machine learning model has not yet reached the expected standard. In this case, it is necessary to continue to conduct targeted optimization training on the machine learning model. By adjusting model parameters, increasing training data or improving algorithms, the recognition accuracy of the model can be continuously improved. This process will continue until the recognition accuracy parameter of the machine learning model is finally greater than or equal to the preset accuracy threshold, ensuring that the final neuron recognition model has efficient and stable recognition performance.

[0115] S6. acquiring unidentified microscopic image data, applying the trained neuron recognition model to process the data, and automatically identifying and labeling neurons in the unidentified microscopic image data;

[0116] In step S6, the trained neuron recognition model is applied to the newly acquired and unrecognized microscopic image data to automatically identify and label the neurons in the unrecognized microscopic image data, thereby completing the entire recognition process. The unrecognized microscopic image data is output and then synchronously input into the neuron recognition model.

[0117] The neuron recognition model analyzes the input microscopic image data and identifies neurons in the unidentified microscopic image data based on the learned baseline feature vectors;

[0118] Among them, during the process of neuron identification in unidentified microscopic image data, the neuron identification model will compare the similarity between the neuron features in the unidentified microscopic image data and the baseline feature vector one by one to determine the category of the neuron, and simultaneously highlight neurons of different categories, and add corresponding identification labels to neurons of different categories.

[0119] Specifically, after the output operation of the unidentified microscopic image data is completed, it will be immediately transmitted and input into the neuron recognition model. After receiving the input microscopic image data, the neuron recognition model will immediately start the parsing program and perform accurate recognition operations on the neurons contained in the unidentified microscopic image data based on the benchmark feature vectors learned and stored in its previous training process. In the specific recognition process, for each neuron in the unidentified microscopic image data, the neuron recognition model will extract features one by one and compare them with the benchmark feature vectors for similarity. Through this comparison analysis, the category of each neuron can be accurately determined. At the same time, the highlighting operation will be performed synchronously to highlight neurons of different categories with different colors or markings for easy observation and analysis. In addition, corresponding identification labels will be automatically added for neurons of different categories to facilitate subsequent data processing and analysis.

[0120] See also Figure 2 , an AI-based neuron morphology image recognition system, using the above-mentioned AI-based neuron morphology image recognition method, comprising:

[0121] An image acquisition module, the image acquisition module is used to obtain microscopic image data containing neuronal morphology, wherein the microscopic image data includes structural information of multiple neurons and their interconnections;

[0122] The edge extraction module is used to extract the edges of the microscopic image data and identify the boundaries of each neuron and the morphological parameters of each neuron;

[0123] The classification module is used to classify neurons according to morphological parameters, determine the category of each neuron, and obtain multiple classification subsets;

[0124] Feature extraction module, which is used to extract features of neurons in each classification subset and perform statistical analysis to obtain the baseline feature vectors under each classification subset;

[0125] The model training module is used to train and verify the preset machine learning model through the benchmark feature vector to obtain a neuron recognition model;

[0126] The recognition module is used to obtain unidentified microscopic image data, apply the trained neuron recognition model to process it, and automatically identify and label neurons in the unidentified microscopic image data.

[0127] In the above, the system includes an image acquisition module, an edge extraction module, a classification module, a feature extraction module, a model training module and a recognition module. The main function of the image acquisition module is to obtain high-resolution microscopic image data containing neuronal morphology. These microscopic image data not only contain a large number of individual neurons, but also record in detail the complex interconnected structure information between these neurons, providing a rich raw data basis for subsequent image processing and analysis. The role of the edge extraction module is to perform precise edge extraction processing on the collected microscopic image data. Through the corresponding image processing algorithm, it can accurately identify the boundary contours of each neuron and further extract the morphological parameters of each neuron, such as size, shape, etc., to provide a corresponding basis for subsequent classification and feature extraction. The main task of the classification module is to perform detailed classification processing on the neurons according to the extracted morphological parameters. This can determine the category of each neuron, thereby dividing all neurons into multiple classification subsets with similar characteristics, which is convenient for subsequent in-depth analysis and research. The feature extraction module is used to perform in-depth feature extraction on neurons in each classification subset and perform corresponding statistical analysis. Through this process, the baseline feature vector under each classification subset can be obtained. The responsibility of the model training module is to systematically train and verify the preset machine learning model by using known baseline feature vectors. After multiple iterative optimizations, the neuron recognition model is finally obtained, which provides strong support for the actual neuron recognition task. The function of the recognition module is to obtain microscopic image data that has not yet been identified and apply the trained neuron recognition model for processing. Then the system can automatically identify and accurately mark each neuron in the unidentified microscopic image data, thereby improving the efficiency and accuracy of neuron recognition.

[0128] See also Figure 3 , an electronic device, the electronic device comprising:

[0129] at least one processor;

[0130] and a memory communicatively coupled to the at least one processor;

[0131] The memory stores a computer program that can be executed by at least one processor, and the computer program is executed by at least one processor so that the at least one processor can execute the above-mentioned AI-based neuronal morphological image recognition method.

[0132] The processor of the above-mentioned electronic device can be a chip with data processing capabilities such as a central processing unit (CPU), a graphics processing unit (GPU) or a neural processing unit (NPU). The memory can include storage media such as random access memory (RAM), read-only memory (ROM), flash memory (Flash) or a hard disk. When the computer program is executed by the processor, it can trigger the electronic device to execute the above-mentioned AI-based neuron morphology image recognition method to achieve automatic acquisition, processing, analysis and recognition of microscopic image data. In addition, the electronic device may also include an arithmetic unit, an input device and an output device. The arithmetic unit is used to perform various arithmetic and logical operations to ensure the normal operation of the electronic device. The input device, such as a keyboard, mouse or touch screen, allows the user to input instructions and data into the electronic device. The output device, such as a display, printer, etc., is used to display the processing results and recognition effects, so that the user can intuitively see the recognition status and analysis results of the neurons.

[0133] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0134] The foregoing is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained herein shall, unless otherwise specified or limited, be implemented in accordance with conventional means in the art.

Claims

1. An AI-based neuron morphology image recognition method, characterized by: include: Acquiring microscopic image data containing neuronal morphology, wherein the microscopic image data includes structural information of a plurality of neurons and their interconnections; Performing edge extraction on the microscopic image data to identify the boundaries of each neuron and the morphological parameters of each neuron; Classifying the neurons according to the morphological parameters, determining the category of each neuron, and obtaining multiple classification subsets; Extracting features of neurons in each of the classification subsets and performing statistical analysis to obtain a baseline feature vector for each of the classification subsets; The preset machine learning model is trained and verified using the benchmark feature vector to obtain a neuron recognition model; Obtain unidentified microscopic image data, apply the trained neuron recognition model to process it, and automatically identify and label neurons in the unidentified microscopic image data; The step of extracting features of neurons in each of the classification subsets and performing statistical analysis to obtain a baseline feature vector for each of the classification subsets includes: Acquiring neurons within the same classification subset, as well as morphological characteristics and color characteristics of each neuron; Counting the differences between the morphological features and color features of each neuron, and recording them as the first feature difference and the second feature difference respectively; Performing offset processing on the first characteristic difference and the second characteristic difference respectively to obtain a first evaluation interval corresponding to the first characteristic difference and a second evaluation interval corresponding to the second characteristic difference; Counting the number of neurons corresponding to the first characteristic difference in the first evaluation interval and recording it as a first condition parameter; Counting the number of neurons corresponding to the second characteristic difference in the second evaluation interval and recording it as a second condition parameter; comparing the first condition parameter with the second condition parameter; If the first condition parameter is greater than the second condition parameter, recording the first evaluation interval as a standard evaluation interval; If the first condition parameter is less than or equal to the second condition parameter, recording the second evaluation interval as the standard evaluation interval; The morphological features and color features within the standard evaluation interval are summarized and output as a baseline feature difference.

2. The AI-based neuron morphology image recognition method according to claim 1, characterized in that: After obtaining the microscopic image data containing neuronal morphology, preprocessing is performed simultaneously, and the preprocessing process includes: Performing denoising on the microscopic image data to improve image clarity and signal-to-noise ratio; Perform contrast enhancement on the denoised microscopic image data to enhance the connection structure between adjacent neurons; The contrast-enhanced microscopic image data are scale-normalized to unify the scales of the microscopic image data.

3. The AI-based neuron morphology image recognition method according to claim 2, characterized in that: The step of performing edge extraction on the microscopic image data to identify the boundaries of each neuron and the morphological parameters of each neuron includes: Acquire pre-processed microscopic image data and record it as an image to be identified; Segmenting the image to be identified layer by layer to obtain the edge contour of the neuron and simultaneously constructing the edge curve of the neuron; The area, perimeter, and aspect ratio of the neuron are calculated based on the edge curve and summarized as the morphological parameters of the neuron.

4. The AI-based neuron morphology image recognition method according to claim 3, characterized in that: The step of calculating the area, perimeter, and aspect ratio of the neuron based on the edge curve includes: Constructing a virtual coordinate system covering the edge curve, and counting the position coordinates of the inflection point of the edge curve in the virtual coordinate system, and synchronously recording them as the inflection point coordinates; Obtaining an area measurement function, inputting the inflection point coordinates into the area measurement function, and recording an output result of the area measurement function as the area of ​​the neuron; Calculate the distance between the coordinates of adjacent inflection points and sum them up to get the perimeter of the neuron; The distances between adjacent inflection point coordinates in the longitudinal state are summed to obtain the long axis length of the neuron. The distances between adjacent inflection point coordinates in the transverse state are summed to obtain the short axis length of the neuron. The ratio between the long axis length and the short axis length is determined as the aspect ratio of the neuron.

5. The AI-based neuron morphology image recognition method according to claim 1, characterized in that: The step of classifying neurons according to the morphological parameters to determine the category of each neuron includes: Obtain the morphological parameters and color feature information of neurons and jointly construct a multi-dimensional classification feature vector; Obtaining a classification function, and inputting the classification feature vectors into the classification function in pairs, and recording the output result of the classification function as a classification condition parameter; Obtaining a classification threshold, and comparing the classification threshold with a classification condition parameter; When the classification condition parameter is greater than or equal to the classification threshold, it indicates that the neurons corresponding to the classification condition parameter are neurons of the same category and are aggregated into the same classification subset; When the classification condition parameter is less than the classification threshold, it indicates that the neurons corresponding to the classification condition parameter are neurons of different categories and are aggregated into different classification subsets.

6. The AI-based neuron morphology image recognition method according to claim 1, characterized in that: The step of training and verifying a preset machine learning model using the benchmark feature vector to obtain a neuron recognition model includes: Obtaining a preset machine learning model and inputting the baseline feature vector into the machine learning model for preliminary training; During the initial training process, the recognition accuracy of the machine learning model is monitored in real time and recorded as the recognition accuracy parameter; When the recognition accuracy parameter is greater than or equal to a preset accuracy threshold, it indicates that the training effect of the machine learning model meets the standard, and the machine learning model at this time is determined as a neuron recognition model; When the recognition accuracy parameter is less than the preset accuracy threshold, it indicates that the training effect of the machine learning model has not met the standard, and the machine learning model will continue to be optimized and trained until the recognition accuracy parameter of the machine learning model is greater than or equal to the preset accuracy threshold.

7. The AI-based neuron morphology image recognition method according to claim 1, characterized in that: After the unidentified microscopic image data is output, it is synchronously input into a neuron recognition model; The neuron recognition model analyzes the input microscopic image data and identifies neurons in the unidentified microscopic image data based on the learned reference feature vectors; In the process of identifying neurons in the unidentified microscopic image data, the neuron recognition model will compare the similarity between the neuron features in the unidentified microscopic image data and the benchmark feature vector one by one to determine the category of the neurons, and simultaneously highlight neurons of different categories, and add corresponding identification labels to neurons of different categories.

8. An AI-based neuron morphology image recognition system, characterized by: The method for neuron morphology recognition based on AI according to any one of claims 1 to 7 comprises: An image acquisition module, the image acquisition module being used to acquire microscopic image data containing neuronal morphology, wherein the microscopic image data includes structural information of multiple neurons and their interconnections; An edge extraction module, configured to extract edges from the microscopic image data and identify the boundaries of each neuron and the morphological parameters of each neuron; A classification module, configured to classify neurons according to the morphological parameters, determine the category of each neuron, and obtain multiple classification subsets; A feature extraction module is used to extract features of neurons in each of the classification subsets and perform statistical analysis to obtain a baseline feature vector for each of the classification subsets; A model training module, the model training module is used to train and verify a preset machine learning model using the benchmark feature vector to obtain a neuron recognition model; The recognition module is used to obtain unrecognized microscopic image data, apply the trained neuron recognition model to process it, and automatically recognize and label neurons in the unrecognized microscopic image data.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the AI-based neuronal morphological image recognition method described in any one of claims 1 to 7.

Images