A method and system for artificial intelligence analysis of bone marrow cell morphology

Through the artificial intelligence analysis method of bone marrow cell morphology and neural network, combined with the image extraction algorithm of the region of interest and the multi-layer convolution feature extraction, the problems of time-consuming and subjective differences in traditional manual examination methods are solved, and efficient and accurate Automatic identification and classification of bone marrow cells.

CN118506361BActive Publication Date: 2024-11-12ZHUJIANG HOSPITAL OF SOUTHERN MEDICAL UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410455333.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-16
Publication Date
2024-11-12
Estimated Expiration
2044-04-16

AI Technical Summary

Technical Problem

Traditional manual examination methods are time-consuming and labor-intensive in the recognition of bone marrow cell morphology, are susceptible to personal professional level and subjective differences, lack efficiency and accuracy, and are difficult to achieve automated diagnosis.

Method used

Using artificial intelligence analysis method for bone marrow cell morphology, combining the image extraction algorithm of the region of interest and the pre-constructed bone marrow cell morphology recognition neural network, feature extraction and cell recognition are performed through multi-layer convolution and attention mechanisms, and the intersection of cell bounding boxes is calculated. to achieve automated identification.

Benefits of technology

提高了骨髓细胞形态识别的效率和准确性,减少了数据处理量,实现了自动化的细胞识别和分类,增强了识别的鲁棒性和可信度。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118506361B_ABST
    Figure CN118506361B_ABST
Patent Text Reader

Abstract

The present application provides a method and system for artificial intelligence analysis of bone marrow cell morphology, wherein the artificial intelligence analysis method for bone marrow cell morphology includes: collecting bone marrow cell sample images, preprocessing the sample images to obtain standardized images; based on the image brightness value, using a region of interest image extraction algorithm, extracting a region of interest from the standardized image to obtain a region of interest image, wherein the region of interest is the area where bone marrow cells are concentrated; using a pre-constructed bone marrow cell morphology recognition neural network, performing feature extraction and bone marrow cell morphology recognition on the region of interest image to generate a bone marrow cell recognition result, wherein the bone marrow cell recognition result includes cell type and cell number.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method and system for artificial intelligence analysis of bone marrow cell morphology. Background Art

[0002] Bone marrow is one of the important organs of the human body, and all blood cells in the human body come from it. Bone marrow lesions can cause abnormal hematopoietic function in the human body, thereby causing related blood diseases. Bone marrow cell morphology recognition is one of the most commonly used basic methods for diagnosing blood diseases in clinical medicine. It generally includes sampling, smearing, staining, and specialized laboratory physicians observing the cell morphology in the bone marrow smear under an optical microscope and counting the proportional relationship between various types of cells. For example, in leukemia, the related pathological cells have obvious pathological characteristics in morphology. Through bone marrow cell morphology recognition, effective and rapid diagnosis can be initially achieved.

[0003] However, traditional manual inspection methods have many disadvantages. They are not only time-consuming and labor-intensive, but also easily affected by factors such as personal professional level and subjective differences. Therefore, using artificial intelligence image analysis technology to study and realize objective, efficient, and automatic bone marrow cell morphology recognition technology to improve work efficiency and accuracy is of great significance for the digital auxiliary diagnosis of hematopoietic system diseases. Summary of the invention

[0004] In view of the above problems, this application is proposed to provide a bone marrow cell morphology artificial intelligence analysis method and system to improve the efficiency and accuracy of bone marrow cell morphology recognition. At the same time, the region of interest image extraction algorithm is adopted to extract the region of interest image using the optimal brightness value, effectively locate the bone marrow cells, eliminate the image noise of the non-interested region, reduce the amount of data processing, and further improve the efficiency and accuracy of bone marrow cell morphology recognition.

[0005] The present application provides a bone marrow cell morphology artificial intelligence analysis method, comprising steps S1 to S3:

[0006] S1: collecting a bone marrow cell sample image, and preprocessing the sample image to obtain a standardized image;

[0007] S2: Based on the image brightness value, using a region of interest image extraction algorithm, extracting a region of interest from the standardized image to obtain a region of interest image, wherein the region of interest image extraction algorithm includes an extraction function, and the extraction function is constructed by the Euclidean distance and clustering degree between the image of interest and the image of no interest, and the region of interest is the region where bone marrow cells are concentrated;

[0008] S3: using a pre-built bone marrow cell morphology recognition neural network to perform feature extraction and bone marrow cell morphology recognition on the image of the region of interest, and generate a bone marrow cell recognition result, wherein the bone marrow cell recognition result includes cell type and cell number;

[0009] The pre-built bone marrow cell morphology recognition neural network is used to perform feature extraction and bone marrow cell morphology recognition on the image of the region of interest to generate a bone marrow cell recognition result, including:

[0010] In the input stage, the region of interest image is input into the bone marrow cell morphology recognition neural network;

[0011] In the feature extraction stage, the input image is quickly encoded and learned through multiple convolutional layers to extract the preliminary feature image of the cell;

[0012] In the feature selection and enhancement stage, a pooling layer is used to extract high-frequency and low-frequency features of bone marrow cells from the preliminary feature image;

[0013] In the attention mechanism application stage, the high-frequency features and the low-frequency features are feature encoded by convolution to generate an attention feature distribution, and a dot multiplication operation is performed with the preliminary feature image;

[0014] Cell detection and positioning stage: Calculate the intersection between the predicted cell bounding box and the actual cell bounding box to generate bone marrow cell recognition results.

[0015] Furthermore, the sample image is preprocessed to obtain a standardized image, which specifically includes the following steps:

[0016] S11: first resize and convert the grayscale of the collected sample image, and then perform normalization processing to obtain a normalized image;

[0017] S12: Based on mathematical morphology, filtering is performed on the normalized image to remove noise to obtain the standardized image.

[0018] Furthermore, the region of interest image extraction algorithm specifically includes the following steps:

[0019] S21: define a brightness value δ, and divide the pixels of the standardized image into an image of interest and an image of non-interest, wherein the image of interest and the image of non-interest are respectively expressed as:

[0020] ;

[0021] ;

[0022] in, , represents the image of interest and the image of no interest respectively, P and Q represent the length and width of the standardized image respectively, F represents the number of brightness values ​​of the standardized image, and the brightness value of the standardized image belongs to , Indicates the number of occurrences of brightness value i.

[0023] S22: Calculate the Euclidean distance D between the image of interest and the image of no interest, specifically:

[0024] ;

[0025] Wherein, the Euclidean distance D represents the degree of distinction between the image of interest and the image of no interest;

[0026] S23: Calculate the clustering degree of the image of interest and the image of no interest, specifically:

[0027] O roi = ∑ i=δ+1 F-1 [ i - ∑ i=δ+1 F-1 ( i t i P×Q / H roi ) ∙ t i P×Q / ∑ i=δ+1 F-1 t i P×Q ] ;

[0028] O not = ∑ i=0 d [ i - ∑ i=0 d i t i P×Q / H not ∙ t i P×Q / ∑ i=0 d t i P×Q ] ;

[0029] in, , Respectively represent the clustering degrees of the image of interest and the image of no interest;

[0030] S24: According to the Euclidean distance D and the clustering degree of the image of interest and the clustering degree of the non-interested image , construct the region of interest image extraction function L, specifically:

[0031] ;

[0032] S25: Calculate the optimal brightness value δ corresponding to the maximum value of the image extraction function L of the region of interest opt , extract the region of interest image.

[0033] Furthermore, the bone marrow cell morphology recognition neural network is a pre-trained model, which is a VGGNet-16 model or a ResNet-50 model. The pre-trained model divides the manually labeled bone marrow cell images into a training set, a validation set, and a test set.

[0034] Furthermore, the pre-trained model: the training set is used to train the neural network and update the network weights through the back-propagation algorithm; the validation set is used to optimize the model parameters and adjust the hyperparameters including the learning rate and batch size to ensure the generalization ability of the model; the test set is used to evaluate the model performance, focusing on indicators including accuracy, recall rate, and F1 score.

[0035] The beneficial effects of this application are:

[0036] (1) This application constructs a bone marrow cell morphology artificial intelligence analysis method and system, which uses a bone marrow cell morphology recognition neural network to objectively, efficiently and automatically realize bone marrow cell morphology recognition, thereby improving work efficiency and accuracy.

[0037] (2) This application also designs a region of interest image extraction algorithm, which uses the optimal brightness value to extract the region of interest image, effectively locates the bone marrow cells, eliminates the image noise of the non-region of interest, reduces the amount of data processing, and further improves the efficiency and accuracy of bone marrow cell morphology recognition.

[0038] (3) This application calculates the intersection between the predicted cell bounding box and the actual cell bounding box, so that the cell can be located and identified at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 A flowchart of the bone marrow cell morphology identification method provided in this application;

[0041] Figure 2 A diagram showing the cross-relationship between the predicted cell boundary box and the actual cell boundary box of the bone marrow cells provided in this application. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0043] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms of "a", "said", and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two, but does not exclude the inclusion of at least one.

[0044] The present application constructs a bone marrow cell morphology artificial intelligence analysis method and system, which adopts a bone marrow cell morphology recognition neural network to objectively, efficiently and automatically realize bone marrow cell morphology recognition, thereby improving work efficiency and accuracy.

[0045] At the same time, this application also designs a region of interest image extraction algorithm, which uses the optimal brightness value to extract the region of interest image, effectively locates the bone marrow cells, eliminates image noise in non-regions of interest, reduces the amount of data processing, and further improves the efficiency and accuracy of bone marrow cell morphology recognition.

[0046] The present application also constructs an intersection function to calculate the intersection between the predicted cell boundary box and the actual cell boundary box, so that the positioning and identification of cells can be achieved simultaneously.

[0047] The present application is further described below in conjunction with the accompanying drawings and specific embodiments.

[0048] First embodiment

[0049] like Figure 1 As shown, the present application provides a method and system for artificial intelligence analysis of bone marrow cell morphology. The bone marrow cell morphology recognition method specifically includes the following steps S1 to S3:

[0050] S1: Collect bone marrow cell sample images, pre-process the sample images, and obtain standardized images.

[0051] As a preferred implementation, the preprocessing of the sample image to obtain a standardized image specifically includes the following steps:

[0052] S11: first resize and convert the grayscale of the collected sample image, and then perform normalization processing to obtain a normalized image;

[0053] S12: Based on mathematical morphology, the normalized image is filtered to remove noise to obtain a standardized image.

[0054] S2: Based on the image brightness value, using the region of interest image extraction algorithm, extracting the region of interest from the standardized image to obtain the region of interest image, wherein the region of interest image extraction algorithm includes an extraction function, and the extraction function is constructed by the Euclidean distance and clustering degree between the image of interest and the image of no interest, and the region of interest is the area where the bone marrow cells are concentrated;

[0055] As a preferred implementation, the region of interest image extraction algorithm specifically includes the following steps:

[0056] S21: Define the brightness value δ, and divide the pixels of the standardized image into an image of interest and an image of non-interest. The image of interest and the image of non-interest are respectively expressed as:

[0057] ;

[0058] ;

[0059] in, , Represent the image of interest and the image of no interest respectively, P and Q represent the length and width of the standardized image respectively, F represents the number of brightness values ​​of the standardized image, and the brightness value of the standardized image belongs to , Indicates the number of occurrences of brightness value i.

[0060] S22: Calculate the Euclidean distance D between the image of interest and the image of non-interest, specifically:

[0061] ;

[0062] The Euclidean distance D represents the degree of distinction between the image of interest and the image of no interest. As the Euclidean distance D increases, the effect of extracting the image of interest gradually becomes better.

[0063] S23: Calculate the clustering degree of the image of interest and the image of non-interest, specifically:

[0064] O roi = ∑ i=δ+1 F-1 [ i - ∑ i=δ+1 F-1 ( i t i P×Q / H roi ) ∙ t i P×Q / ∑ i=δ+1 F-1 t i P×Q ] ;

[0065] O not = ∑ i=0 d [ i - ∑ i=0 d i t i P×Q / H not ∙ t i P×Q / ∑ i=0 d t i P×Q ] ;

[0066] in, , Represent the clustering degree of the image of interest and the image of non-interest, respectively. , As the value decreases, the clustering degree of the images of interest and the images of non-interest gradually becomes better, and the effect of extracting the images of interest also gradually becomes better.

[0067] S24: Based on the Euclidean distance D and the clustering degree of the image of interest and clustering degree of non-interest images , construct the region of interest image extraction function L, specifically:

[0068] ;

[0069] S25: Calculate the optimal brightness value δ corresponding to the maximum value of the image extraction function L in the region of interest opt , extract the region of interest image to optimize the region of interest image extraction effect.

[0070] This application designs an algorithm for extracting images of regions of interest, which uses the optimal brightness value to extract images of regions of interest, effectively locates bone marrow cells, eliminates image noise in non-regions of interest, reduces the amount of data processing, and improves the efficiency and accuracy of bone marrow cell morphology recognition.

[0071] S3: using a pre-built bone marrow cell morphology recognition neural network to perform feature extraction and bone marrow cell morphology recognition on the image of the region of interest to generate a bone marrow cell recognition result, wherein the bone marrow cell recognition result includes cell type and cell number;

[0072] The pre-built bone marrow cell morphology recognition neural network is used to perform feature extraction and bone marrow cell morphology recognition on the image of the region of interest to generate a bone marrow cell recognition result, including:

[0073] In the input stage, the region of interest image is input into the bone marrow cell morphology recognition neural network;

[0074] In the feature extraction stage, the input image is quickly encoded and learned through multiple convolutional layers to extract the preliminary feature image of the cell;

[0075] Among them, the preliminary feature image is W 1 ×H 1 ×C 1 , where W 1 , H 1 are the length and width of the preliminary feature map, respectively, C 1 is the number of channels of the preliminary feature image.

[0076] In the feature selection and enhancement stage, a pooling layer is used to extract high-frequency features and low-frequency features of bone marrow cells from the preliminary feature image;

[0077] Among them, according to the characteristics of bone marrow cells of different forms, the characteristics of the bone marrow cells may include: cell image major axis, minor axis, major-minor axis ratio, cell circumference, cell area, cell image mean, cell image variance, cell image contrast, cell image root mean square brightness, etc.

[0078] In the attention mechanism application stage, the high-frequency features and the low-frequency features are feature encoded by convolution to generate an attention feature distribution, and a dot multiplication operation is performed with the preliminary feature image;

[0079] Cell detection and positioning stage: Calculate the intersection between the predicted cell bounding box and the actual cell bounding box, and generate a bone marrow cell recognition result based on the intersection.

[0080] like Figure 2 As shown, the intersection relationship between the predicted cell bounding box of the bone marrow cells and the actual cell bounding box is calculated, and the intersection degree C is calculated as follows:

[0081] ;

[0082] Among them, b, b z are the center points of the predicted cell bounding box and the actual cell bounding box, represents the intersection of the predicted cell bounding box and the actual cell bounding box, represents the union of the predicted cell bounding box and the actual cell bounding box, represents the Euclidean distance between the predicted cell bounding box and the actual cell bounding box, and c is the diagonal of the minimum enclosing rectangle.

[0083] When the cross-degree is less than or equal to a preset cross-degree threshold, determining that the bone marrow cell detection and identification is correct;

[0084] When the cross-degree is greater than the cross-degree threshold, it is determined that the bone marrow cell detection and recognition is wrong.

[0085] As a preferred embodiment, the bone marrow cell morphology recognition neural network adopts a pre-training model, which is a VGGNet-16 model or a ResNet-50 model.

[0086] The pre-trained model divides manually labeled bone marrow cell images into a training set, a validation set, and a test set.

[0087] The training set is used to train the neural network and update the network weights through a back-propagation algorithm;

[0088] The validation set is used to optimize model parameters and adjust hyperparameters including learning rate and batch size to ensure the generalization ability of the model;

[0089] The test set is used to evaluate the performance of the model, and the focus indicators include accuracy, recall, and F1 score.

[0090] Second embodiment

[0091] The present application also provides a bone marrow cell morphology artificial intelligence analysis system. The bone marrow cell morphology analysis system specifically includes:

[0092] An image acquisition and preprocessing module acquires bone marrow cell sample images, preprocesses the sample images, and obtains standardized images;

[0093] As a preferred implementation, the image acquisition and preprocessing module specifically includes:

[0094] The normalization processing module first resizes and converts the grayscale of the collected sample image, and then performs normalization processing to obtain a normalized image;

[0095] The filtering processing module performs filtering processing on the normalized image based on mathematical morphology to remove noise and obtain a standardized image.

[0096] An area of ​​interest image acquisition module is used to extract an area of ​​interest from a standardized image based on an image brightness value and to acquire an area of ​​interest image by using an area of ​​interest image extraction algorithm, wherein the area of ​​interest image extraction algorithm includes an extraction function, and the extraction function is constructed by the Euclidean distance and clustering degree between an image of interest and an image of non-interest, and the area of ​​interest is an area where bone marrow cells are concentrated;

[0097] As a preferred implementation, the region of interest image extraction algorithm specifically includes the following steps:

[0098] S21: Define the brightness value δ, and divide the pixels of the standardized image into an image of interest and an image of non-interest. The image of interest and the image of non-interest are respectively expressed as:

[0099] ;

[0100] ;

[0101] in, , Represent the image of interest and the image of no interest respectively, P and Q represent the length and width of the standardized image respectively, F represents the number of brightness values ​​of the standardized image, and the brightness value of the standardized image belongs to , Indicates the number of occurrences of brightness value i.

[0102] S22: Calculate the Euclidean distance D between the image of interest and the image of non-interest, specifically:

[0103] ;

[0104] The Euclidean distance D represents the degree of distinction between the image of interest and the image of no interest. As the Euclidean distance D increases, the image extraction effect of the region of interest gradually improves.

[0105] S23: Calculate the clustering degree of the image of interest and the image of non-interest, specifically:

[0106] O roi = ∑ i=δ+1 F-1 [ i - ∑ i=δ+1 F-1 ( i t i P×Q / H roi ) ∙ t i P×Q / ∑ i=δ+1 F-1 t i P×Q ] ;

[0107] O not = ∑ i=0 d [ i - ∑ i=0 d i t i P×Q / H not ∙ t i P×Q / ∑ i=0 d t i P×Q ] ;

[0108] in, , Represent the clustering degree of the image of interest and the image of non-interest, respectively. , As the value decreases, the clustering degree of the images of interest and the images of no interest gradually becomes better, and the image extraction effect of the region of interest also gradually becomes better.

[0109] S24: Based on the Euclidean distance D and the clustering degree of the image of interest and clustering degree of non-interest images , construct the region of interest image extraction function L, specifically:

[0110] ;

[0111] S25: Calculate the optimal brightness value δ corresponding to the maximum value of the image extraction function L in the region of interest opt , extract the region of interest image to optimize the region of interest image extraction effect.

[0112] A neural network recognition module uses a pre-built bone marrow cell morphology recognition neural network to perform feature extraction and bone marrow cell morphology recognition on the image of the region of interest to generate a bone marrow cell recognition result, wherein the bone marrow cell recognition result includes cell type and cell number;

[0113] The pre-built bone marrow cell morphology recognition neural network is used to perform feature extraction and bone marrow cell morphology recognition on the image of the region of interest to generate a bone marrow cell recognition result, including:

[0114] In the input stage, the region of interest image is input into the bone marrow cell morphology recognition neural network;

[0115] In the feature extraction stage, the image of the region of interest is quickly encoded and learned through multiple convolutional layers to extract the preliminary feature image of the cell;

[0116] Among them, the preliminary feature image is W 1 ×H 1 ×C 1 , where W 1 , H 1 are the length and width of the preliminary feature map, respectively, C 1 is the number of channels of the preliminary feature image.

[0117] In the feature selection and enhancement stage, a pooling layer is used to extract high-frequency features and low-frequency features of bone marrow cells from the preliminary feature image;

[0118] Among them, according to the characteristics of bone marrow cells of different forms, the characteristics of the bone marrow cells may include: cell image major axis, minor axis, major-minor axis ratio, cell circumference, cell area, cell image mean, cell image variance, cell image contrast, cell image root mean square brightness, etc.

[0119] In the attention mechanism application stage, the high-frequency features and the low-frequency features are feature encoded by convolution to generate an attention feature distribution, and a dot multiplication operation is performed with the preliminary feature image;

[0120] Cell detection and positioning stage: Calculate the intersection between the predicted cell bounding box and the actual cell bounding box, and generate a bone marrow cell recognition result based on the intersection.

[0121] like Figure 2 As shown, the intersection relationship between the predicted cell bounding box of the bone marrow cells and the actual cell bounding box is calculated, and the intersection degree C is calculated as follows:

[0122] ;

[0123] Among them, b, b z are the center points of the predicted cell bounding box and the actual cell bounding box, represents the intersection of the predicted cell bounding box and the actual cell bounding box, represents the union of the predicted cell bounding box and the actual cell bounding box, represents the Euclidean distance between the predicted cell bounding box and the actual cell bounding box, and c is the diagonal of the minimum enclosing rectangle.

[0124] When the cross-degree is less than or equal to a preset cross-degree threshold, determining that the bone marrow cell detection and identification is correct;

[0125] When the cross-degree is greater than the cross-degree threshold, it is determined that the bone marrow cell detection and recognition is wrong.

[0126] As a preferred embodiment, the bone marrow cell morphology recognition neural network adopts a pre-training model, which is a VGGNet-16 model or a ResNet-50 model.

[0127] The pre-trained model divides manually labeled bone marrow cell images into a training set, a validation set, and a test set.

[0128] The training set is used to train the neural network and update the network weights through a back-propagation algorithm;

[0129] The validation set is used to optimize model parameters and adjust hyperparameters including learning rate, batch size, etc. to ensure the generalization ability of the model;

[0130] The test set is used to evaluate the performance of the model, and the focus indicators include accuracy, recall, and F1 score.

[0131] In the embodiment of the present application, the bone marrow cell sample image is first preprocessed and standardized, which can eliminate noise in the image, enhance cell features, and ensure that the image has consistent brightness and contrast. Beneficial effects include: improving image quality, which is conducive to the accuracy of subsequent analysis and recognition. Making the image consistent, which is convenient for subsequent processing and comparative analysis. Based on the image brightness value, the region of interest image extraction algorithm is used to obtain the region of interest image of the bone marrow cell concentration. Beneficial effects include:

[0132] Reduce unnecessary analysis overhead and improve processing efficiency. Ensure that the required cell area is accurately extracted to avoid interference from processing irrelevant areas. Use the pre-built neural network to extract features from the image of the region of interest and recognize the morphology of bone marrow cells to generate bone marrow cell recognition results. Beneficial effects include: automatic recognition and classification of bone marrow cells, improving the accuracy and speed of recognition. Feature extraction through neural networks helps to discover hidden features of cells and improve the robustness of recognition. Multi-layer convolution and attention feature distribution calculation: Multiple convolution layers are used in the neural network for feature encoding learning, combining technologies such as pooling layers, convolution operations, and attention feature distribution calculation. Beneficial effects include: improving feature extraction and encoding capabilities, and enhancing recognition accuracy and robustness. By calculating the intersection of cell bounding boxes, bone marrow cell recognition results can be generated more accurately, improving the credibility of recognition. In general, this process combines image processing and deep learning technologies, and has many beneficial effects in the field of bone marrow cell recognition, including improving image quality, accurately extracting regions of interest, and automatically identifying and classifying cells.

[0133] The above description shows and describes several preferred embodiments of the present application, but as mentioned above, it should be understood that the present application is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the invention concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present application, and should be within the scope of protection of the claims attached to the present application.

Claims

1. A bone marrow cell morphology artificial intelligence analysis method, characterized in that: include: Collecting a bone marrow cell sample image, and preprocessing the sample image to obtain a standardized image; Based on the image brightness value, using a region of interest image extraction algorithm, extracting a region of interest from the standardized image to obtain a region of interest image, wherein the region of interest image extraction algorithm includes an extraction function, and the extraction function is constructed by the Euclidean distance and clustering degree between the image of interest and the image of no interest, and the region of interest is the region where bone marrow cells are concentrated; Using a pre-built bone marrow cell morphology recognition neural network, feature extraction and bone marrow cell morphology recognition are performed on the image of the region of interest to generate a bone marrow cell recognition result, wherein the bone marrow cell recognition result includes cell type and cell number; The pre-built bone marrow cell morphology recognition neural network is used to perform feature extraction and bone marrow cell morphology recognition on the image of the region of interest to generate a bone marrow cell recognition result, including: Inputting the region of interest image into the bone marrow cell morphology recognition neural network; Performing fast encoding learning on the image of the region of interest through multiple convolutional layers to extract a preliminary characteristic image of the cell; A pooling layer is used to extract high-frequency features and low-frequency features of bone marrow cells from the preliminary feature image; Performing feature encoding on the high-frequency features and the low-frequency features by using convolution to generate an attention feature distribution, and performing a dot multiplication operation with the preliminary feature image; The intersection degree between the predicted cell boundary box and the actual cell boundary box is calculated, and a bone marrow cell recognition result is generated based on the intersection degree.

2. The method according to claim 1, characterized in that The preprocessing of the sample image to obtain a standardized image specifically includes the following steps: The sample image is sequentially resized, gray-scale converted, and normalized to obtain a normalized image; Based on mathematical morphology, the normalized image is filtered and denoised to obtain the standardized image.

3. The method according to claim 1, characterized in that The method of extracting a region of interest from the standardized image based on the image brightness value and obtaining the region of interest image by using a region of interest image extraction algorithm comprises: S21: define a brightness value δ, and divide the pixels of the standardized image into an image of interest and an image of non-interest, wherein the image of interest and the image of non-interest are respectively expressed as: ; ; in, , represents the image of interest and the image of no interest respectively, P and Q represent the length and width of the standardized image respectively, F represents the number of brightness values ​​of the standardized image, and the brightness value of the standardized image belongs to , Indicates the number of occurrences of brightness value i; S22: Calculate the Euclidean distance D between the image of interest and the image of no interest, specifically: ; Wherein, the Euclidean distance D represents the degree of distinction between the image of interest and the image of no interest; S23: Calculate the clustering degree of the image of interest and the image of no interest, specifically: ; ; in, , Respectively represent the clustering degrees of the image of interest and the image of no interest; S24: According to the Euclidean distance D and the clustering degree of the image of interest and the clustering degree of the non-interested image , construct the region of interest image extraction function L, specifically: ; S25: Calculate the optimal brightness value δ corresponding to the maximum value of the image extraction function L of the region of interest opt , extract the region of interest image.

4. The method according to claim 1, characterized in that: The bone marrow cell morphology recognition neural network is a pre-trained model, which is a VGGNet-16 model or a ResNet-50 model. The bone marrow cell images used in the pre-training model are divided into a training set, a validation set and a test set. The training set is used to train the neural network and update the network weights through a back-propagation algorithm. The validation set is used to optimize model parameters, and hyperparameters including learning rate and batch size are adjusted; The test set is used to evaluate the performance of the model, and the focus indicators include accuracy, recall, and F1 score.

5. The method according to claim 1, characterized in that Based on the cross-degree, a bone marrow cell identification result is generated, include: The calculation formula of the crossover degree C is: ; Among them, b, b z are the center points of the predicted cell bounding box and the actual cell bounding box, represents the intersection of the predicted cell bounding box and the actual cell bounding box, represents the union of the predicted cell bounding box and the actual cell bounding box, represents the Euclidean distance between the predicted cell bounding box and the actual cell bounding box, and c is the diagonal of the minimum enclosing rectangle; When the cross-degree is less than or equal to a preset cross-degree threshold, determining that the bone marrow cell detection and identification is correct; When the cross-degree is greater than the cross-degree threshold, it is determined that the bone marrow cell detection and recognition is wrong.

6. A bone marrow cell morphology artificial intelligence analysis system, characterized in that: Specifically include: An image acquisition and preprocessing module acquires bone marrow cell sample images, preprocesses the sample images, and obtains standardized images; An area of ​​interest image acquisition module is used to extract an area of ​​interest from the standardized image based on the image brightness value and to acquire the area of ​​interest image by using an area of ​​interest image extraction algorithm, wherein the area of ​​interest image extraction algorithm includes an extraction function, and the extraction function is constructed by the Euclidean distance and clustering degree between the image of interest and the image of no interest, and the area of ​​interest is the area where bone marrow cells are concentrated; A neural network recognition module uses a pre-built bone marrow cell morphology recognition neural network to perform feature extraction and bone marrow cell morphology recognition on the image of the region of interest to generate a bone marrow cell recognition result, wherein the bone marrow cell recognition result includes cell type and cell number; The pre-built bone marrow cell morphology recognition neural network is used to perform feature extraction and bone marrow cell morphology recognition on the image of the region of interest to generate a bone marrow cell recognition result, including: In the input stage, the region of interest image is input into the bone marrow cell morphology recognition neural network; In the feature extraction stage, the input image is quickly encoded and learned through multiple convolutional layers to extract the preliminary feature image of the cell; In the feature selection and enhancement stage, a pooling layer is used to extract high-frequency features and low-frequency features of bone marrow cells from the preliminary feature image; In the attention mechanism application stage, the high-frequency features and the low-frequency features are feature encoded by convolution to generate an attention feature distribution, and a dot multiplication operation is performed with the preliminary feature image; Cell detection and positioning stage: Calculate the intersection between the predicted cell bounding box and the actual cell bounding box, and generate a bone marrow cell recognition result based on the intersection.

7. The system according to claim 6, characterized in that The image acquisition and preprocessing module specifically includes: The normalization processing module first resizes and converts the grayscale of the collected sample image, and then performs normalization processing to obtain a normalized image; The filtering processing module performs filtering processing on the normalized image based on mathematical morphology to remove noise so as to obtain the standardized image.

8. The system according to claim 6, characterized in that The region of interest image extraction algorithm specifically includes the following steps: S21: define a brightness value δ, and divide the pixels of the standardized image into an image of interest and an image of non-interest, wherein the image of interest and the image of non-interest are respectively expressed as: ; ; in, , represents the image of interest and the image of no interest respectively, P and Q represent the length and width of the standardized image respectively, F represents the number of brightness values ​​of the standardized image, and the brightness value of the standardized image belongs to , Indicates the number of occurrences of brightness value i; S22: Calculate the Euclidean distance D between the image of interest and the image of no interest, specifically: ; Wherein, the Euclidean distance D represents the degree of distinction between the image of interest and the image of no interest; S23: Calculate the clustering degree of the image of interest and the image of no interest, specifically: ; ; in, , Respectively represent the clustering degrees of the image of interest and the image of no interest; S24: According to the Euclidean distance D and the clustering degree of the image of interest and the clustering degree of the non-interested image , construct the region of interest image extraction function L, specifically: ; S25: Calculate the optimal brightness value δ corresponding to the maximum value of the image extraction function L of the region of interest opt , extract the region of interest image.

9. The system according to claim 6, characterized in that The bone marrow cell morphology recognition neural network is a pre-trained model, which is a VGGNet-16 model or a ResNet-50 model. The pre-training model divides the manually labeled bone marrow cell images into a training set, a validation set, and a test set, wherein the training set is used to train the neural network and update the network weights through a back-propagation algorithm; The validation set is used to optimize model parameters, and hyperparameters including learning rate and batch size are adjusted; The test set is used to evaluate the performance of the model, and the focus indicators include accuracy, recall, and F1 score.

10. The system according to claim 6, characterized in that The crossover degree C is specifically: ; Among them, b, b z are the center points of the predicted cell bounding box and the actual cell bounding box, represents the intersection of the predicted cell bounding box and the actual cell bounding box, represents the union of the predicted cell bounding box and the actual cell bounding box, represents the Euclidean distance between the predicted cell bounding box and the actual cell bounding box, and c is the diagonal of the minimum enclosing rectangle; When the cross-degree is less than or equal to a preset cross-degree threshold, determining that the bone marrow cell detection and identification is correct; When the cross-degree is greater than the cross-degree threshold, it is determined that the bone marrow cell detection and recognition is wrong.