A heart organ abnormality rapid detection method based on search space pruning strategy

Abstract

The present application relates to the technical field of intelligent medical treatment, and particularly relates to a heart organ abnormality rapid detection method based on a search space pruning strategy, which obtains a region of interest on a heart CT image, and obtains a heart image by edge recognition and cutting in the region of interest, so as to effectively reduce the segmentation area of the image, improve the time performance of the segmentation operation while maintaining the segmentation accuracy, and perform a pruning strategy in a search space, so as to use only the data of n heart image classes with the lowest class difference degree with a patient heart correction image of a patient to be detected to construct a convolutional neural network model, so as to accelerate the model training speed. The present application can play an early screening role in the process of heart organ abnormality diagnosis, and is helpful for assisting in determining a diagnosis and treatment result and helping to shorten a diagnosis cycle.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical treatment, and particularly relates to a heart organ abnormality rapid detection method based on a search space pruning strategy. BACKGROUND

[0002] In recent years, with the wide application of intelligent medical treatment, artificial intelligence algorithm models based on CT images and other related medical data have achieved good results in various disease early screening (Yang Shunxiang. Disease early screening method based on multi-modal medical images [D]. Shandong University, 2021.). The heart is one of the most important organs in the human body. In order to improve the diagnosis accuracy and prediction efficiency of heart-related diseases, many engineering and technical personnel have conducted a lot of research on related artificial intelligence methods.

[0003] Before processing the heart image data, it is necessary to extract the key area for processing. At present, there are many application researches on artificial intelligence methods related to heart image recognition and segmentation, for example, Qiao Yan uses a convolutional neural network to segment the left ventricle of a heart MR image and realizes automatic segmentation (Qiao Yan. Key technology research on left ventricular myocardial segmentation of heart MRI image based on full convolution network [D]. Nanjing University of Posts and Telecommunications, 2022.). Fu Dianchen realizes cross-modal medical image domain adaptation based on a generative adversarial network, and fuses a traditional Unet model to achieve good results in the task of heart image segmentation (Fu Dianchen. Research on cross-modal heart medical image domain adaptive segmentation algorithm [D]. Changchun University of Technology, 2021.). In recent years, neural networks have achieved very good results in many complex tasks, but the computing power and operation time consumed are also very amazing.

[0004] In summary, for the research problem of heart organ abnormality rapid detection, it is necessary to further improve the detection efficiency and early screening accuracy. Heart organ abnormality rapid detection can play a role in early screening in the process of heart organ abnormality diagnosis, which helps to assist in determining the diagnosis and treatment results and helps to shorten the diagnosis cycle, and has important practical significance. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a heart organ abnormality rapid detection method based on a search space pruning strategy to solve the problem of low heart organ abnormality detection efficiency.

[0006] In order to achieve the above purpose, the present application provides a heart organ abnormality rapid detection method based on a search space pruning strategy, which comprises the following steps:

[0007] S1, acquiring a heart CT image of a patient to be detected;

[0008] S2, performing edge recognition on the CT image of the heart of the patient to be detected to extract a region of interest image including only the heart of the patient to be detected, and obtaining a raw image of the heart of the patient to be detected;

[0009] S3, performing a correction operation on the raw image of the heart of the patient to be detected to obtain a corrected image of the heart of the patient to be detected;

[0010] S4, performing a pruning strategy in a search space, classifying the corrected images of the hearts of all patients in the database according to corresponding heart-related disease types to obtain a plurality of heart image classes;

[0011] S5, for each heart image class, calculating a mean heart image of the class, comparing the corrected image of the heart of the patient to be detected with the mean heart image of each class, and calculating a class difference degree;

[0012] S6, extracting n heart image classes with the smallest difference degrees, and using data of the n heart image classes to construct a convolutional neural network;

[0013] S7, inputting the corrected image of the heart of the patient to be detected into the convolutional neural network to calculate probability values of the patient to be detected belonging to each heart image class.

[0014] Preferably, in step S2, the edge recognition is performed, and the extraction of the region of interest image includes the following sub-steps:

[0015] S21, constructing a standard Faster R-CNN neural network algorithm model, and using the model to perform target recognition on the input CT image of the heart of the patient to extract a region of interest image including only the heart of the patient;

[0016] S22, using a Sobel edge detection algorithm to further recognize the obtained rectangular region of interest image, so as to further obtain a heart edge in an irregular shape based on the rectangular region of interest image;

[0017] S23, performing image cropping based on the obtained heart edge to obtain a raw image of the heart of the patient.

[0018] Preferably, in step S3, the correction operation on the raw image includes:

[0019] S31, expanding the raw image of the heart of the patient into a regular circular image;

[0020] S32, obtaining an example image, and calling a PSO particle swarm optimization algorithm to perform a rotation operation with the objective of minimizing the difference between the raw image of the heart of the patient expanded into a regular circle and the example heart image;

[0021] S33, adjust the brightness of the image so that the mean value of all pixel points in the image is the mean value of the minimum and maximum allowed values.

[0022] Preferably, step S31 specifically comprises the following sub-steps:

[0023] convert the original image of the patient's heart into a single-channel gray-scale image;

[0024] extend the obtained gray-scale image into a 2000DPI*2000DPI regular circular image, and the smallest rectangle that can contain the image is referred to as the outer rectangle, and the intersection of the two diagonals of the outer rectangle is referred to as the center of the image. During the extension, the center of the image remains unchanged, and other pixel points are uniformly scaled in the direction of any one day diameter to achieve the extension operation.

[0025] Save the image in PNG format, and store each pixel point using 8 bits of data. The pixel points outside the regular circular image and within the outer rectangle are saved in a 100% transparent form.

[0026] Preferably, in step S32, the rotation angle of the original patient's heart pattern extended into a regular circle is determined according to the following steps:

[0027] Obtain ten standard 2000DPI*2000DPI regular circular heart example images;

[0028] Construct an optimization model based on the PSO particle swarm optimization algorithm, wherein the target function to be optimized is:

[0029]

[0030] wherein G is the target function to be optimized, I is the number of samples in the heart image database, i is the sample number of the current comparison, which takes the values 1, 2, 3, 4, 5, 6, 7, 8, 9 or 10, std(·) is a standard deviation calculation function, R represents the rotation angle, Set i (R) is a set of absolute values of the difference between each pixel point in the original patient's heart pattern extended into a regular circle of the current patient and the gray-scale value of the corresponding pixel point in the i-th sample in the example heart image database under the condition of rotation R degrees, n represents the number of pixel points in the 2000DPI*2000DPI regular circular heart image, j represents the number of the currently processed pixel point when the pixel points in the 2000DPI*2000DPI regular circular heart image are calculated in the order from left to right and from top to bottom and numbered from 1, and P(j, R) represents the gray-scale value of the j-th pixel point of the original patient's heart pattern extended into a regular circle of the current patient under the condition of rotation R degrees.S i (j) represents the gray-scale value of the j-th pixel point of the i-th sample in the example heart image database;

[0031] The parameters of the PSO algorithm are set as: inertia weight W=0.4, the PSO algorithm is initialized according to the following parameters: population size N=100, maximum evaluation number maxFE=1000, minimization of the objective function G as the optimization direction, and the evaluation number FE=0 in the initial state;

[0032] The evaluation number FE is increased by one each time the value of the objective function G is calculated, and the PSO algorithm ends when FE is greater than or equal to maxFE;

[0033] The optimal individual obtained from the PSO algorithm is extracted, and the optimal rotation angle R is obtained by decoding the optimal individual, wherein R=0 represents no rotation, a positive value represents clockwise rotation, and a negative value represents counterclockwise rotation.

[0034] Preferably, step S5 includes the following sub-steps:

[0035] S51, for each heart image class, the mean heart image of the class is calculated, and when calculating, the mean value of the pixel points of all patient heart CT images in the heart image class is taken as the mean value of the corresponding position pixel points in the mean heart image;

[0036] S52, the patient heart correction image of the patient to be detected is compared with the mean heart image of each class, and the class difference degree is calculated according to the following formula:

[0037]

[0038] Wherein, k is the number of the class, CD k represents the class difference degree of the patient heart correction image of the patient to be detected and the mean heart image of the kth class, std(·) is the standard deviation calculation function, R best is the optimal rotation angle for performing the rotation operation, Set k (R best ) is the absolute value of the difference between the gray value of each pixel point in the obtained patient heart correction image of the patient whose original heart shape is extended to a positive circle and the gray value of the kth class in the case of rotation R best degree, n represents the number of pixel points in the positive circular heart image of 2000DPI*2000DPI, q represents the number of the pixel point currently processed when the pixel points in the positive circular heart image of 2000DPI*2000DPI are calculated in the order from left to right and from top to bottom and numbered from 1, and P(q, R best ) represents the absolute value of the difference between the gray value of each pixel point in the obtained patient heart correction image of the patient whose original heart shape is extended to a positive circle and the gray value of the kth class in the case of rotation R bestIn the case of obtaining the corrected image of the heart of the patient, for each pixel point in the corrected image of the heart of the patient, the result of extracting the gray value of the qth pixel point is Ave k (q) represents the gray value of the qth pixel point of the mean heart image of the kth class.

[0039] Preferably, before classifying the corrected images of the hearts of all patients in the database according to the corresponding heart-related disease types in step S4, the method further comprises the following sub-steps:

[0040] S41, checking all the heart CT images of the patients in the database, and setting the set of the corresponding row numbers of the images in the database in which the edge recognition and correction operation is not performed as SC;

[0041] S42, checking whether the set SC satisfies:

[0042] length(SC) = 0

[0043] If satisfied, ending the image processing, otherwise going to step S43, wherein length(·) is an element number calculation function, that is, calculating the number of elements of the input set, and the value is 0 when the input is an empty set;

[0044] S43, taking the first element in the set SC as the row number and inputting it, and calling the edge recognition and correction operation in steps S2 and S3 to obtain the corrected image of the heart of the patient, and deleting the first element in the set SC.

[0045] Preferably, after step S7, the method further comprises:

[0046] Performing an archiving operation on the data of the current patient.

[0047] Preferably, before step S1, the method further comprises:

[0048] S0, loading the patient information database and the patient heart CT image database required for constructing the heart organ abnormality rapid detection model, and the data entry in the patient heart CT image database includes a patient ID and a patient heart CT image pointer, and the pointer is used to index the storage location of the patient heart CT image;

[0049] Step S1 specifically comprises the following sub-steps:

[0050] S11, inserting an empty row above the first row of the patient heart CT image database, storing the patient ID and the patient heart CT image pointer of the patient to be detected in the empty row, calling the constructed heart organ abnormality rapid detection model, inputting the detection row number 1, and executing the detection command to obtain the patient heart CT image file of the patient.

[0051] Preferably, the data entries in the patient information database include patient ID, patient gender, patient age, heart-related disease type, medical history, entry date, and the information of each patient is stored as a row of data. For patients diagnosed as healthy or having no disease related to the heart organ, the data entry of heart-related disease type is stored as "normal", otherwise, the name of the heart-related disease and its medical judgment level are stored.

[0052] The present application provides a heart organ abnormality rapid detection method based on a search space pruning strategy. The method obtains a region of interest on a heart CT image, and crops a heart image in the region of interest through edge recognition, thereby effectively reducing the segmentation area of the image, improving the time performance of the segmentation operation while maintaining the segmentation accuracy, and accelerating the model training speed by performing a pruning strategy in the search space and only using the data of the n heart image classes with the lowest class difference degree with the patient heart correction image of the patient to be detected to construct a convolutional neural network model. The present application can play an early screening role in the heart organ abnormality diagnosis process, and is helpful for assisting in the diagnosis and treatment result judgment and shortening the diagnosis cycle. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only illustrate the present application, and other drawings can also be obtained by those skilled in the art without any creative effort.

[0054] Figure 1 The present application provides a heart organ abnormality rapid detection method based on a search space pruning strategy. The method obtains a region of interest on a heart CT image, and crops a heart image in the region of interest through edge recognition, thereby effectively reducing the segmentation area of the image, improving the time performance of the segmentation operation while maintaining the segmentation accuracy, and accelerating the model training speed by performing a pruning strategy in the search space and only using the data of the n heart image classes with the lowest class difference degree with the patient heart correction image of the patient to be detected to construct a convolutional neural network model. The present application can play an early screening role in the heart organ abnormality diagnosis process, and is helpful for assisting in the diagnosis and treatment result judgment and shortening the diagnosis cycle. DETAILED DESCRIPTION

[0055] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only illustrate the present application, and other drawings can also be obtained by those skilled in the art without any creative effort.

[0056] As shown in the drawings, Figure 1 A heart organ abnormality rapid detection method based on a search space pruning strategy includes the following steps:

[0057] S1, obtaining a heart CT image of a patient to be detected;

[0058] S2, performing edge recognition on the heart CT image of the patient to be detected, extracting a region of interest image including only the heart of the patient to be detected, and obtaining a heart original image of the patient to be detected;

[0059] For example, performing edge recognition and extracting the region of interest image includes the following sub-steps:

[0060] S21, construct a standard Faster R-CNN neural network algorithm model, and use it to perform target recognition operation on the input patient heart CT image to extract a region of interest image including only the patient heart;

[0061] S22, for the obtained rectangular region of interest image, further identify using a Sobel edge detection algorithm, so as to further obtain a heart edge of irregular shape on the basis of the rectangular region of interest image;

[0062] S23, image cutting according to the obtained heart edge to obtain a patient heart original image.

[0063] S3, correct the patient heart original image to be detected to obtain a patient heart corrected image to be detected;

[0064] For example, the correction operation on the original image includes the following sub-steps:

[0065] S31, expand the patient heart original image into a regular circular image;

[0066] S32, obtain an example image, and call a PSO particle swarm optimization algorithm to perform a rotation operation with the minimum difference between the patient heart original image expanded into a regular circular image and the example heart image as the target;

[0067] S33, adjust the brightness of the image so that the mean value of all pixel points in the image is the mean value of the minimum value and the maximum value allowed.

[0068] By using the Faster R-CNN neural network algorithm to obtain the region of interest on the heart CT image, and using the Sobel edge detection algorithm to crop the heart image in the region of interest, the segmentation area of the image is effectively reduced, and the time performance of the segmentation operation is improved while maintaining the segmentation accuracy.

[0069] Wherein, as an embodiment, step S31 specifically includes the following sub-steps:

[0070] Convert the patient heart original image into a single-channel grayscale image;

[0071] Expand the obtained grayscale image into a 2000DPI*2000DPI regular circular image, the smallest rectangle that can contain the image is called an outer rectangle, and the intersection of the two diagonals of the outer rectangle is called the center of the image. During expansion, the center of the image remains unchanged, and other pixel points are uniformly scaled in any one-day diameter direction to achieve expansion.

[0072] The image is saved in PNG format, and each pixel point is stored in 8 bits of data. The pixel points outside the circular image and in the outer square are saved in 100% transparent form.

[0073] As an embodiment, in step S32, the rotation angle of the patient's original heart shape extended into a circle is determined according to the following steps:

[0074] Ten standard 2000DPI*2000DPI circular heart sample images are obtained;

[0075] An optimization model based on the PSO particle swarm optimization algorithm is constructed, wherein the target function to be optimized is:

[0076]

[0077] Wherein, G is the target function to be optimized, I is the number of samples in the heart image database, i is the number of the current sample being compared, which takes the values 1, 2, 3, 4, 5, 6, 7, 8, 9 or 10, std(·) is the standard deviation calculation function, R represents the angle of rotation, Set i (R) is the set of absolute values of the difference between the gray value of each pixel point in the patient's original heart shape extended into a circle and the corresponding pixel point in the i-th sample in the sample heart image database, n represents the number of pixel points in the 2000DPI*2000DPI circular heart image, j represents the number of the current pixel point being processed, P(j, R) represents the gray value of the j-th pixel point in the patient's original heart shape extended into a circle at a rotation angle of R degrees.S i (j) represents the gray value of the j-th pixel point in the i-th sample in the sample heart image database;

[0078] The parameters of the PSO algorithm are set as follows: inertia weight W=0.4, the PSO algorithm is initialized according to the following parameters: population size N=100, maximum evaluation number maxFE=1000, the minimization of the target function G is taken as the optimization direction, and the PSO algorithm is run. In the initial state, the evaluation number FE=0;

[0079] Each time the value of the target function G is calculated, the evaluation number FE is incremented by one, and when FE is greater than or equal to maxFE, the PSO algorithm ends running;

[0080] The optimal individual obtained from the PSO algorithm is extracted, and the optimal individual is decoded to obtain the best rotation angle R, wherein R=0 represents no rotation, a positive value represents clockwise rotation, and a negative value represents counterclockwise rotation.

[0081] By extending the irregularly shaped heart image to a regular circle for storage and using the PSO optimization algorithm to calibrate the rotation angle, the influence of patient posture, angle and other factors on the detection result is avoided.

[0082] S4, a pruning strategy is performed in the search space, and the heart correction images of all patients in the database are classified according to the corresponding heart related disease types to obtain a plurality of heart image classes;

[0083] S5, for each heart image class, the mean heart image of the class is calculated, and the heart correction image of the patient to be detected is compared with the mean heart image of each class to calculate the class difference degree;

[0084] For example, step S5 includes the following sub-steps:

[0085] S51, for each heart image class, the mean heart image of the class is calculated, and for each pixel point in the mean heart image, the mean value of the corresponding position pixel points of all patient heart CT images in the heart image class is taken;

[0086] S52, the patient heart correction image of the patient to be detected is compared with the mean heart image of each class, and the class difference degree is calculated according to the following formula:

[0087]

[0088] Wherein, k is the number of the class, CD k represents the class difference degree of the patient heart correction image of the patient to be detected and the mean heart image of the kth class, std(·) is the standard deviation calculation function, that is, the standard deviation of the input set is calculated, R best is the optimal rotation angle for performing the rotation operation, Set k (R best ) is the set of absolute values of the difference between each pixel point in the patient heart correction image obtained by extending the patient heart original image to a regular circle at a rotation angle R best of the current patient and the gray value of the mean heart image of the kth class, n represents the number of pixel points in the 2000DPI*2000DPI regular circle heart image, q represents the number of the pixel point currently processed when the pixel points in the 2000DPI*2000DPI regular circle heart image are calculated in the order from left to right and from top to bottom and numbered from 1, P(q, R best) represents the current patient's patient heart original pattern extended as a positive circle in the case of rotating R best k (q) represents the gray value of the qth pixel point of the mean heart image of the kth class.

[0089] S6, extract n heart image classes with the smallest difference, and use the data of the n heart image classes to construct a convolutional neural network.

[0090] For example, n=5, to speed up the search speed by reducing the search space.

[0091] S7, input the patient heart correction image to be detected into the convolutional neural network, and calculate the probability value of the patient heart correction image to be detected belonging to each heart image class.

[0092] For example, in the convolutional neural network, the convolutional layer is connected with the pooling layer, and then there is a convolutional layer and a pooling layer. The last pooling layer is connected with two consecutive fully connected layers, and the last output layer uses the Softmax activation function. The output layer outputs a one-dimensional vector of 1*n, and the ith element of the vector is used to represent the probability value of the current patient's patient heart CT image belonging to the ith heart image class. If n=5, the output layer outputs a one-dimensional vector of 1*5.

[0093] As an implementation, after step S7, the method further comprises:

[0094] S8, perform archiving operation on the data of the current patient. The doctor uses the rapid detection result to assist in diagnosis, and obtains the final diagnosis result of the patient. An empty row is inserted above the first row of the patient information database, and the patient ID, patient gender, patient age, heart-related disease type, medical history, and entry date of the current patient are stored in the inserted row.

[0095] As an implementation, before step S1, the method further comprises:

[0096] ​S0, load the patient information database, the patient heart CT image database and the example heart image database required for constructing the heart organ anomaly rapid detection model. The data entries in the patient information database include patient ID, patient gender, patient age, heart related disease type, medical history, entry date, and the information of each patient is stored as a row of data. For patients diagnosed as healthy or having no heart related diseases, the data entry of heart related disease type should store the word "normal", otherwise, the name of the heart related disease and its medical judgment grade should be stored. The data entries in the patient heart CT image database include patient ID and patient heart CT image pointer, which is used to index the storage location of the patient heart CT image. Ten standard 2000DPI*2000DPI circular heart images are stored in the example heart image database;

[0097] Step S1 specifically includes the following sub-steps:

[0098] S11, insert an empty row above the first row of the patient heart CT image database, store the patient ID and patient heart CT image pointer of the patient to be detected in the empty row, call the constructed heart organ anomaly rapid detection model, input the detection row number 1, and execute the detection command to obtain the patient heart CT image file of the patient.

[0099] Then, when performing the edge recognition and correction operation on the patient heart CT image, the model automatically obtains the input row number and indexes the row in the patient heart CT image database to obtain the patient heart CT image file of the patient.

[0100] As an implementation, before classifying the heart corrected images of all patients in the patient heart CT image database according to their corresponding heart related disease types in step S4, the following sub-steps are further included:

[0101] S41, check all patient heart CT images in the database, and set the set of row numbers corresponding to the images in the database that have not undergone edge recognition and correction operation in the set SC;

[0102] S42, check whether the set SC satisfies:

[0103] length(SC)=0

[0104] If satisfied, end the image processing, otherwise go to step S43, wherein length(·) is an element number calculation function, that is, the element number of the input set is calculated, and the value is 0 when the input is an empty set;

[0105] S43, take out the first element in the set SC, input its value as the row number and call the edge recognition and correction operation in steps S2 and S3 to obtain the patient heart corrected image, and delete the first element in the set SC.

[0106] After completion, all patient heart CT images in the database are completed with the edge recognition and correction operation.

[0107] To sum up, the application provides a heart organ abnormality rapid detection method based on a search space pruning strategy, which adopts a Faster R-CNN neural network algorithm to obtain a region of interest on a heart CT image, and adopts a Sobel edge detection algorithm to crop a heart image in the region of interest; the irregularly shaped heart image is expanded into a regular circle for storage, and a PSO optimization algorithm is used to calibrate the rotation angle, so as to avoid the influence of patient posture, angle and other factors on the detection result; a pruning strategy is performed in the search space, only the data of the five heart image classes with the lowest class difference degree of the patient heart corrected image of the patient to be detected are used to construct a convolutional neural network model, so as to speed up the model training speed. The application can play a role in early screening in the process of heart organ abnormality diagnosis, and is helpful for assisting in the judgment of diagnosis and treatment results and helping to shorten the diagnosis period.

[0108] Those skilled in the art will understand that the discussion of any of the above embodiments is merely exemplary and is not intended to suggest that the scope of the application (including the claims) is limited to these examples; the above embodiments or technical features among different embodiments can also be combined, steps can be implemented in any order, and there are many other changes of the different aspects of the application as described above, which are not provided in detail for the sake of brevity.

[0109] The present application is intended to cover all such alternatives, modifications and variations as fall within the broad scope of the appended claims. Accordingly, any and all such modifications, variations or equivalents that fall within the spirit and scope of the application are intended to be included within the scope of the application.

Claims

1. A rapid detection method for cardiac organ abnormalities based on a search space pruning strategy, characterized in that, The method includes the following steps: S1. Acquire cardiac CT images of the patient to be tested; S2. Perform edge recognition on the cardiac CT image of the patient to be tested, and extract the region of interest image that includes only the heart of the patient to be tested, to obtain the original image of the heart of the patient to be tested. S3. Perform a correction operation on the original heart image of the patient to be tested to obtain a corrected heart image of the patient to be tested. S4. Perform a pruning strategy in the search space to classify the cardiac correction images of all patients in the database according to their corresponding cardiac-related disease types, resulting in multiple cardiac image classes. S5. For each cardiac image class, calculate the mean cardiac image of that class, compare the corrected cardiac image of the patient to be tested with the mean cardiac image of each class, and calculate the class difference. Includes the following sub-steps: S51. For each cardiac image class, calculate the mean cardiac image of that class. When calculating, for each pixel in the mean cardiac image, take the mean of the corresponding pixel in all patient cardiac CT images in that cardiac image class. S52. Compare the patient's corrected cardiac image with the mean cardiac image for each class, and calculate the class dissimilarity using the following formula: Where k is the class number, CD k This represents the class dissimilarity between the patient's corrected cardiac image and the mean cardiac image of the k-th class, where std(·) is the standard deviation calculation function, and Rk is the standard deviation. best To determine the optimal rotation angle for performing the rotation operation, Set k (R best The original image of the patient's heart is expanded into a perfect circle by rotating R. best In the case of a certain degree, for each pixel in the obtained corrected image of the patient's heart, the set of absolute values ​​of the differences between the gray values ​​and the mean gray value of the heart image of the k-th class, where n represents the number of pixels in a 2000DPI*2000DPI circular heart image, q represents the number of the currently processed pixel in the 2000DPI*2000DPI circular heart image, calculated from left to right and top to bottom, starting from 1, and P(q,R) best The original image of the patient's heart, expanded into a perfect circle, is rotated by R. best In the case of a certain degree, for each pixel in the obtained corrected image of the patient's heart, the gray value of the q-th pixel is extracted, Ave k (q) represents the gray value of the q-th pixel in the heart image, which represents the mean of the k-th class. S6. Extract the n heart image classes with the smallest differences, and use the data from these n heart image classes to construct a convolutional neural network; S7. Input the corrected cardiac image of the patient to be detected into the convolutional neural network and calculate the probability value of the patient to be detected belonging to each cardiac image class.

2. The rapid detection method for cardiac organ abnormalities based on search space pruning strategy according to claim 1, characterized in that, In step S2, edge recognition is performed to extract the image of interest, which includes the following sub-steps: S21. Construct a standard Faster R-CNN neural network algorithm model and use it to perform target recognition operations on the input patient cardiac CT images to extract images of regions of interest that include only the patient's heart. S22. For the obtained rectangular region of interest image, the Sobel edge detection algorithm is used for further identification, thereby obtaining the irregularly shaped heart edge based on the rectangular region of interest image; S23. Based on the obtained heart edge, crop the image to obtain the original image of the patient's heart.

3. The method for rapid detection of cardiac organ abnormalities based on search space pruning strategy according to claim 1, characterized in that, Step S3, the correction operation on the original image includes: S31. Expand the patient's original heart image into a circular image; S32. Obtain the example image and call the PSO particle swarm optimization algorithm to perform a rotation operation with the goal of minimizing the difference between the original patient heart image expanded into a perfect circle and the example heart image. S33. Adjust the brightness of the image so that the average value of all pixels in the image is the average of the minimum and maximum allowed values.

4. The method for rapid detection of cardiac organ abnormalities based on search space pruning strategy according to claim 3, characterized in that, Step S31 specifically includes the following sub-steps: Convert the patient's raw heart image into a single-channel grayscale image; The obtained grayscale image is expanded into a 2000DPI*2000DPI circular image. The smallest rectangle that can contain the image is called the outer rectangle, and the intersection of the two diagonals of the outer rectangle is called the center of the image. During expansion, the center of the image remains unchanged, and other pixels are scaled uniformly in the direction of any diameter to achieve the expansion operation. Images are saved in PNG format, with each pixel stored using 8 bits. Pixels outside the circular image and inside the outer rectangle are saved with 100% transparency.

5. The method for rapid detection of cardiac organ abnormalities based on search space pruning strategy according to claim 3, characterized in that, In step S32, the rotation angle of the original patient heart graphic, which expands into a perfect circle, is determined according to the following steps: Obtain ten standard 2000DPI*2000DPI circular heart example images; An optimization model based on the Particle Swarm Optimization (PSO) algorithm is constructed, where the objective function to be optimized is: Where G is the objective function to be optimized, l is the number of samples in the cardiac image database, i is the sample number being compared, and its value is 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10, std(·) is the standard deviation calculation function, R represents the rotation angle, and Set i (R) represents the set of absolute differences between the grayscale values ​​of each pixel in the original circular image of the current patient's heart, rotated by R degrees, and the corresponding pixel in the i-th sample of the example heart image database; n represents the number of pixels in the 2000 DPI * 2000 DPI circular heart image; j represents the pixel number currently being processed in the 2000 DPI * 2000 DPI circular heart image, calculated from left to right and top to bottom, starting from 1; P(j,R) represents the grayscale value of the j-th pixel in the original circular image of the current patient's heart, rotated by R degrees; S i (j) represents the gray value of the j-th pixel in the i-th sample in the example cardiac image database; Set the parameters of the PSO algorithm as follows: inertia weight W = 0.

4. Initialize the PSO algorithm with the following parameters: population size N = 100, maximum number of evaluations maxFE = 1000. With minimizing the objective function G as the optimization direction, run the PSO algorithm. In the initial state, the number of evaluations FE = 0. Each time the objective function G is calculated, the evaluation count FE is incremented by one. When FE is greater than or equal to maxFE, the PSO algorithm ends. The optimal individual is extracted from the PSO algorithm, and the optimal rotation angle R is decoded from the optimal individual. R = 0 means no rotation is needed, a positive value means clockwise rotation is needed, and a negative value means counterclockwise rotation is needed.

6. The method for rapid detection of cardiac organ abnormalities based on search space pruning strategy according to claim 1, characterized in that, In step S4, before classifying the cardiac correction images of all patients in the database according to their corresponding heart-related disease types, the following sub-steps are also included: S41. Examine all patient cardiac CT images in the database and set the set of row numbers in the database corresponding to images that have not undergone edge recognition and correction to SC; S42. Check if set SC satisfies: length(SC) = 0 If satisfied, end image processing; otherwise, proceed to step S43, where length(·) is the element count calculation function, which calculates the number of elements in the input set. When the input set is empty, its value is 0. S43. Take out the first element in set SC, input its value as the row number and call the edge recognition and correction operations in steps S2 and S3 to obtain the corrected image of the patient's heart, and delete the first element in set SC.

7. The method for rapid detection of cardiac organ abnormalities based on search space pruning strategy according to claim 1, characterized in that, After step S7, the method further includes: Perform an archiving operation on the current patient's data.

8. The method for rapid detection of cardiac organ abnormalities based on search space pruning strategy according to claim 1, characterized in that, Before step S1, the method further includes: S0. Load the patient information database and patient cardiac CT image database required to construct the rapid detection model for cardiac organ abnormalities. The data entries in the patient cardiac CT image database include the patient ID and the patient cardiac CT image pointer, which is used to index the storage location of the patient cardiac CT image. Step S1 specifically includes the following sub-steps: S11. Insert a blank line above the first line of the patient's cardiac CT image database, store the patient ID and the patient's cardiac CT image pointer in it, call the constructed rapid detection model for cardiac organ abnormalities, input the detection line number 1, and execute the detection command to obtain the patient's cardiac CT image file.

9. The method for rapid detection of cardiac organ abnormalities based on search space pruning strategy according to claim 8, characterized in that, The data entries in the patient information database include patient ID, patient gender, patient age, type of heart-related disease, medical history, and entry date. Each patient's information is stored as one line of data. For patients who have been diagnosed as healthy or whose disease is unrelated to the heart, the data entry for their type of heart-related disease is stored as "normal". Otherwise, the name of their heart-related disease and its medical assessment level are stored.

Images