A medical image intelligent recognition method and system based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based medical image intelligent recognition method and system, and relates to the field of image data processing and analysis. The present invention collects multiple medical image samples of multiple modality types through a medical image acquisition module, a preprocessing module, a labeling module, a model building module, a recognition module, a node parameter acquisition module, a decision data visualization module and a recognition result calculation module, thereby improving the richness and comprehensiveness of the samples, and having a good recognition ability for false positives and false negatives that may appear in medical images; by setting a node parameter output interface in each final Alexnet neural network model, various parameters of each node can be output, and the decision value can be calculated and visualized for display, thereby helping medical personnel to have an in-depth understanding of the decision-making process of the model, and thus know the reasons for the decision.

Description

Technical Field

[0001] The present invention belongs to the field of image data processing and analysis, and in particular, relates to a medical image intelligent recognition method and system based on artificial intelligence. Background Art

[0002] Chinese patent CN113177923B discloses a medical image content recognition method, electronic device and storage medium. The method first determines several target image layers in the medical image data to be recognized, and then inputs these target image layers into a human body part recognition model to obtain human body part labels. Then, based on these labels and position information, part recognition results are generated. If the model is also used for anatomical structure recognition, recognition results will also be obtained, which helps to improve the recognition accuracy of human body parts and anatomical structures.

[0003] At present, the decision-making process of identifying medical images through artificial intelligence-related algorithms is usually a black box, and it is difficult to explain the reasons for making decisions. Medical personnel need to understand the decision-making basis of the algorithm. However, most artificial intelligence algorithms are relatively complex, and it is difficult to directly understand their operating steps; secondly, medical images may contain false negative or false positive data. If these data are not labeled, the artificial intelligence algorithm will give incorrect recognition results. Summary of the invention

[0004] In view of the problems in the related art, the present invention proposes a medical image intelligent recognition method and system based on artificial intelligence to overcome the above-mentioned technical problems existing in the existing related technology.

[0005] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0006] The present invention is a medical image intelligent recognition method based on artificial intelligence, comprising the following steps:

[0007] S1, the medical image acquisition module acquires multiple medical image samples of multiple modality types; the preprocessing module performs preprocessing operations on the multiple medical image samples;

[0008] The labeling module labels the preprocessed medical image sample matrix, and the labeled data includes false positive data and false negative data;

[0009] S2, the model building module uses the labeled medical image sample matrix to train, optimize and test multiple initial Alexnet neural network models to obtain multiple final Alexnet neural network models;

[0010] S3, the recognition module uses multiple final Alexnet neural network models to recognize the medical image set to be recognized, and obtains a recognition result set;

[0011] The node parameter acquisition module sets a node parameter output interface for each node of each final Alexnet neural network model, and obtains the parameter data of each node after the recognition is completed;

[0012] S4, the decision data visualization module calculates the decision value according to the parameter data of each node, and visualizes the decision value;

[0013] S5, the recognition result calculation module calculates the recognition result of the medical image set to be recognized according to the recognition result set;

[0014] By collecting multiple medical image samples, data support is provided for the subsequent construction of a medical image recognition model; among them, since medical images have multiple modality types, each modality type has its own advantages and can reflect different aspects of the characteristics of the lesions, therefore, in this scheme, multiple medical image samples of multiple modality types are collected, which improves the richness and comprehensiveness of the samples, thereby improving the accuracy of the subsequent construction of the medical image recognition model, and ultimately improving the accuracy of the identification of the medical image set to be identified; by preprocessing multiple medical image samples, the quality of multiple medical image samples is improved, which facilitates the subsequent data labeling of medical image samples and the training and testing of the medical image recognition model; and then by labeling the preprocessed medical image sample matrix, the medical image recognition model can identify each feature in the medical image sample when training the medical image recognition model, and ultimately obtain a medical image recognition model with good recognition ability for the medical image samples; among them, due to the labeling The data includes false positive data and false negative data, which improves the recognition feature dimension of medical image samples, so that the medical image recognition model finally constructed can also have good recognition ability for false positive and false negative situations that may appear in medical images, further improving the recognition ability of the medical image recognition model; by using labeled medical image sample matrices to train, optimize and test multiple initial Alexnet neural network models, the final Alexnet neural network model obtained has good recognition ability for medical images; by setting a node parameter output interface in each final Alexnet neural network model, the parameters of each node can be output when the final Alexnet neural network model identifies the medical image to be identified, and the decision value is calculated based on these parameters and visualized, which can help medical personnel to have an in-depth understanding of the decision-making process of the model, so as to know the reasons for the decision, and provide a decision-making basis for medical personnel to further analyze medical images.

[0015] Preferably, the S1 comprises the following steps:

[0016] S11. Set multiple medical imaging modality types to obtain a modality type set , Indicates the setting modal type, Indicates the total number of set modal types; according to the modal type set In each modality type, multiple medical image sample sets are collected to obtain a medical image sample matrix ;as follows,

[0017] ;

[0018] in, Indicates Medical imaging collection image samples of modality types, Indicates the total number of medical image sets collected;

[0019] Set the medical image sample matrix The medical image label corresponding to each medical image sample set in the image is obtained by , Indicates The labels corresponding to the medical image sets;

[0020] S12, the medical image sample matrix Perform preprocessing operations on each medical image sample in to obtain the preprocessed medical image sample matrix ;as follows,

[0021] ;

[0022] in, Indicates Medical image sample set Medical image samples obtained after preprocessing image samples of modality types;

[0023] S13, the preprocessed medical image sample matrix Data annotation is performed on each medical image sample in the data set; a labeled medical image sample matrix is ​​obtained ;as follows,

[0024] ;

[0025] in, Express Medical image samples obtained after data annotation;

[0026] Preferably, the data annotation includes false positive data and false negative data;

[0027] Preferably, the modality types include X-ray imaging, computed tomography (CT), magnetic resonance imaging (MRI), ultrasound imaging, nuclear medicine imaging, optical imaging, digital subtraction angiography (DSA), functional magnetic resonance imaging (fMRI);

[0028] These modalities have their own characteristics and can provide information at different levels. Multiple modalities are often used in combination to obtain more comprehensive and accurate information. Different modalities have their own advantages and scope of application in different medical fields and diseases. By comprehensively collecting medical imaging data of these modalities, the imaging data can be made richer and more reliable.

[0029] Preferably, the S12 comprises the following steps:

[0030] S121, the medical image sample matrix Perform denoising on each medical image sample in to obtain a denoised medical image sample matrix;

[0031] S122, using a histogram equalization algorithm to perform contrast enhancement processing on the denoised medical image sample set to obtain the pre-processed medical image sample matrix ;

[0032] By performing denoising and contrast enhancement processing on each medical image sample in the medical image sample matrix, each medical image is made clearer and of higher quality.

[0033] Preferably, the S121 comprises the following steps:

[0034] S1211, the medical image sample matrix Perform non-subsampled Contourlet transform on each medical image sample in to obtain the medical image high frequency coefficient matrix set and the medical image low frequency coefficient matrix set ; They are as follows,

[0035] ; ;

[0036] and Respectively represent Medical imaging collection The high-frequency coefficient matrix and low-frequency coefficient matrix of the image samples of each modality type;

[0037] S1212, using a context model to analyze the medical image high frequency coefficient matrix set Each high-frequency coefficient matrix in is separated and processed to obtain a high-frequency coefficient sub-matrix set matrix ;as follows,

[0038] ; ;

[0039] in, Indicates Medical imaging collection The high-frequency coefficient sub-matrix set of the high-frequency coefficient matrix of the image samples of the modality type, ; express The A high-frequency coefficient sub-matrix, Indicates Medical imaging collection The total number of high-frequency coefficient sub-matrices obtained by dividing the high-frequency coefficient matrix of image samples of modality types;

[0040] S1213, the high frequency coefficient sub-matrix set matrix Perform threshold processing on each high frequency coefficient sub-matrix in to obtain a processed high frequency coefficient sub-matrix set matrix;

[0041] S1214, the processed high-frequency coefficient sub-matrix set matrix and the medical image low-frequency coefficient matrix set Performing a non-subsampled Contourlet inverse transform to obtain the denoised medical image sample set;

[0042] The non-subsampled Contourlet transform overcomes the problem of a large range of coefficient variations in the same scale space, breaks its spatial constraints, enhances certain local features that are neglected in image denoising, and greatly increases the adaptive ability of denoising.

[0043] Preferably, S2 comprises the following steps:

[0044] S21. Construction An initial Alexnet neural network model is generated, and a model interface is set for each of the initial Alexnet neural network models to obtain a model interface set. , Indicates An interface to the initial Alexnet neural network model;

[0045] The activation function of each of the initial Alexnet neural network models is set to RELU;

[0046] S22. Setting the first training data ratio set , Indicates The training data ratio set for each modality type; according to the first training data ratio set For the labeled medical image sample matrix Each column of data and medical image label set Perform data division to obtain a medical image training sample matrix, a medical image test sample matrix, a medical image training label set, and a medical image test label set;

[0047] S23. Set the training error accuracy threshold set , Indicates the setting training error accuracy threshold; inputting each column of image data in the medical image training sample matrix and the medical image training label set into the model interface set Each interface corresponds to Trained in an initial Alexnet neural network model;

[0048] During the training process, The training error accuracy of the initial Alexnet neural network model is less than When The training operation of the initial Alexnet neural network model; when When all initial Alexnet neural network models are trained, we get A trained Alexnet neural network model;

[0049] S24. Set the accuracy threshold set , Indicates the setting accuracy threshold; input each column of image data in the medical image test sample matrix and the medical image test label set into the model interface set The corresponding interfaces in A trained Alexnet neural network model is used for testing. After all the trained Alexnet neural network models are tested, the test accuracy set is obtained. , Indicates The test accuracy obtained by testing a trained Alexnet neural network model;

[0050] when When The trained Alexnet neural network model is used as the The final Alexnet neural network model; otherwise, The trained Alexnet neural network model is adjusted and optimized to obtain the A final Alexnet neural network model;

[0051] when After all the tests and optimizations of the trained Alexnet neural network model are completed, we get A final Alexnet neural network model;

[0052] There are large differences between the medical imaging data of each modality type. Therefore, this scheme adopts multiple Alexnet neural network models to train them separately, so that a single Alexnet neural network model can have excellent recognition ability for medical imaging data of a single modality type; then these Alexnet neural network models that can recognize each single modality type of medical imaging data are combined to further improve the accuracy of the final recognition result.

[0053] Preferably, in S24, The trained Alexnet neural network model is adjusted and optimized to obtain the The final Alexnet neural network model consists of the following steps:

[0054] S241, setting the The initial node value set of a trained Alexnet neural network model , and Respectively represent the The first trained Alexnet neural network model The weight and correction value of each node, Indicates the The total number of nodes in the trained Alexnet neural network model;

[0055] S242. Constructing dragonfly populations , represents the first A dragonfly, represents the size of the dragonfly population; sets the maximum number of iterations of the dragonfly population to , the current number of iterations is And the search space dimension is dimension;

[0056] S243, according to the initial node value set Set the initial position of each dragonfly in the dragonfly population to obtain the initial position matrix set , Indicates the number of The initial position matrix of the dragonfly is as follows:

[0057] ;

[0058] in, and Respectively represent the first The initial position matrix of the dragonfly corresponds to The first trained Alexnet neural network model The weight value and position component of the correction value of each node; the calculation formulas are as follows:

[0059] ; ;

[0060] In the formula, and Respectively represent The first trained Alexnet neural network model The lower and upper limits of the weight values ​​of each node; and Respectively represent The first trained Alexnet neural network model The lower and upper limits of the correction value of each node; and Respectively represent the The initial position matrix of the dragonfly corresponds to The first trained Alexnet neural network model The weight value of each node and the random number selected for the correction value;

[0061] S244, setting the dragonfly population The fitness function ;as follows,

[0062] ;

[0063] In the formula, A positive number indicates a protection parameter;

[0064] S245, start the iteration operation, and set the current number of iterations Set to 1; update the position of each dragonfly in the dragonfly population during each iteration; the update formula is as follows,

[0065] ;

[0066] ;

[0067] In the formula, , , and They represent the number of dragonfly populations in the next iteration. The position of the dragonfly, the first dragonfly in the dragonfly population in the current iteration The position of the dragonfly in the next iteration The moving step length of each dragonfly and the The dragonfly's step length; , , , , and They represent the step size change weight, collision avoidance behavior weight, teaming behavior weight, collective behavior weight, predation behavior weight, and enemy avoidance behavior weight respectively; , , , and They represent the collision avoidance behavior step length, teaming behavior step length, collective behavior step length, predation behavior step length, and enemy avoidance behavior step length respectively;

[0068] After each round of iteration, the current number of iterations is Perform plus one;

[0069] S246, when When , stop the iteration operation and output the final position matrix ;as follows,

[0070] ;

[0071] in, and Respectively represent the corresponding The first trained Alexnet neural network model The weight value and position component of the correction value of each node;

[0072] S247, the final position matrix Substitute into In the trained Alexnet neural network model, we get An optimized Alexnet neural network model; the first The image data and the medical image test label set are input into the The optimized Alexnet neural network model was tested to obtain the first test accuracy;

[0073] When the first test accuracy is greater than or equal to , the The optimized Alexnet neural network model is used as the A final Alexnet neural network model; otherwise, continue the iterative operation in S244 until the first test accuracy is greater than or equal to So far, we have obtained A final Alexnet neural network model;

[0074] The dragonfly optimization algorithm has the advantages of fast convergence and jumping out of local optimal solutions, ensuring the constant convergence of the optimization results; therefore, the dragonfly optimization algorithm is used to continuously iteratively optimize the parameters of each node in the Alexnet neural network model, so that the Alexnet neural network model can find a set of optimal parameters with better efficiency, making the accuracy of the medical image recognition results of each modality higher.

[0075] Preferably, S3 comprises the following steps:

[0076] S31. Setting the medical image set to be identified , Indicates The medical images to be identified correspond to the modality types; and the identification feature-value database is constructed;

[0077] exist A node parameter output interface is set at each node in the final Alexnet neural network model to obtain a node parameter output interface matrix ;as follows,

[0078] ;

[0079] in, Indicates The final Alexnet neural network model Node parameter output interface at each node;

[0080] S32, the medical image set to be identified Each medical image in is input into the model interface set Identify in each interface in;

[0081] During the identification process, the interface matrix is ​​output through the node parameters Each node parameter output interface in the output of the medical image set to be identified Multiple feature types, corresponding feature data, corresponding weights and correction values ​​of each medical image in the image are obtained to obtain the feature type matrix , feature data matrix , weight matrix , Correction value matrix And the identification result set , Indicates The final Alexnet neural network model is The result of identifying the medical images to be identified corresponding to the modality type; , , and They are as follows:

[0082] ; ;

[0083] ; ;

[0084] in, , , and They represent the first The first medical image to be identified corresponding to the modality type The feature type, feature value, feature weight and feature correction value of each feature; Represents the medical image set to be identified Middle The total number of features in the medical images to be identified corresponding to each modality type;

[0085] S33, the feature type matrix , feature data matrix , weight matrix and the correction matrix Stored in a recognition feature-value database;

[0086] After the identification of the medical image set to be identified is completed, the feature type matrix, feature data matrix, weight matrix, and correction value matrix are output through multiple node parameter output interfaces. These data can reflect the final decision steps of the Alexnet neural network model and provide data support for the subsequent calculation of the decision value of each node.

[0087] Preferably, S4 comprises the following steps:

[0088] S41, according to the feature data matrix in the identification feature-value database , weight matrix , Correction value matrix Calculate the decision value matrix ;as follows,

[0089] ;

[0090] in, Indicates the first The first medical image to be identified corresponding to the modality type The decision value of a feature; ;

[0091] S42, the decision value matrix Visualize the decision values ​​and corresponding feature types in ;

[0092] By calculating the decision value of each node in the final Alexnet neural network model and visually displaying these decision values, medical personnel can easily observe and analyze these decision data, providing more basis for medical personnel to make subsequent decisions.

[0093] Preferably, S5 comprises the following steps:

[0094] S51, calculating the recognition result set The probability of each recognition result appearing in the result set is obtained , Indicates The probability value of the recognition result appearing in Indicates the total number of recognition results that appear;

[0095] S52: Select the recognition result probability set The recognition result corresponding to the maximum probability value is taken as the medical image set to be recognized The result of the recognition is to obtain the recognition result of the medical image set to be recognized;

[0096] By calculating the probability of occurrence of each final decision result of the Alexnet neural network model, the decision result with the highest probability is selected as the final decision result, so that the final decision result has a quantitative basis and standard, and the purpose of identifying the image to be identified is achieved.

[0097] An artificial intelligence-based medical image intelligent recognition system, comprising a medical image acquisition module, a preprocessing module, a labeling module, a model building module, a recognition module, a node parameter acquisition module, a decision data visualization module and a recognition result calculation module;

[0098] The medical image acquisition module is used to acquire multiple medical image samples of multiple modality types to obtain a medical image sample matrix;

[0099] The preprocessing module is used to perform a preprocessing operation on each medical image sample in the medical image sample matrix to obtain a preprocessed medical image sample matrix;

[0100] The labeling module is used to label the preprocessed medical image sample matrix to obtain a labeled medical image sample matrix;

[0101] The model building module is used to train, optimize and test multiple initial Alexnet neural network models using the labeled medical image sample matrix to obtain multiple final Alexnet neural network models;

[0102] The recognition module is used to use multiple final Alexnet neural network models to recognize the medical image set to be recognized, and obtain a recognition result set;

[0103] The node parameter acquisition module is used to set a node parameter output interface at each node of each final Alexnet neural network model, and obtain the parameter data of each node through the node parameter output interface after the recognition is completed;

[0104] The decision data visualization module is used to calculate the decision value according to the parameter data of each node and visualize the decision value;

[0105] The recognition result calculation module is used to calculate the recognition result of the medical image set to be recognized based on the recognition result set.

[0106] The present invention has the following beneficial effects:

[0107] 1. In the present invention, by collecting multiple medical image samples of multiple modality types, the richness and comprehensiveness of the samples are improved, thereby improving the accuracy of the subsequent construction of the medical image recognition model; the annotated data includes false positive data and false negative data, which improves the recognition feature dimension of the medical image samples, so that it can have better recognition ability for false positives and false negatives that may appear in medical images; by setting a node parameter output interface in each final Alexnet neural network model, the various parameters of each node can be output, and the decision value can be calculated based on these parameters and visualized, which can help medical personnel to have an in-depth understanding of the decision-making process of the model, thereby knowing the reasons for the decision, and providing a decision-making basis for medical personnel to further analyze medical images.

[0108] 2. In the present invention, multiple Alexnet neural network models are used for separate training, so that a single Alexnet neural network model can have excellent recognition ability for medical imaging data of a single modality type, further improving the accuracy of the final recognition result.

[0109] 3. In the present invention, the parameters of each node in the Alexnet neural network model are continuously iterated and optimized by using the dragonfly optimization algorithm. The dragonfly optimization algorithm has the advantages of fast convergence and jumping out of the local optimal solution. The Alexnet neural network model can find a set of optimal parameters with better efficiency, so that the accuracy of the results of medical image recognition of each modality is higher.

[0110] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0111] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.

[0112] Figure 1 The present invention is a schematic diagram of a process of recognizing medical images by an artificial intelligence-based medical image intelligent recognition system. DETAILED DESCRIPTION

[0113] The following will be combined with the drawings in the embodiments of the invention to clearly and completely describe the technical solutions in the embodiments of the invention. Obviously, the described embodiments are only part of the embodiments of the invention, not all of the embodiments. Based on the embodiments in the invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the invention.

[0114] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0115] Embodiment 1

[0116] This embodiment is a medical image intelligent recognition method based on artificial intelligence, comprising the following steps:

[0117] S1, the medical image acquisition module acquires multiple medical image samples of multiple modality types; the preprocessing module performs preprocessing operations on the multiple medical image samples;

[0118] The labeling module labels the preprocessed medical image sample matrix, and the labeled data includes false positive data and false negative data;

[0119] The S1 comprises the following steps:

[0120] S11. Set multiple medical imaging modality types to obtain a modality type set , Indicates the setting modal type, Indicates the total number of set modal types; according to the modal type set In each modality type, multiple medical image sample sets are collected to obtain a medical image sample matrix ;as follows,

[0121] ;

[0122] in, Indicates Medical imaging collection Image samples of modality types, Indicates the total number of medical image sets collected;

[0123] Set the medical image sample matrix The medical image label corresponding to each medical image sample set in the image is obtained by , Indicates The labels corresponding to the medical image sets;

[0124] S12, the medical image sample matrix Perform preprocessing operations on each medical image sample in to obtain the preprocessed medical image sample matrix ;as follows,

[0125] ;

[0126] in, Indicates Medical image sample set Medical image samples obtained after preprocessing image samples of modality types;

[0127] The S12 comprises the following steps:

[0128] S121, the medical image sample matrix Perform denoising on each medical image sample in to obtain a denoised medical image sample matrix;

[0129] The S121 includes the following steps:

[0130] S1211, the medical image sample matrix Perform non-subsampled Contourlet transform on each medical image sample in to obtain the medical image high frequency coefficient matrix set and the medical image low frequency coefficient matrix set ; They are as follows,

[0131] ; ;

[0132] and Respectively represent Medical imaging collection The high-frequency coefficient matrix and low-frequency coefficient matrix of the image samples of each modality type;

[0133] S1212, using a context model to analyze the medical image high frequency coefficient matrix set Each high-frequency coefficient matrix in is separated and processed to obtain a high-frequency coefficient sub-matrix set matrix ;as follows,

[0134] ; ;

[0135] in, Indicates Medical imaging collection The high-frequency coefficient sub-matrix set of the high-frequency coefficient matrix of the image samples of the modality type, ; express The A high-frequency coefficient sub-matrix, Indicates Medical imaging collection The total number of high-frequency coefficient sub-matrices obtained by dividing the high-frequency coefficient matrix of image samples of modality types;

[0136] S1213, the high frequency coefficient sub-matrix set matrix Perform threshold processing on each high frequency coefficient sub-matrix in to obtain a processed high frequency coefficient sub-matrix set matrix;

[0137] S1214, the processed high-frequency coefficient sub-matrix set matrix and the medical image low-frequency coefficient matrix set Performing a non-subsampled Contourlet inverse transform to obtain the denoised medical image sample set;

[0138] S122, using a histogram equalization algorithm to perform contrast enhancement processing on the denoised medical image sample set to obtain the pre-processed medical image sample matrix ;

[0139] S13, the preprocessed medical image sample matrix Data annotation is performed on each medical image sample in the data set; a labeled medical image sample matrix is ​​obtained ;as follows,

[0140] ;

[0141] in, Express Medical image samples obtained after data annotation;

[0142] The data annotation includes false positive data and false negative data;

[0143] S2, the model building module uses the labeled medical image sample matrix to train, optimize and test multiple initial Alexnet neural network models to obtain multiple final Alexnet neural network models;

[0144] The S2 comprises the following steps:

[0145] S21. Construction An initial Alexnet neural network model is generated, and a model interface is set for each of the initial Alexnet neural network models to obtain a model interface set. , Indicates An interface to the initial Alexnet neural network model;

[0146] The activation function of each of the initial Alexnet neural network models is set to RELU;

[0147] S22. Setting the first training data ratio set , Indicates The training data ratio set for each modality type; according to the first training data ratio set For the labeled medical image sample matrix Each column of data and medical image label set Perform data division to obtain a medical image training sample matrix, a medical image test sample matrix, a medical image training label set, and a medical image test label set;

[0148] S23. Set the training error accuracy threshold set , Indicates the setting training error accuracy threshold; inputting each column of image data in the medical image training sample matrix and the medical image training label set into the model interface set Each interface corresponds to Trained in an initial Alexnet neural network model;

[0149] During the training process, The training error accuracy of the initial Alexnet neural network model is less than When The training operation of the initial Alexnet neural network model; when When all initial Alexnet neural network models are trained, we get A trained Alexnet neural network model;

[0150] S24. Set the accuracy threshold set , Indicates the setting accuracy threshold; input each column of image data in the medical image test sample matrix and the medical image test label set into the model interface set The corresponding interfaces in A trained Alexnet neural network model is used for testing. After all the trained Alexnet neural network models are tested, the test accuracy set is obtained. , Indicates The test accuracy obtained by testing a trained Alexnet neural network model;

[0151] when When The trained Alexnet neural network model is used as the The final Alexnet neural network model; otherwise, The trained Alexnet neural network model is adjusted and optimized to obtain the A final Alexnet neural network model;

[0152] when After all the tests and optimizations of the trained Alexnet neural network model are completed, we get A final Alexnet neural network model;

[0153] S24 The trained Alexnet neural network model is adjusted and optimized to obtain the The final Alexnet neural network model consists of the following steps:

[0154] S241, setting the The initial node value set of a trained Alexnet neural network model , and Respectively represent the The first trained Alexnet neural network model The weight and correction value of each node, Indicates the The total number of nodes in the trained Alexnet neural network model;

[0155] S242. Constructing dragonfly populations , represents the first A dragonfly, represents the size of the dragonfly population; sets the maximum number of iterations of the dragonfly population to , the current number of iterations is And the search space dimension is dimension;

[0156] S243, according to the initial node value set Set the initial position of each dragonfly in the dragonfly population to obtain the initial position matrix set , Indicates the number of The initial position matrix of the dragonfly is as follows:

[0157] ;

[0158] in, and Respectively represent the first The initial position matrix of the dragonfly corresponds to The first trained Alexnet neural network model The weight value and position component of the correction value of each node; the calculation formulas are as follows:

[0159] ; ;

[0160] In the formula, and Respectively represent The first trained Alexnet neural network model The lower and upper limits of the weight values ​​of each node; and Respectively represent The first trained Alexnet neural network model The lower and upper limits of the correction value of each node; and Respectively represent the The initial position matrix of the dragonfly corresponds to The first trained Alexnet neural network model The weight value of each node and the random number selected for the correction value;

[0161] S244, setting the dragonfly population The fitness function ;as follows,

[0162] ;

[0163] In the formula, A positive number indicates a protection parameter;

[0164] S245, start the iteration operation, and set the current number of iterations Set to 1; update the position of each dragonfly in the dragonfly population during each iteration; the update formula is as follows,

[0165] ;

[0166] ;

[0167] In the formula, , , and They represent the number of dragonfly populations in the next iteration. The position of the dragonfly, the first dragonfly in the dragonfly population in the current iteration The position of the dragonfly in the next iteration The moving step length of each dragonfly and the The dragonfly's step length; , , , , and They represent the step size change weight, collision avoidance behavior weight, teaming behavior weight, collective behavior weight, predation behavior weight, and enemy avoidance behavior weight respectively; , , , and They represent the collision avoidance behavior step length, teaming behavior step length, collective behavior step length, predation behavior step length, and enemy avoidance behavior step length respectively;

[0168] After each round of iteration, the current number of iterations is Perform plus one;

[0169] S246, when When , stop the iteration operation and output the final position matrix ;as follows,

[0170] ;

[0171] in, and Respectively represent the corresponding The first trained Alexnet neural network model The weight value and position component of the correction value of each node;

[0172] S247, the final position matrix Substitute into In the trained Alexnet neural network model, we get An optimized Alexnet neural network model; the first The image data and the medical image test label set are input into the The optimized Alexnet neural network model was tested to obtain the first test accuracy;

[0173] When the first test accuracy is greater than or equal to , the The optimized Alexnet neural network model is used as the A final Alexnet neural network model; otherwise, continue the iterative operation in S244 until the first test accuracy is greater than or equal to So far, we have obtained A final Alexnet neural network model;

[0174] S3, the recognition module uses multiple final Alexnet neural network models to recognize the medical image set to be recognized, and obtains a recognition result set;

[0175] The node parameter acquisition module sets a node parameter output interface for each node of each final Alexnet neural network model, and obtains the parameter data of each node after the recognition is completed;

[0176] The S3 comprises the following steps:

[0177] S31. Setting the medical image set to be identified , Indicates The medical images to be identified correspond to the modality types; and the identification feature-value database is constructed;

[0178] exist A node parameter output interface is set at each node in the final Alexnet neural network model to obtain a node parameter output interface matrix ;as follows,

[0179] ;

[0180] in, Indicates The final Alexnet neural network model Node parameter output interface at each node;

[0181] S32, the medical image set to be identified Each medical image in is input into the model interface set Identify in each interface in;

[0182] During the identification process, the interface matrix is ​​output through the node parameters Each node parameter output interface in the output of the medical image set to be identified Multiple feature types, corresponding feature data, corresponding weights and correction values ​​of each medical image in the image are obtained to obtain the feature type matrix , feature data matrix , weight matrix , Correction value matrix And the identification result set , Indicates The final Alexnet neural network model is used for The result of identifying the medical images to be identified corresponding to the modality type; , , and They are as follows:

[0183] ; ;

[0184] ; ;

[0185] in, , , and They represent the first The first medical image to be identified corresponding to the modality type The feature type, feature value, feature weight and feature correction value of each feature; Represents the medical image set to be identified Middle The total number of features in the medical images to be identified corresponding to each modality type;

[0186] S33, the feature type matrix , feature data matrix , weight matrix and the correction matrix Stored in a recognition feature-value database;

[0187] S4, the decision data visualization module calculates the decision value according to the parameter data of each node, and visualizes the decision value;

[0188] The S4 comprises the following steps:

[0189] S41, according to the feature data matrix in the identification feature-value database , weight matrix , Correction value matrix Calculate the decision value matrix ;as follows,

[0190] ;

[0191] in, Indicates the first The first medical image to be identified corresponding to the modality type The decision value of a feature; ;

[0192] S42, the decision value matrix Visualize the decision values ​​and corresponding feature types in ;

[0193] S5, the recognition result calculation module calculates the recognition result of the medical image set to be recognized according to the recognition result set;

[0194] The S5 comprises the following steps:

[0195] S51, calculating the recognition result set The probability of each recognition result appearing in the result set is obtained , Indicates The probability value of the recognition result appearing in Indicates the total number of recognition results that appear;

[0196] S52: Select the recognition result probability set The recognition result corresponding to the maximum probability value is taken as the medical image set to be recognized The result of the recognition is to obtain the recognition result of the medical image set to be recognized.

[0197] Embodiment 2

[0198] This embodiment discloses an artificial intelligence-based medical image intelligent recognition system, which can implement the method of the above embodiment, including a medical image acquisition module, a preprocessing module, a labeling module, a model building module, a recognition module, a node parameter acquisition module, a decision data visualization module and a recognition result calculation module;

[0199] The medical image acquisition module is used to acquire multiple medical image samples of multiple modality types to obtain a medical image sample matrix;

[0200] The preprocessing module is used to perform a preprocessing operation on each medical image sample in the medical image sample matrix to obtain a preprocessed medical image sample matrix;

[0201] The labeling module is used to label the preprocessed medical image sample matrix to obtain a labeled medical image sample matrix;

[0202] The model building module is used to train, optimize and test multiple initial Alexnet neural network models using the labeled medical image sample matrix to obtain multiple final Alexnet neural network models;

[0203] The recognition module is used to use multiple final Alexnet neural network models to recognize the medical image set to be recognized, and obtain a recognition result set;

[0204] The node parameter acquisition module is used to set a node parameter output interface at each node of each final Alexnet neural network model, and obtain the parameter data of each node through the node parameter output interface after the recognition is completed;

[0205] The decision data visualization module is used to calculate the decision value according to the parameter data of each node and visualize the decision value;

[0206] The recognition result calculation module is used to calculate the recognition result of the medical image set to be recognized based on the recognition result set.

[0207] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0208] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. A medical image intelligent recognition method based on artificial intelligence, characterized in that: The following steps are involved: S1. The medical image acquisition module acquires multiple medical image samples of multiple modality types; the preprocessing module performs preprocessing operations on the multiple medical image samples; the labeling module labels the preprocessed medical image sample matrix, and the labeled data includes false positive data and false negative data; S2, the model building module uses the labeled medical image sample matrix to train, optimize and test multiple initial Alexnet neural network models to obtain multiple final Alexnet neural network models; S3, the recognition module uses multiple final Alexnet neural network models to recognize the medical image set to be recognized, and obtains a recognition result set; The node parameter acquisition module sets a node parameter output interface for each node of each final Alexnet neural network model, and obtains the parameter data of each node after the recognition is completed; S4, the decision data visualization module calculates the decision value according to the parameter data of each node, and visualizes the decision value; S5, the recognition result calculation module calculates the recognition result of the medical image set to be recognized according to the recognition result set; The S2 comprises the following steps: S21, constructing c′ initial Alexnet neural network models, setting a model interface for each of the initial Alexnet neural network models, and obtaining a model interface set; c′ represents the total number of set modality types; S22, setting a first training data ratio set; performing data division on each column of the labeled medical image sample matrix and the medical image label set according to the first training data ratio set to obtain a medical image training sample matrix, a medical image test sample matrix, a medical image training label set, and a medical image test label set; S23, setting a training error accuracy threshold set; inputting each column of image data in the medical image training sample matrix and the medical image training label set into c′ initial Alexnet neural network models corresponding to each interface in the model interface set for training; During the training process, when the training error accuracy of the i-th initial Alexnet neural network model is less than the corresponding training error accuracy threshold, the training operation of the i-th initial Alexnet neural network model is stopped; when the training of c′ initial Alexnet neural network models is completed, c′ trained Alexnet neural network models are obtained; S24. Set the accuracy threshold set represents the set i-th accuracy threshold; input each column of image data in the medical image test sample matrix and the medical image test label set into the c′ trained Alexnet neural network models corresponding to each interface in the model interface set for testing operation, and when all the c′ trained Alexnet neural network models are tested, the test accuracy set is obtained. represents the test accuracy of the i-th trained Alexnet neural network model; when When , the i-th trained Alexnet neural network model is used as the i-th final Alexnet neural network model; otherwise, the i-th trained Alexnet neural network model is adjusted and optimized to obtain the i-th final Alexnet neural network model; when all c′ trained Alexnet neural network models are tested and optimized, c′ final Alexnet neural network models are obtained.

2. The medical image intelligent recognition method based on artificial intelligence according to claim 1, characterized in that: The S1 comprises the following steps: S11, setting multiple medical image modality types to obtain a modality type set; collecting multiple medical image sample sets according to each modality type in the modality type set to obtain a medical image sample matrix; setting a medical image label corresponding to each medical image sample set in the medical image sample matrix to obtain a medical image label set; S12, performing a preprocessing operation on each medical image sample in the medical image sample matrix to obtain a preprocessed medical image sample matrix; S13, performing data labeling on each medical image sample in the preprocessed medical image sample matrix; obtaining a labeled medical image sample matrix; the data labeling includes false positive data and false negative data.

3. The medical image intelligent recognition method based on artificial intelligence according to claim 2, characterized in that: The S12 comprises the following steps: S121, performing denoising processing on each medical image sample in the medical image sample matrix to obtain a denoised medical image sample matrix; S122, using a histogram equalization algorithm to perform contrast enhancement processing on the denoised medical image sample set to obtain the preprocessed medical image sample matrix.

4. The medical image intelligent recognition method based on artificial intelligence according to claim 3 is characterized in that: The S121 includes the following steps: S1211, performing non-subsampled Contourlet transform on each medical image sample in the medical image sample matrix to obtain a medical image high-frequency coefficient matrix set and a medical image low-frequency coefficient matrix set; S1212, using a context model to separate each high-frequency coefficient matrix in the medical image high-frequency coefficient matrix set to obtain a high-frequency coefficient sub-matrix set matrix; S1213, performing threshold processing on each high frequency coefficient submatrix in the high frequency coefficient submatrix set matrix to obtain a processed high frequency coefficient submatrix set matrix; S1214, performing non-subsampled Contourlet inverse transform on the processed high-frequency coefficient sub-matrix set and the medical image low-frequency coefficient matrix set to obtain the denoised medical image sample set.

5. The medical image intelligent recognition method based on artificial intelligence according to claim 4, characterized in that: In S24, adjusting and optimizing the i-th trained Alexnet neural network model to obtain the i-th final Alexnet neural network model includes the following steps: S241, setting the initial node value set of the i-th trained Alexnet neural network model; S242, constructing a dragonfly population; setting the maximum number of iterations of the dragonfly population to , the current number of iterations is ; S243, setting the initial position of each dragonfly in the dragonfly population according to the initial node value set to obtain an initial position matrix set; S244, setting the fitness function of the dragonfly population; S245, start the iteration operation, and set the current number of iterations Set to 1; in each round of iteration, the position of each dragonfly in the dragonfly population is updated according to the fitness function of the dragonfly population; after each round of iteration, the current number of iterations is updated. Perform plus one; S246, when When , stop the iteration operation and output the final position matrix; S247, substitute the final position matrix into the i-th trained Alexnet neural network model to obtain the i-th optimized Alexnet neural network model; input the i-th column of image data in the medical imaging test sample matrix in S24 and the medical imaging test label set into the i-th optimized Alexnet neural network model for testing to obtain a first test accuracy; when the first test accuracy is greater than or equal to When the i-th optimized Alexnet neural network model is used as the i-th final Alexnet neural network model; otherwise, continue the iterative operation in S244 until the first test accuracy is greater than or equal to So far, the i-th final Alexnet neural network model is obtained.

6. The medical image intelligent recognition method based on artificial intelligence according to claim 5, characterized in that: The S3 comprises the following steps: S31, setting a medical image set to be identified; and constructing an identification feature-value database; setting a node parameter output interface at each node in the c′ final Alexnet neural network model to obtain a node parameter output interface matrix; S32, inputting each medical image in the medical image set to be identified into each interface in the model interface set for identification; during the identification process, outputting multiple feature types, corresponding feature data, corresponding weights and correction values ​​of each medical image in the medical image set to be identified through each node parameter output interface in the node parameter output interface matrix, and obtaining a feature type matrix, a feature data matrix, a weight matrix, a correction value matrix and an identification result set; S33, storing the feature type matrix, feature data matrix, weight matrix and correction value matrix into a recognition feature-value database.

7. The medical image intelligent recognition method based on artificial intelligence according to claim 6, characterized in that: The S4 comprises the following steps: S41, calculating a decision value matrix according to the feature data matrix, weight matrix, and correction value matrix in the identification feature-value database; S42: Visually display each decision value in the decision value matrix and the corresponding feature type.

8. The medical image intelligent recognition method based on artificial intelligence according to claim 7, characterized in that: The S5 comprises the following steps: S51, calculating the probability of occurrence of each recognition result in the recognition result set to obtain a recognition result probability set; S52: Select the recognition result corresponding to the maximum probability value in the recognition result probability set as the result of recognizing the medical image set to be recognized, and obtain the recognition result of the medical image set to be recognized.

9. A system for implementing the medical image intelligent recognition method based on artificial intelligence as claimed in any one of claims 1 to 8, characterized in that: It includes a medical image acquisition module, a preprocessing module, a labeling module, a model building module, a recognition module, a node parameter acquisition module, a decision data visualization module and a recognition result calculation module.

Images