Medical image classification method and system based on deep multi-feature enhancement network, and computer readable storage medium

Abstract

The invention relates to a medical image classification method and system based on a deep multi-feature enhancement network, and a computer storage medium. The method comprises the following steps: acquiring a pancreas CT medical image of a patient; performing image feature extraction on the pancreas CT medical image to obtain morphological feature information, textural feature information and depth feature information corresponding to the pancreas CT medical image; fusing the morphological feature information, the texture feature information and the depth feature information to obtain comprehensive feature information; performing image reconstruction on the comprehensive feature information to obtain enhanced reconstruction image information after the texture feature information and the morphological feature information are enhanced and reconstructed; and based on a Softmax cross entropy loss function, performing classification training on the enhanced reconstructed image information to obtain classification information corresponding to the pancreas CT medical image. According to the embodiment, intelligent grading of the pancreatic CT images of the patient can be automatically achieved, targeted treatment on the patient clinically is effectively assisted, and the clinical cure rate of the patient is increased.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a medical image classification method, system and computer-readable storage medium based on a deep multi-feature enhancement network. Background Art

[0002] Acute pancreatitis is a common disease. This disease comes on suddenly and changes rapidly. If you do not seek medical treatment in time, the condition will quickly turn into severe pancreatitis. The complications of severe pancreatitis are very dangerous. Generally speaking, they are often accompanied by obstruction and failure of important organs such as the heart, lungs, and kidneys, and the mortality rate can reach 30%. Balthazar CT plain scan morphological scoring is commonly used in clinical practice to score pancreatic morphological changes, which is divided into five levels: AE. The above-mentioned grade A pancreas is basically normal in morphology, grade B pancreas is slightly edematous, the pancreas is significantly enlarged, grade C pancreas is edematous, there is peripancreatic exudation, and there is no obvious peripancreatic effusion, grade D is peripancreatic edema, exudation is relatively obvious, accompanied by a single peripancreatic effusion, grade E pancreas is swollen, accompanied by low-density foci inside the pancreas, indicating pancreatic necrosis, and there are multiple fluid accumulations around the pancreas.

[0003] Currently, clinical medicine generally uses manual methods to grade pancreatic CT images, and occasionally uses deep learning technology for grading.

[0004] The main disadvantages of manual scoring are as follows: 1. There are large differences in grading. Although the guidelines have defined each grade, the rating does not have a strict definition, resulting in inconsistent understanding of the guidelines by different doctors; 2. The grading process is highly dependent on expertise, and some junior doctors are often unable to perform the task; 3. The grading workload is large. For critically ill hospitalized patients, a large number of CT images need to be graded during the entire hospitalization period; 4. Grading is often not timely enough, and even a large number of omissions occur, resulting in a waste of valuable clinical process data.

[0005] Using deep learning technology: Although deep learning technology has become a popular technology for medical image feature extraction in recent years, it can automatically extract visual features through convolution operations. However, pancreatic CT image grading involves manually designed features such as pancreatic geometric features and contrast. Due to the poor interpretability of deep learning technology, it is impossible to extract specific features of medical images. Therefore, deep learning technology has not made much progress in the field of pancreatic CT grading. Summary of the invention

[0006] Based on this, it is necessary to provide a medical image classification method based on a deep multi-feature enhancement network, which causes large differences in grading due to manual scoring, relies on the subjective judgment of doctors, and causes waste of clinical process data; currently deep learning technology cannot extract the specific features of medical images of CT images.

[0007] A medical image classification method based on a deep multi-feature enhancement network, comprising:

[0008] Obtain pancreatic CT medical images of patients;

[0009] Performing image feature extraction on the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image;

[0010] Fusing the morphological feature information, texture feature information, and depth feature information to obtain comprehensive feature information;

[0011] Reconstructing the comprehensive feature information to obtain enhanced reconstructed image information after the texture feature information and the morphological feature information are enhanced and reconstructed;

[0012] Based on the Softmax cross entropy loss function, the enhanced and reconstructed image information is classified and trained to obtain the grading information corresponding to the pancreatic CT medical image.

[0013] In one preferred embodiment, the image feature extraction of the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image includes:

[0014] Using a generalized linear model method to extract morphological feature information from the CT medical image to obtain morphological feature information corresponding to the CT medical image;

[0015] In one preferred embodiment, image feature extraction is performed on the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image, including:

[0016] The local binary pattern is used to extract texture features of the CT medical image.

[0017] In one preferred embodiment, image feature extraction is performed on the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image, including:

[0018] The CT medical image is subjected to deep feature extraction by a deep feature extractor, wherein the deep feature extractor is composed of a convolutional layer containing L layers, and the output feature map of each layer can be expressed as:

[0019]

[0020] Among them, w k and b kare the weights and bias parameters of the kth convolutional layer, σ is the activation function, and 1 <k<L。

[0021] In one preferred embodiment, before fusing the morphological feature information, texture feature information and depth feature information to obtain comprehensive feature information, the method further includes:

[0022] The morphological feature information and the texture feature information are processed by dimension reduction through principal component analysis so that the dimensions of the morphological feature information, the texture feature information and the depth feature information are consistent, and the morphological feature information after dimension reduction satisfies:

[0023]

[0024] Among them, i represents a natural number, f f (x i ) represents morphological feature extraction, represents the feature information after dimensionality reduction processing, Represents the original feature information.

[0025] In one preferred embodiment, the morphological feature information, texture feature information and depth feature information are fused to obtain comprehensive feature information, including:

[0026] The morphological feature information, texture feature information and depth feature information are fused by using an element-by-element fusion method, and the comprehensive feature information satisfies:

[0027]

[0028] in, Represents morphological feature information, Represents texture feature information, Represents deep feature information, Represents comprehensive feature information.

[0029] In one preferred embodiment, the reconstruction process of reconstructing the image of the comprehensive feature information is constrained by a reconstruction loss function.

[0030] The above-mentioned embodiment of the present invention uses a medical image classification method based on a deep multi-feature enhancement network to fuse the morphological features, texture features, etc. of the pancreatic CT image with the deep features to form a deep enhancement network, and then optimizes the network through semi-supervised learning. In the case of a small number of labeled samples, an effective classification network is trained to achieve intelligent CT grading of acute pancreatitis. The above-mentioned embodiment can automatically achieve intelligent grading of the patient's pancreatic CT images, effectively assist the clinic in targeted treatment of patients, and improve the clinical cure rate of patients. At the same time, the grading data is stored in the patient's treatment process data in a structured form, thereby better supporting subsequent medical research.

[0031] A grading system for pancreatitis based on CT images includes:

[0032] A CT image acquisition module, used for acquiring a pancreatic CT medical image of a patient;

[0033] A feature extraction module, used to extract image features from the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image;

[0034] A feature fusion module, used to fuse the morphological feature information, texture feature information and depth feature information to obtain comprehensive feature information;

[0035] An image reconstruction module, used for reconstructing the comprehensive feature information to obtain enhanced reconstructed image information after the texture feature information and the morphological feature information are enhanced and reconstructed;

[0036] The classification training module is used to perform classification training on the enhanced reconstructed image information based on the Softmax cross entropy loss function to obtain the classification information corresponding to the pancreatic CT medical image.

[0037] In one preferred embodiment, the feature extraction module includes:

[0038] The CT medical image is subjected to deep feature extraction by a deep feature extractor, wherein the deep feature extractor is composed of a stack of l convolutional layers, and the output feature map of each layer can be expressed as:

[0039]

[0040] Among them, w k and b k are the weights and bias parameters of the kth convolutional layer, σ is the activation function, and 1 <k<L。

[0041] The above-mentioned embodiment of the present invention uses a medical image classification system based on a deep multi-feature enhancement network to fuse the morphological features, texture features, etc. of the pancreatic CT image with the deep features to form a deep enhancement network, and then optimizes the network through semi-supervised learning. In the case of a small number of labeled samples, an effective classification network is trained to achieve intelligent CT grading of acute pancreatitis. The above-mentioned embodiment can automatically achieve intelligent grading of the patient's pancreatic CT images, effectively assist the clinic in targeted treatment of patients, and improve the clinical cure rate of patients. At the same time, the grading data is stored in the patient's treatment process data in a structured form, thereby better supporting subsequent medical research.

[0042] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the medical image classification method based on a deep multi-feature enhancement network described above is implemented.

[0043] The above-mentioned embodiment of the present invention uses a computer-readable storage medium to execute the above-mentioned classification method, fuses the morphological features, texture features, etc. of the pancreatic CT image with the depth features to form a deep enhancement network, and then optimizes the network through semi-supervised learning. In the case of a small number of labeled samples, an effective classification network is trained to achieve intelligent CT grading of acute pancreatitis. Through the above-mentioned implementation, the intelligent grading of the patient's pancreatic CT images can be automatically achieved, effectively assisting the clinic to carry out targeted treatment of the patient and improving the patient's clinical cure rate. At the same time, the grading data is stored in the patient's treatment process data in a structured form, thereby better supporting subsequent medical research. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flowchart of a medical image classification method based on a deep multi-feature enhancement network disclosed in the first preferred embodiment of the present invention;

[0045] Figure 2 This is a flowchart of a medical image classification system based on a deep multi-feature enhancement network disclosed in the second preferred embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] It should be noted that when an element is referred to as being "disposed on" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only and do not represent the only implementation method.

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0049] like Figure 1 As shown, the first preferred embodiment of the present invention discloses a medical image classification method based on a deep multi-feature enhancement network, the method comprising:

[0050] S10: Acquire a pancreatic CT medical image of the patient.

[0051] In this step, the pancreatic CT medical image corresponding to the target patient is acquired through the CT imaging device, and the pancreatic CT medical image is imported into the image acquisition system corresponding to this step to facilitate the analysis in subsequent steps. In this embodiment, the above CT medical image is recorded as Xi.

[0052] S20: performing image feature extraction on the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image;

[0053] In this step, feature extraction is mainly performed on the acquired pancreatic CT medical image. The above feature extraction process mainly includes morphological feature extraction, texture feature extraction and depth feature extraction, and corresponding morphological feature information, texture feature information and depth feature information are obtained respectively.

[0054] Specifically, in the above process of extracting morphological features, the shape, edge contour and other information of the pancreas are mainly extracted. This embodiment uses a generalized linear model method to extract morphological features of the CT medical image. The characteristic of this method is that the natural measurement of the data is not forcibly changed, and the data can have a nonlinear and non-constant variance structure. It is a development of the linear model in studying the non-normal distribution of the response value and the concise and direct linear transformation of the nonlinear model.

[0055] After extracting the morphological feature information obtained by the generalized linear model method, in order to facilitate subsequent fusion, in this implementation, the principal component analysis method is also used to reduce the dimension of the extracted morphological features to ensure that the morphological feature information is consistent with the latitude of the subsequent texture features and depth features. It can be expressed as:

[0056]

[0057] Among them, i above represents a natural number, x i represents the pancreatic CT medical image corresponding to the i-th copy, f f (x i ) represents the morphological feature extraction of the corresponding pancreatic CT medical image, Represents the final morphological feature information, Represents the morphological feature value information of the original input.

[0058] On the other hand, in general, the patient's pancreatic edema, swelling, peripancreatic fluid accumulation and other phenomena will show certain local serial repetition, non-random arrangement and basically uniform texture area in the CT medical image. In this embodiment, the local binary pattern (LBP) is used to extract texture features of the CT medical image. Specifically, in this embodiment, a window with a threshold size of 3×3 is set to analyze and compare the values ​​of the central pixel and its 8 adjacent pixels. The texture feature value in pancreatic CT can be obtained by the LBP function.

[0059] LBP (Local Binary Pattern) is an operator used to describe local texture features of an image; it has significant advantages such as rotation invariance and grayscale invariance. In this embodiment, the above method is used for texture feature extraction. Moreover, the extracted features are local texture features of the image.

[0060] The original LBP operator is defined as taking the central pixel of the window as the threshold within a 3*3 window, comparing the grayscale values ​​of the 8 adjacent pixels with it, and if the surrounding pixel values ​​are greater than the central pixel value, the position of the pixel is marked as 1, otherwise it is 0. In this way, the 8 points in the 3*3 neighborhood can generate an 8-bit binary number (usually converted to a decimal number, namely the LBP code, a total of 256 types) after comparison, that is, the LBP value of the central pixel of the window is obtained, and this value is used to reflect the texture information of the area.

[0061] Similarly, in order to facilitate subsequent fusion, the principal component analysis (PCA) is also used to reduce the dimension of the extracted texture features to ensure consistency with other morphological feature information and depth feature latitude. The final morphological feature hi t It can be expressed as:

[0062]

[0063] Among them, i above represents a natural number, x i represents the pancreatic CT medical image corresponding to the i-th copy, f f (x i ) represents the texture feature extraction of the corresponding pancreatic CT medical image, Represents the final texture feature information, Represents the texture feature value information of the original input.

[0064] The purpose of the PCA (Principal Component Analysis) algorithm is to convert high-dimensional data to low-dimensional data with less "information" loss. By extracting the largest individual differences revealed by the principal components, it can also be used to reduce the number of variables in regression analysis and cluster analysis, thereby reducing the amount of calculation.

[0065] PCA (Principal Component Analysis) is a technique used to explore the structure of high-dimensional data. The idea of ​​PCA is to map the dimensional feature to the dimension, which is a new orthogonal feature. This dimensional feature is called the principal component, which is a reconstructed dimensional feature. In PCA, the data is converted from the original coordinate system to the new coordinate system, and the choice of the new coordinate system is closely related to the data itself. The first new coordinate axis selects the direction with the largest variance in the original data, and the second new coordinate axis selects the direction that is orthogonal to the first coordinate axis and has the largest variance. This process is repeated, and the number of repetitions is the number of features in the original data. Most of the variance is contained in the first few new coordinate axes. Therefore, the remaining coordinate axes can be ignored, that is, the data is subjected to dimensionality reduction.

[0066] The essence of the PCA algorithm is to find some projection directions so that the variance of the data in these projection directions is the largest, and these projection directions are mutually orthogonal. This is actually the process of finding a new orthogonal basis and calculating the variance of the original data projected on these orthogonal bases. The larger the variance, the more information is contained in the corresponding orthogonal basis. The larger the eigenvalue of the covariance matrix of the original data, the larger the corresponding variance, and the greater the amount of information projected on the corresponding eigenvector. On the contrary, if the eigenvalue is small, it means that the amount of information projected on these eigenvectors is very small, and the data in the direction corresponding to the small eigenvalue can be deleted, thereby achieving the purpose of dimensionality reduction.

[0067] PCA synthesizes high-dimensional variables that may be correlated into linearly independent low-dimensional variables, called principal components. The new low-dimensional data set will retain the variables of the original data as much as possible.

[0068] The depth feature extractor is composed of L stacked convolutional layers, and the output feature map of each layer can be expressed as

[0069]

[0070] where w k and b k are the weight and bias parameters of the k-th convolutional layer, σ is the activation function, and 1 < k < L. The final morphological feature h i d can be expressed as:

[0071]

[0072] S30: Fuse the morphological feature information, texture feature information, and depth feature information to obtain comprehensive feature information;

[0073] Specifically, in this step, the morphological feature information, texture feature information, and depth feature information are fused using the element-wise multiplication fusion method, and the comprehensive feature information satisfies:

[0074]

[0075] where represents the morphological feature information, represents the texture feature information, represents the depth feature information, represents the comprehensive feature information.

[0076] S40: Perform image reconstruction on the comprehensive feature information to obtain enhanced reconstructed image information enhanced by texture feature information and morphological feature information;

[0077] In this step, the reconstruction process of performing image reconstruction on the comprehensive feature information is constrained by a reconstruction loss function. Specifically, the above reconstruction loss function:

[0078]

[0079] where represents the image function after reconstruction, and x i represents the input image information.

[0080] S50: Based on the Softmax cross-entropy loss function, perform classification training on the enhanced reconstructed image information to obtain the grading information corresponding to the pancreatic CT medical image.

[0081] In this step, the Softmax Cross-Entropy Loss Function is a loss function in multi-classification problems that combines the advantages of the Softmax function and the cross-entropy loss function. The Softmax function is used to convert the output of the model into a probability distribution, while the cross-entropy loss function is used to measure the difference between the predicted probability distribution and the true label.

[0082] The above-mentioned embodiment of the present invention uses a medical image classification method based on a deep multi-feature enhancement network to fuse the morphological features, texture features, etc. of the pancreatic CT image with the deep features to form a deep enhancement network, and then optimizes the network through semi-supervised learning. In the case of a small number of labeled samples, an effective classification network is trained to achieve intelligent CT grading of acute pancreatitis. The above-mentioned embodiment can automatically achieve intelligent grading of the patient's pancreatic CT images, effectively assist the clinic in targeted treatment of patients, and improve the clinical cure rate of patients. At the same time, the grading data is stored in the patient's treatment process data in a structured form, thereby better supporting subsequent medical research.

[0083] like Figure 2 As shown, the second preferred embodiment of the present invention discloses a pancreatitis CT image-based grading system 100 , the system 100 includes a CT image acquisition module 110 , a feature extraction module 120 , a feature fusion module 130 , an image reconstruction module 140 and a classification training module 150 .

[0084] The CT image acquisition module 110 is used to acquire a pancreatic CT medical image of a patient;

[0085] Specifically, the CT image acquisition module 110 acquires the pancreatic CT medical image corresponding to the target patient through the CT imaging device, and imports the pancreatic CT medical image into the image acquisition system corresponding to this step, so as to facilitate the analysis in the subsequent steps. In this embodiment, the CT medical image is recorded as x i .

[0086] The feature extraction module 120 is used to extract image features from the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image;

[0087] Specifically, the feature extraction module 120 mainly performs feature extraction on the acquired pancreatic CT medical image. The feature extraction process mainly includes morphological feature extraction, texture feature extraction and depth feature extraction, and obtains corresponding morphological feature information, texture feature information and depth feature information respectively.

[0088] In more detail, the feature extraction module 120 includes a morphological feature extraction unit 121 , a texture feature extraction unit 122 , and a depth feature extraction unit 123 .

[0089] In the above-mentioned morphological feature extraction unit 121, the shape, edge contour and other information of the pancreas are mainly extracted during the morphological feature extraction process. In this embodiment, the above-mentioned morphological feature extraction unit 121 uses a generalized linear model method to extract morphological features from the CT medical image. The characteristic of this method is that the natural measurement of the data is not forcibly changed, and the data can have a nonlinear and non-constant variance structure. It is a development of the linear model in studying the non-normal distribution of the response value and the concise and direct linear transformation of the nonlinear model.

[0090] After extracting the morphological feature information obtained by the generalized linear model method, in order to facilitate subsequent fusion, in this embodiment, the morphological feature extraction unit 121 also uses the principal component analysis method to reduce the dimension of the extracted morphological features to ensure that the morphological feature information is consistent with the latitude of the subsequent texture features and depth features. It can be expressed as:

[0091]

[0092] Among them, i above represents a natural number, x i represents the pancreatic CT medical image corresponding to the i-th copy, f f (x i ) represents the morphological feature extraction of the corresponding pancreatic CT medical image, Represents the final morphological feature information, Represents the morphological feature value information of the original input.

[0093] On the other hand, in general, the patient's pancreatic edema, swelling, peripancreatic fluid accumulation and other phenomena will show certain local serial repetition, non-random arrangement and basically uniform texture area in the CT medical image. In this embodiment, the texture feature extraction unit 122 uses a local binary pattern (LBP) to extract texture features from the CT medical image. Specifically, in this embodiment, a window with a threshold size of 3×3 is set to analyze and compare the values ​​of the central pixel and its 8 adjacent pixels. The texture feature value in the pancreatic CT can be obtained by the LBP function.

[0094] LBP (Local Binary Pattern) is an operator used to describe local texture features of an image; it has significant advantages such as rotation invariance and grayscale invariance. In this embodiment, the above method is used for texture feature extraction. Moreover, the extracted features are local texture features of the image.

[0095] The original LBP operator is defined as taking the central pixel of the window as the threshold within a 3*3 window, comparing the grayscale values ​​of the 8 adjacent pixels with it, and if the surrounding pixel values ​​are greater than the central pixel value, the position of the pixel is marked as 1, otherwise it is 0. In this way, the 8 points in the 3*3 neighborhood can generate an 8-bit binary number (usually converted to a decimal number, namely the LBP code, a total of 256 types) after comparison, that is, the LBP value of the central pixel of the window is obtained, and this value is used to reflect the texture information of the area.

[0096] Similarly, in order to facilitate the subsequent fusion, the texture feature extraction unit 122 also uses the principal component analysis (PCA) method to reduce the dimension of the extracted texture features to ensure that they are consistent with other morphological feature information and depth feature latitude. The final morphological feature h i t It can be expressed as:

[0097]

[0098] Among them, i above represents a natural number, x i represents the pancreatic CT medical image corresponding to the i-th copy, f f (x i ) represents the texture feature extraction of the corresponding pancreatic CT medical image, Represents the final texture feature information, Represents the texture feature value information of the original input.

[0099] The purpose of the PCA (Principal Component Analysis) algorithm is to convert high-dimensional data to low-dimensional data with less "information" loss. By extracting the largest individual differences revealed by the principal components, it can also be used to reduce the number of variables in regression analysis and cluster analysis, thereby reducing the amount of calculation.

[0100] Principal Component Analysis (PCA) is a technique used to explore the structure of high-dimensional data. The idea of PCA is to map the p-dimensional features onto k-dimensional space, where these k dimensions are new orthogonal features. These k-dimensional features are called principal components, which are re-constructed p-dimensional features. In PCA, the data is transformed from the original coordinate system to a new coordinate system, and the choice of the new coordinate system is closely related to the data itself. The first new coordinate axis is chosen in the direction of the largest variance in the original data, and the second new coordinate axis is chosen to be orthogonal to the first axis and have the largest variance. This process is repeated for the number of features in the original data. Most of the variance is contained in the first few new coordinate axes. Therefore, the remaining coordinate axes can be ignored, i.e., dimensionality reduction of the data is achieved.

[0101] The essence of the PCA algorithm is to find some projection directions such that the variance of the data in these projection directions is the largest, and these projection directions are orthogonal to each other. This is actually a process of finding new orthogonal bases. Calculate the variance of the projection of the original data on these orthogonal bases. The larger the variance, the more information is contained in the corresponding orthogonal basis. The larger the eigenvalue of the covariance matrix of the original data, the larger the corresponding variance, and the more information is projected on the corresponding eigenvector. Conversely, if the eigenvalue is small, it means that the information projected on these eigenvectors is very small, and the data in the directions corresponding to the small eigenvalues can be deleted, thus achieving the purpose of dimensionality reduction.

[0102] PCA combines potentially correlated high-dimensional variables into linearly independent low-dimensional variables, called principal components. The new low-dimensional dataset will retain as much of the original data's variables as possible.

[0103] B3. The deep feature extractor is composed of a stack of L convolutional layers. The output feature map of each layer can be expressed

[0104]

[0105] where, w k and b k are the weight and bias parameters of the k-th convolutional layer, σ is the activation function, and 1 < k < L. The final morphological feature h i d can be expressed as:

[0106]

[0107] The above-mentioned feature fusion module 130 is used to fuse the morphological feature information, texture feature information, and depth feature information to obtain comprehensive feature information.

[0108] Specifically, the feature fusion module 130 uses an element-by-element fusion method to fuse the morphological feature information, texture feature information, and depth feature information, and the comprehensive feature information satisfies:

[0109]

[0110] in, Represents morphological feature information, Represents texture feature information, Represents deep feature information, Represents comprehensive feature information.

[0111] The image reconstruction module 140 is used to reconstruct the comprehensive feature information to obtain enhanced reconstructed image information after the texture feature information and the morphology feature information are enhanced and reconstructed;

[0112] The reconstruction process of the image reconstruction module 140 for reconstructing the comprehensive feature information is constrained by a reconstruction loss function. Specifically, the reconstruction loss function is:

[0113]

[0114] in, Represents the image function after reconstruction, x i Represents the input image information.

[0115] The classification training module 150 is used to perform classification training on the enhanced reconstructed image information based on the Softmax cross entropy loss function to obtain the classification information corresponding to the pancreatic CT medical image.

[0116] In this step, the Softmax Cross-Entropy Loss Function is a loss function in multi-classification problems that combines the advantages of the Softmax function and the cross-entropy loss function. The Softmax function is used to convert the output of the model into a probability distribution, while the cross-entropy loss function is used to measure the difference between the predicted probability distribution and the true label.

[0117] The above-mentioned embodiment of the present invention uses a medical image classification system based on a deep multi-feature enhancement network to fuse the morphological features, texture features, etc. of the pancreatic CT image with the deep features to form a deep enhancement network, and then optimizes the network through semi-supervised learning. In the case of a small number of labeled samples, an effective classification network is trained to achieve intelligent CT grading of acute pancreatitis. The above-mentioned embodiment can automatically achieve intelligent grading of the patient's pancreatic CT images, effectively assist the clinic in targeted treatment of patients, and improve the clinical cure rate of patients. At the same time, the grading data is stored in the patient's treatment process data in a structured form, thereby better supporting subsequent medical research.

[0118] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, a medical image classification method based on a deep multi-feature enhancement network in a first preferred embodiment of the present invention is implemented.

[0119] The above-mentioned embodiment of the present invention executes the above-mentioned medical image classification system based on deep multi-feature enhancement network through a computer-readable storage medium, fuses the morphological features, texture features, etc. of the pancreatic CT image with the deep features, forms a deep enhancement network, and then optimizes the network through semi-supervised learning. In the case of a small number of labeled samples, an effective classification network is trained to realize intelligent CT grading of acute pancreatitis. Through the above-mentioned embodiment, the intelligent grading of the patient's pancreatic CT images can be automatically realized, which effectively assists the clinic in carrying out targeted treatment of the patient and improves the clinical cure rate of the patient. At the same time, the grading data is stored in the patient's treatment process data in a structured form, thereby better supporting subsequent medical research.

[0120] It should be noted that the computer-readable storage medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable storage medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0121] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0122] The computer-readable storage medium may be included in the electronic device, or may exist independently without being installed in the electronic device.

[0123] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device:

[0124] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0125] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A medical image classification method based on a deep multi-feature enhancement network, characterized in that: include: Obtain pancreatic CT medical images of patients; Performing image feature extraction on the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image; Fusing the morphological feature information, texture feature information, and depth feature information to obtain comprehensive feature information; Reconstructing the comprehensive feature information to obtain enhanced reconstructed image information after the texture feature information and the morphological feature information are enhanced and reconstructed; Based on the Softmax cross entropy loss function, the enhanced and reconstructed image information is classified and trained to obtain the grading information corresponding to the pancreatic CT medical image.

2. According to claim 1, a medical image classification method based on a deep multi-feature enhancement network is characterized in that: The extracting of image features from the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image includes: A generalized linear model method is used to extract morphological feature information from the CT medical image to obtain morphological feature information corresponding to the CT medical image.

3. The medical image classification method based on deep multi-feature enhancement network according to claim 1 is characterized in that: Performing image feature extraction on the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image includes: The local binary pattern is used to extract texture features of the CT medical image.

4. The medical image classification method based on deep multi-feature enhancement network according to claim 1 is characterized in that: Performing image feature extraction on the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image includes: The CT medical image is subjected to deep feature extraction by a deep feature extractor, wherein the deep feature extractor is composed of a convolutional layer containing L layers, and the output feature map of each layer can be expressed as: Among them, w k and b k are the weights and bias parameters of the kth convolutional layer, σ is the activation function, and 1 <k<L。 5. The medical image classification method based on deep multi-feature enhancement network according to claim 1, characterized in that: Before fusing the morphological feature information, the texture feature information and the depth feature information to obtain the comprehensive feature information, the method further includes: The morphological feature information and the texture feature information are processed by dimension reduction through principal component analysis so that the dimensions of the morphological feature information, the texture feature information and the depth feature information are consistent, and the morphological feature information after dimension reduction satisfies: Among them, i represents a natural number, f f (x i ) represents morphological feature extraction, represents the feature information after dimensionality reduction processing, Represents the original feature information.

6. The medical image classification method based on deep multi-feature enhancement network according to claim 1 is characterized in that: The fusion of the morphological feature information, the texture feature information and the depth feature information to obtain comprehensive feature information includes: The morphological feature information, texture feature information and depth feature information are fused by using an element-by-element fusion method, and the comprehensive feature information satisfies: in, Represents morphological feature information, Represents texture feature information, Represents deep feature information, Represents comprehensive feature information.

7. The medical image classification method based on deep multi-feature enhancement network according to claim 1 is characterized in that: The reconstruction process of reconstructing the image of the comprehensive feature information is constrained by a reconstruction loss function.

8. A pancreatitis CT imaging-based grading system, characterized in that: include: A CT image acquisition module, used for acquiring a pancreatic CT medical image of a patient; A feature extraction module, used to extract image features from the pancreatic CT medical image to obtain morphological feature information, texture feature information and depth feature information corresponding to the pancreatic CT medical image; A feature fusion module, used to fuse the morphological feature information, texture feature information and depth feature information to obtain comprehensive feature information; An image reconstruction module, used for reconstructing the comprehensive feature information to obtain enhanced reconstructed image information after the texture feature information and the morphological feature information are enhanced and reconstructed; The classification training module is used to perform classification training on the enhanced reconstructed image information based on the Softmax cross entropy loss function to obtain the classification information corresponding to the pancreatic CT medical image.

9. The pancreatitis CT image-based grading system according to claim 8, characterized in that: The feature extraction module comprises: The CT medical image is subjected to deep feature extraction by a deep feature extractor, wherein the deep feature extractor is composed of a convolutional layer containing L layers, and the output feature map of each layer can be expressed as: Among them, w k and b k are the weights and bias parameters of the kth convolutional layer, σ is the activation function, and 1 <k<L。 10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements a medical image classification method based on a deep multi-feature enhancement network as described in any one of claims 1 to 7.

Images