A method and system for judging radiation alveolar damage based on machine learning

Abstract

The embodiment of the present invention provides a method and system for judging radiation alveolar damage based on machine learning, which belongs to the field of artificial intelligence technology. The method includes: obtaining a target CT image corresponding to a target patient, and performing edge recognition on the target CT image to obtain a corresponding target edge image; performing data clustering on the corresponding initial edge in the target edge image to obtain a corresponding target clustering result; performing type recognition on each subclass cluster in the target clustering result to obtain a target type corresponding to the subclass cluster; performing type labeling on the target edge image according to the target type to obtain a corresponding target labeling image; performing density analysis on the target labeling image to obtain a type distribution result corresponding to the target labeling image; and classifying and judging the radiation alveolar damage corresponding to the target patient using the type distribution result according to the damage recognition model to obtain a target judgment result corresponding to the target patient.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for determining radiation alveolar damage based on machine learning. Background Art

[0002] Radiotherapy is one of the conventional treatments for malignant thoracic tumors. Radiotherapy can cause irreversible damage to nearby lung tissues due to radiation doses exceeding their biological effect threshold. Therefore, radiation-induced alveolar damage has become a complication of radiotherapy for patients with thoracic tumors, affecting the quality of life of patients and even leading to death. After radiotherapy, ground-glass changes appear in the clinical target volume range, and the characteristics of interphase transition are reflected. During the interphase, the alveoli are damaged by radiation, and exudative organization can be observed, showing fibrotic characteristics. However, the relevant technology cannot promptly determine whether there is radiation-induced alveolar damage, and thus cannot promptly switch to precise radiotherapy plans, which may cause unnecessary harm to patients.

[0003] Therefore, it is urgent to design a technical solution to solve at least one of the above technical problems. Summary of the invention

[0004] The main purpose of the embodiments of the present invention is to provide a method and system for determining radiation alveolar damage based on machine learning, aiming to solve the problem in the related technology that it is impossible to timely determine whether radiation alveolar damage exists, and thus it is impossible to timely switch to precise radiotherapy plans, which may cause unnecessary harm to patients.

[0005] In a first aspect, an embodiment of the present invention provides a method for determining radiation alveolar damage based on machine learning, comprising:

[0006] Obtaining a target CT image corresponding to a target patient, and performing edge recognition on the target CT image to obtain a corresponding target edge image;

[0007] Performing data clustering on the corresponding initial edges in the target edge image to obtain corresponding target clustering results;

[0008] Performing type identification on each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster;

[0009] Performing type labeling on the target edge image according to the target type to obtain a corresponding target labeling image;

[0010] Performing density analysis on the target annotated image to obtain a type distribution result corresponding to the target annotated image;

[0011] The radiation alveolar injury corresponding to the target patient is classified and judged using the type distribution result according to the injury recognition model to obtain a target judgment result corresponding to the target patient.

[0012] In a second aspect, an embodiment of the present invention provides a radiation alveolar damage judgment system based on machine learning, comprising:

[0013] A data processing module, used to obtain a target CT image corresponding to a target patient, and perform edge recognition on the target CT image to obtain a corresponding target edge image;

[0014] A data clustering module, used for performing data clustering on the corresponding initial edges in the target edge image to obtain the corresponding target clustering results;

[0015] A type identification module, used to perform type identification on each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster;

[0016] A type labeling module, used for labeling the target edge image according to the target type to obtain a corresponding target labeling image;

[0017] A density analysis module, used to perform density analysis on the target annotated image to obtain a type distribution result corresponding to the target annotated image;

[0018] The injury judgment module is used to classify and judge the radiation alveolar injury corresponding to the target patient using the type distribution result according to the injury recognition model, and obtain the target judgment result corresponding to the target patient.

[0019] In a third aspect, an embodiment of the present invention further provides a terminal device, comprising a processor, a memory, a computer program stored in the memory and executable by the processor, and a data bus for realizing connection and communication between the processor and the memory, wherein when the computer program is executed by the processor, the steps of any one of the methods for determining radiation alveolar damage based on machine learning provided in the specification of the present invention are implemented.

[0020] In a fourth aspect, an embodiment of the present invention further provides a storage medium for computer-readable storage, characterized in that the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any method for determining radiation alveolar damage based on machine learning as provided in the specification of the present invention.

[0021] The embodiment of the present invention provides a method and system for judging radiation alveolar damage based on machine learning, the method comprising: obtaining a target CT image related to a target patient, then performing edge recognition on the target CT image, extracting key structures and contours in the target CT image to obtain a target edge image. Then, edge shape clustering is performed on the initial edge in the target edge image to obtain a target clustering result, so that through edge recognition and data clustering, the structural features in the target CT image can be more clearly extracted, providing support for subsequent damage judgment. After obtaining the target clustering result, type recognition is performed on each subclass cluster in the target clustering result to determine the target type corresponding to each subclass cluster. Subsequently, according to the identified target type, the target edge image is type-labeled to obtain a target labeled image, thereby not only making the distribution of different shapes in the target CT image clear at a glance, but also providing a basis for subsequent quantitative and qualitative analysis, and then performing density analysis on the target labeled image to obtain a type distribution result. The type distribution result can quantify the distribution of different target types in the target CT image, providing a basis for judging whether there is damage, and then classifying and judging the possible radiation alveolar damage of the target patient based on the damage recognition model using the type distribution result to obtain the target judgment result corresponding to the target patient. This method achieves efficient and accurate analysis from target CT image acquisition to injury judgment. It not only improves the accuracy and efficiency of diagnosis, but also provides a scientific basis for clinical decision-making, which helps to improve the treatment effect and prognosis of patients. It also solves the problem that related technologies cannot timely determine whether there is radiation alveolar damage, and thus cannot promptly switch to precise radiotherapy plans, which may cause unnecessary harm to patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 A schematic diagram of a flow chart of a method for determining radiation alveolar damage based on machine learning provided in an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of the module structure of a radiation alveolar damage judgment system based on machine learning provided in an embodiment of the present invention;

[0025] Figure 3 A schematic block diagram of the structure of a terminal device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0028] It should be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0029] The embodiment of the present invention provides a method and system for judging radiation alveolar damage based on machine learning. The method for judging radiation alveolar damage based on machine learning can be applied to a terminal device, which can be an electronic device such as a tablet computer, a laptop computer, a desktop computer, a personal digital assistant, and a wearable device. The terminal device can be a server or a server cluster.

[0030] Radiotherapy is one of the conventional treatments for malignant tumors of the chest, and 70% of lung cancers require radiotherapy. Radiotherapy can cause irreversible damage to nearby lung tissues due to radiation doses exceeding their biological effect threshold. Therefore, radiation alveolar damage has become a complication of radiotherapy for patients with chest tumors, affecting the quality of life of patients and even leading to death. After radiotherapy, ground-glass changes appear in the clinical target volume range, and the characteristics of interphase transition are reflected. In the interphase, the alveoli are damaged by radiation, and exudative organization can be observed, showing fibrotic characteristics. However, the relevant technology cannot promptly determine whether there is radiation alveolar damage, and thus cannot promptly switch to precise radiotherapy plans, which may cause unnecessary harm to patients.

[0031] Therefore, it is urgent to design a technical solution to solve at least one of the above technical problems.

[0032] To this end, an embodiment of the present invention provides a method and system for judging radiation alveolar damage based on machine learning, the method comprising: obtaining a target CT image related to a target patient, then performing edge recognition on the target CT image, extracting key structures and contours in the target CT image to obtain a target edge image. Then, edge shape clustering is performed on the initial edge in the target edge image to obtain a target clustering result, so that through edge recognition and data clustering, the structural features in the target CT image can be more clearly extracted, providing support for subsequent damage judgment. After obtaining the target clustering result, type recognition is performed on each subclass cluster in the target clustering result to determine the target type corresponding to each subclass cluster. Subsequently, according to the identified target type, the target edge image is type-labeled to obtain a target labeled image, thereby not only making the distribution of different shapes in the target CT image clear at a glance, but also providing a basis for subsequent quantitative and qualitative analysis, and then performing density analysis on the target labeled image to obtain a type distribution result. The type distribution result can quantify the distribution of different target types in the target CT image, providing a basis for judging whether there is damage, and then classifying and judging the possible radiation alveolar damage of the target patient based on the damage recognition model using the type distribution result to obtain the target judgment result corresponding to the target patient. This method achieves efficient and accurate analysis from target CT image acquisition to injury judgment. It not only improves the accuracy and efficiency of diagnosis, but also provides a scientific basis for clinical decision-making, which helps to improve the treatment effect and prognosis of patients. It also solves the problem that related technologies cannot timely determine whether there is radiation alveolar damage, and thus cannot promptly switch to precise radiotherapy plans, which may cause unnecessary harm to patients.

[0033] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0034] Please refer to Figure 1 , Figure 1 A flowchart of a method for determining radiation alveolar damage based on machine learning is provided in an embodiment of the present invention.

[0035] like Figure 1 As shown, the method for determining radiation alveolar damage based on machine learning includes steps S101 to S106.

[0036] Step S101: obtaining a target CT image corresponding to a target patient, and performing edge recognition on the target CT image to obtain a corresponding target edge image.

[0037] Exemplarily, the target CT image of the target patient is retrieved from the electronic medical record system based on the target patient's identification (such as name, ID number) or medical record.

[0038] Exemplarily, edge detection is performed on the target CT image using an edge detection algorithm to obtain a target edge image corresponding to the target CT image. The edge detection algorithm includes but is not limited to Canny edge detection, Sobel operator or Laplace operator, etc.

[0039] Step S102: performing data clustering on the corresponding initial edges in the target edge image to obtain corresponding target clustering results.

[0040] Exemplarily, when radiation-induced alveolar damage occurs, ground-glass changes, interphase transformations, and fibrosis may appear in the target CT image, affecting the performance. Ground-glass changes are manifested as reduced transparency of lung tissue in the target CT image, with the appearance of fuzzy, translucent shadows, which usually do not obscure the edges of the pulmonary vessels and bronchi, and in specific areas of the lobes, with variable shapes. Interphase transformations may manifest as interstitial thickening or fibrosis, leading to structural and functional changes in lung tissue. Diffuse lines or nodules may be seen on the target CT image, which may be accompanied by ground-glass changes, or may present a grid-like or band-like appearance. Fibrosis is manifested as significant thickening of the interstitium and remodeling of tissue. In the target CT image, fibrosis usually manifests as band-like or reticular changes, which may be accompanied by the appearance of honeycomb lungs (small vesicular changes), and is usually accompanied by features associated with ground-glass changes.

[0041] Exemplarily, based on the above information, the present application clusters the initial edges in the target edge image to analyze the distribution characteristics of edges with the same or similar contours, thereby providing support for the subsequent judgment of whether there is radiation alveolar damage.

[0042] Exemplarily, initial coordinate information corresponding to the initial edge in the target CT image is obtained, and then the initial coordinate information is standardized so that features such as scales and directions of different initial edges are comparable, thereby obtaining target coordinate information.

[0043] Exemplarily, useful features such as the length, curvature, and angle change of the initial edge are extracted from the target coordinate information, and the extracted features are encoded into a feature vector, which is convenient for computer processing and the application of clustering algorithms. Then, a clustering algorithm such as K-means clustering, hierarchical clustering, DBSCAN, etc. is selected to perform cluster analysis on the feature vectors corresponding to the initial edge, thereby obtaining the target clustering result corresponding to the initial edge.

[0044] In some embodiments, data clustering is performed on the corresponding initial edges in the target edge image to obtain the corresponding target clustering results, including: obtaining a target coordinate set corresponding to the initial edge in a target coordinate system; determining an initial characterization vector corresponding to the initial edge in the target coordinate system based on the target coordinate set; performing dimensionality reduction processing on the initial characterization vector to obtain a target characterization vector corresponding to the initial edge; and data clustering is performed on the initial edge based on the target characterization vector to obtain the target clustering results.

[0045] Exemplarily, the initial coordinate set corresponding to the initial edge in the target edge image is obtained, and then the algorithm for obtaining the geometric center is used to obtain the edge center corresponding to the initial edge in consideration of the characteristics of the initial edge, and then the edge center is determined as the origin of the target coordinate system, so that the origin of the coordinate system corresponding to the target edge image is converted to the origin of the target coordinate system, and then the geometric transformation parameters (such as translation, rotation, scaling, etc.) required for the target edge image corresponding coordinate system to be transferred to the target coordinate system are obtained, and then each coordinate in the initial coordinate set is transformed according to the geometric transformation parameters to obtain the target coordinate set corresponding to the initial edge in the target coordinate system. The target coordinate system is a coordinate system with the edge center of the initial edge as the origin.

[0046] Exemplarily, the initial coordinate set of each initial edge in the target edge image is transferred to the target coordinate system with the edge center of the initial edge as the origin, that is, each initial edge in the target edge image is under the same reference object, thereby providing support for the subsequent improvement of data clustering results.

[0047] Exemplarily, each coordinate information in the target coordinate set is connected to the corresponding origin of the target coordinate system, so as to calculate the edge vector between each coordinate information in the target coordinate set and the corresponding origin of the target coordinate system, and then jointly determine multiple edge vectors as the initial characterization vector.

[0048] For example, the target representation vector is obtained by performing dimensionality reduction processing on the initial representation vector using principal component analysis, and the purpose of dimensionality reduction is to reduce the dimension of the initial representation vector while retaining key information, so that the target representation vector can better reflect the essential characteristics of the initial edge while reducing the computational complexity.

[0049] Exemplarily, the target representation vector is obtained by processing the initial representation vector through a dimensionality reduction algorithm. The dimensionality reduction process not only reduces the dimension of the feature space, but also extracts key information, so that the target representation vector can restore the contour information of the initial edge.

[0050] For example, before using the target representation vector for cluster analysis, an important preprocessing step is to transfer each subvector in the target representation vector to the same direction. This operation is intended to ignore the influence of the protruding or concave parts of the initial edge in different directions. By unifying the direction, the error caused by the change of the local features of the initial edge can be reduced, and the accuracy and consistency of the clustering results can be improved.

[0051] Exemplarily, according to the target representation vector after dimensionality reduction, an appropriate clustering algorithm can be selected for clustering analysis. Common clustering algorithms include K-means, hierarchical clustering, DBSCAN, etc. According to the intrinsic structure of the target representation vector, similar initial edges are classified into the same cluster to obtain the target clustering result. Each clustering cluster in the target clustering result should be able to clearly reflect the outline of the same feature. This means that the edge data in the same clustering cluster should have similar morphological and structural characteristics.

[0052] In some embodiments, obtaining the target coordinate set corresponding to the initial edge in the target coordinate system includes: determining the initial coordinate system, and obtaining the first coordinate set corresponding to the initial edge in the initial coordinate system; calculating the average value of all horizontal coordinates in the first coordinate set to determine the first horizontal center corresponding to the initial edge, and calculating the average value of all vertical coordinates in the first coordinate set to determine the first vertical center corresponding to the initial edge; collecting edge coordinates of the initial edge using a target angle according to the first horizontal center and the first vertical center, and obtaining a second coordinate set corresponding to the initial edge; calculating the area of ​​the initial edge according to the second coordinate set, the first horizontal center, and the first vertical center, and obtaining a target area corresponding to the initial edge; and obtaining a target area corresponding to the initial edge according to the target area and the first vertical center. The second coordinate set recalculates the target center corresponding to the initial edge to obtain the second horizontal center and the second vertical center; translates the origin corresponding to the initial coordinate system to the second horizontal center and the second vertical center to obtain the target coordinate system; processes the second coordinate set according to the initial coordinate system and the target coordinate system to obtain the third coordinate set; performs distance calculation on any two coordinates in the third coordinate set to obtain the target major axis corresponding to the initial edge, and determines the maximum edge point corresponding to the initial edge according to the target major axis; determines the target rotation information corresponding to the initial edge according to the target major axis and the maximum edge point; adjusts the third coordinate set in the target coordinate system according to the target rotation information to obtain the target coordinate set corresponding to the initial edge in the target coordinate system.

[0053] Exemplarily, the image coordinate system corresponding to the target edge image is determined as the initial coordinate system, and then the first coordinate set corresponding to the initial edge in the initial coordinate system is obtained.

[0054] Exemplarily, all horizontal coordinates corresponding to the initial edge are obtained from the first coordinate set, and all horizontal coordinates are summed and averaged, and then the corresponding horizontal average value is determined as the first horizontal center corresponding to the initial edge; and all vertical coordinates corresponding to the initial edge are obtained from the first coordinate set, and all vertical coordinates are summed and averaged, and then the corresponding vertical average value is determined as the first vertical center corresponding to the initial edge.

[0055] Exemplarily, the target angle is the collection interval when collecting the initial edge. The smaller the target angle, the more the number of collections, the more coordinates in the second coordinate set corresponding to the initial edge, and the more accurate the representation of the initial edge. The target angle can be set to 1 degree or 2 degrees, etc., and this application does not limit it.

[0056] Exemplarily, the first horizontal center and the first vertical center are used as acquisition centers, and then a acquisition line is emitted from the acquisition center at every target angle, and then the intersection coordinates corresponding to the intersection of the acquisition line and the initial edge are obtained, so that all the intersection coordinates are combined into a second coordinate set corresponding to the initial edge.

[0057] Exemplarily, two adjacent coordinates are obtained from the second coordinate set and determined as the first coordinate and the second coordinate, and then the first coordinate and the second coordinate are combined with the first horizontal center and the first vertical center to form a triangle, and then the area of ​​the triangle is solved, so that the target area corresponding to the initial edge is obtained by summing the areas of all triangles formed by the two adjacent coordinates in the second coordinate set and the first horizontal center and the first vertical center.

[0058] Exemplarily, the first result is obtained by multiplying the area of ​​each triangle by the horizontal coordinate of the first coordinate corresponding to the triangle, and then the first result corresponding to each triangle is obtained and summed to obtain the first summation result, and then the first summation result is divided by the target area to obtain the second horizontal center of the target center corresponding to the initial edge. Similarly, the second result is obtained by multiplying the area of ​​each triangle by the vertical coordinate of the first coordinate corresponding to the triangle, and then the second result corresponding to each triangle is obtained and summed to obtain the second summation result, and then the second summation result is divided by the target area to obtain the second vertical center of the target center corresponding to the initial edge.

[0059] Exemplarily, the origin corresponding to the initial coordinate system is translated to the second horizontal center and the second vertical center to obtain the target coordinate system, that is, the second horizontal center and the second vertical center are determined as the coordinate origin corresponding to the target coordinate system. In this way, the geometric transformation parameters (such as translation, rotation, scaling, etc.) required for transferring the initial coordinate system to the target coordinate system are obtained, and then the second coordinate set is processed according to the geometric transformation parameters (such as translation, rotation, scaling, etc.) to obtain the third coordinate set.

[0060] Exemplarily, the distance between any two coordinates in the third coordinate set is calculated to obtain the third coordinate and the fourth coordinate corresponding to the maximum distance, and then the coordinate vector between the third coordinate and the fourth coordinate is calculated, and then the coordinate vector is determined as the target long axis corresponding to the initial edge. The target long axis is used to characterize the direction information corresponding to the maximum edge distance in the initial edge.

[0061] Exemplarily, the third coordinate and the fourth coordinate corresponding to the long axis of the target are obtained, and then the third coordinate and the distance between the second horizontal center and the second vertical center are calculated to obtain the first distance, and the distance between the fourth coordinate and the second horizontal center and the second vertical center are calculated to obtain the second distance, and then the coordinate corresponding to the maximum of the first distance and the second distance is determined as the maximum edge point corresponding to the initial edge.

[0062] Exemplarily, the target major axis is determined as the horizontal reference line of the initial edge, and the direction of the second horizontal center and the second vertical center pointing to the maximum edge point is determined as the arrow direction of the horizontal reference line. That is, an x-axis is reset according to the target major axis and the maximum edge point, the x-axis is parallel to the horizontal reference line, and the direction of the x-axis is the same as the direction of the second horizontal center and the second vertical center pointing to the maximum edge point, thereby obtaining the target rotation information required for converting the x-axis to the x-axis in the target coordinate system, and then adjusting each coordinate in the third coordinate set according to the target rotation information, thereby obtaining the target coordinate set corresponding to the initial edge in the target coordinate system.

[0063] Step S103: performing type identification on each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster.

[0064] Exemplarily, feature extraction is performed on the initial edges contained in each subclass cluster in the target clustering result to obtain edge features corresponding to each initial edge in the subclass cluster, and then feature fusion is performed on all edge features of each subclass cluster to obtain target features corresponding to the subclass cluster, and then the target classification model is used to perform type identification on the target features to obtain the target type corresponding to the subclass cluster. The target types include but are not limited to strips, strips, circles, etc.

[0065] In some embodiments, the type identification of each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster includes: extracting features of each target edge in the subclass cluster to obtain initial color features, initial texture features, and initial morphological features; performing feature fusion based on the initial color features, the initial texture features, and the initial morphological features to obtain the target features corresponding to the target edge; performing type identification on the target edge using the target features according to a type identification model to obtain an initial type; and summarizing the initial types using a voting mechanism to determine the target type corresponding to the subclass cluster.

[0066] Exemplarily, the type of color feature to be extracted is determined, such as average color, color histogram, principal component color, etc. The region corresponding to each target edge in the subclass cluster is separated from the target CT image, so as to perform color analysis on the extracted and separated region, calculate the intensity and distribution of each color, and then generate the initial color feature of each target edge.

[0067] Exemplarily, a texture feature extraction algorithm such as a local binary pattern (LBP), a gray level co-occurrence matrix (GLCM), etc. is determined, and the area corresponding to the target edge in the target CT image is extracted according to the texture feature extraction algorithm to extract texture features, including texture roughness, direction, contrast, etc., thereby obtaining an initial texture feature containing multiple texture features.

[0068] Exemplarily, the type of morphological features is determined, such as area, perimeter, shape moment, edge morphology, etc. Morphological processing is performed on the target edge, including contour detection, normalization, etc., to extract key morphological features, and the geometric parameters of the target edge are calculated by numerical methods to obtain initial morphological features describing the morphology.

[0069] Exemplarily, the extracted initial color features, initial texture features, and initial morphological features are weightedly fused to form target features, and the target features serve as the final feature description of the target edge.

[0070] Exemplarily, a machine learning or deep learning model is selected for training (such as random forest, support vector machine, neural network, etc.). A training set containing labeled data is used to train the model so that it can identify the target type through input features. Thus, the fused target features are input into the trained type recognition model, the output of the model is obtained, and the initial type corresponding to the target edge is obtained.

[0071] Exemplarily, for the initial types of all target edges in the same subclass cluster, a voting mechanism is used for inductive processing to obtain the frequency of occurrence of each initial type in the target edge, and then the initial type corresponding to the maximum frequency is used as the target type of all target edges in the subclass cluster.

[0072] In some embodiments, the obtaining the target feature corresponding to the target edge by performing feature fusion according to the initial color feature, the initial texture feature and the initial morphological feature includes: performing information analysis on all the initial color features to obtain a first entropy value corresponding to the subclass cluster in the color dimension; performing information analysis on all the initial texture features to obtain a second entropy value corresponding to the subclass cluster in the texture dimension; performing information analysis on all the initial morphological features to obtain a third entropy value corresponding to the subclass cluster in the morphological dimension; determining a first weight corresponding to the subclass cluster in the color dimension, a second weight corresponding to the subclass cluster in the texture dimension and a third weight corresponding to the subclass cluster in the morphological dimension according to the first entropy value, the second entropy value and the third entropy value; performing feature fusion on the initial color feature, the initial texture feature and the initial morphological feature according to the first weight, the second weight and the third weight to obtain the target feature corresponding to the target edge; wherein the first weight, the second weight and the third weight are obtained according to the following formula:

[0073]

[0074] Among them, w1 represents the first weight, w2 represents the second weight, w3 represents the third weight, S1 represents the first entropy value, S2 represents the second entropy value, S3 represents the third entropy value, and m represents the number corresponding to the target edge in the subclass cluster.

[0075] Exemplarily, the information amount of the color feature set is quantified using information entropy or mutual information, etc. The information entropy of the initial color features of all target edges in the subclass cluster is calculated to obtain a first entropy value in the color dimension.

[0076] Exemplarily, an algorithm such as information entropy or mutual information is used to quantify the amount of information in the texture feature set. The information entropy of the initial texture features of all target edges in the subclass cluster is calculated to obtain a second entropy value in the texture dimension.

[0077] Exemplarily, an algorithm such as information entropy or mutual information is used to quantify the amount of information in the morphological feature set. The information entropy of the initial morphological features of all target edges in the subclass cluster is calculated to obtain a third entropy value in the morphological dimension.

[0078] Exemplarily, a first weight corresponding to the subclass cluster in the color dimension, a second weight corresponding to the subclass cluster in the texture dimension, and a third weight corresponding to the subclass cluster in the morphology dimension are determined according to the first entropy value, the second entropy value, and the third entropy value; wherein the first weight, the second weight, and the third weight are obtained according to the following formula:

[0079]

[0080] Among them, w1 represents the first weight, w2 represents the second weight, w3 represents the third weight, S1 represents the first entropy value, S2 represents the second entropy value, S3 represents the third entropy value, and m represents the number of target edges corresponding to the subclass cluster.

[0081] Exemplarily, the above formula balances the impact of different feature dimensions on the final result. For example, if the change in the color dimension is large (high entropy value), it has a stronger ability to distinguish the edge of the target and should be given a higher weight; on the contrary, if the morphological dimension is relatively simple (low entropy value), its weight can be reduced. In the above formula, the distribution of weights is further adjusted by introducing the number of target edges in the subclass cluster. When the number is large, the impact of the entropy value will be more significant, making the model more adaptable to large-scale subclass clusters. Therefore, by automatically calculating the weights based on the entropy value, the process of manually adjusting the weights can be reduced. In traditional methods, weights may need to be determined by experiments or experience, while this method automatically calculates through mathematical formulas, simplifies the tuning process, and improves efficiency.

[0082] Exemplarily, the initial color features, the initial texture features, and the initial morphological features are weightedly fused according to the first weight, the second weight, and the third weight to obtain the target features corresponding to the target edge.

[0083] Step S104: annotate the target edge image according to the target type to obtain a corresponding target annotated image.

[0084] Exemplarily, a target label corresponding to the target type is determined, and then the target region corresponding to the target edge in the target edge image is labeled with the target label, thereby obtaining a corresponding target labeled image.

[0085] In some embodiments, the type labeling of the target edge image according to the target type to obtain a corresponding target labeled image includes: obtaining a target label corresponding to the target type; setting an initial labeled image that is the same size as the target edge image and has a pixel value that is a preset value; obtaining edge information corresponding to the initial edge in the target edge image and internal information contained in the edge information; setting a first pixel value corresponding to the edge information in the initial labeled image to the target label, and setting a second pixel value corresponding to the internal information to the target label, to obtain the target labeled image.

[0086] Exemplarily, a unique target label is assigned to each target type. The target label can be a number or a color value, which is used to identify the type in subsequent annotated images.

[0087] Exemplarily, a blank image of the same size as the target edge image is created as the initial labeled image. Each pixel value of the image is initially set to a preset value (eg, 0) to indicate an unlabeled area.

[0088] Exemplarily, edge information corresponding to the target edge is extracted from the target edge image. The edge information refers to the outline or boundary portion of the target edge, indicating the outer boundary of the target edge. The internal information refers to the internal area surrounded by the target edge. The pixel area inside the target edge can be determined by image processing technology (e.g., edge-based filling algorithm or connected region analysis). The internal information refers to the internal part of the target edge, that is, all pixel information surrounded by the target edge.

[0089] Exemplarily, in the initial annotated image, the pixel value corresponding to the edge information is updated to the target label (i.e., the label value previously assigned to each target type). Each pixel in the edge information is traversed, and its corresponding position in the initial annotated image is set to the label value of the target type. In the initial annotated image, the pixel value corresponding to the internal information is also updated to the same target label. Each pixel in the internal information is traversed, and its corresponding position in the initial annotated image is set to the label value of the target type.

[0090] Exemplarily, after the above-mentioned labeling steps, each pixel in the initial labeling image has been updated to the target label value, and the edge area and the internal area of ​​the target edge are labeled, so that the initial labeling image has become the target labeling image.

[0091] Step S105: performing density analysis on the target annotated image to obtain a type distribution result corresponding to the target annotated image.

[0092] Exemplarily, the number of edges corresponding to each target type in a unit area is obtained according to the target label image, thereby determining the distribution density corresponding to the target type in the unit area according to the number of edges, and then obtaining the type distribution result corresponding to the target type according to the distribution density.

[0093] In some embodiments, the density analysis of the target annotated image to obtain the type distribution result corresponding to the target annotated image includes: determining a target window and a target step size, and determining a sliding direction corresponding to the sliding of the target window in the target annotated image; sliding the target window in the target annotated image according to the target step size and the sliding direction to obtain a window annotated image corresponding to the target window; performing a density analysis on the target label in the window annotated image to obtain window density information corresponding to the target label under the target window; arranging the window density information in sequence according to the sliding direction to obtain sequence density information corresponding to the target label; performing feature extraction on the sequence density information to obtain a label distribution result corresponding to the target label; and determining the type distribution result corresponding to the target annotated image according to the target label and the label distribution result.

[0094] Exemplarily, a target window is selected to slide in the target annotated image. The size of the target window should be able to effectively capture the target label information. For example, 20x20 pixels, 50x50 pixels, etc. can be selected and adjusted according to the size and needs of the target. A target step size is set for the sliding of the target window in the target annotated image. The step size should be less than or equal to the size of the window to ensure comprehensive coverage. For example, the target step size can be set to half the size of the window (if the window is 50x50, the step size is 25px) to make the target window move more smoothly. Determine the direction in which the target window slides in the target annotated image. There are usually two basic directions, horizontal sliding is from left to right, and continues to slide downward for a new row. Vertical sliding moves the window from top to bottom, and you can choose to slide horizontally and vertically at the same time to obtain window annotations for the entire image.

[0095] Exemplarily, according to the set target step size and sliding direction, the target window is gradually moved in the target annotated image, and the target window is first placed in the upper left corner of the target annotated image. After each movement, the annotated image in the target window is obtained until the window reaches the lower right corner of the target annotated image.

[0096] Exemplarily, after each slide, the annotated images in the target window are extracted according to the position of the current target window. These images should contain target label information (for example, 1, 2, 3, etc. represent different target types). Then, the target labels in the window annotated images are counted, and the number of occurrences of each target label in the window annotated images is calculated to generate window density information. For example, the number of pixels of each target label can be divided by the total number of pixels of the target window to obtain the density value of the target label in the window.

[0097] Exemplarily, the density information corresponding to each target window is arranged in sequence according to the sliding direction to form a continuous sequence of density information. For example, if horizontal sliding is used, the density data of each window in the first row is recorded first, then the second row, and so on.

[0098] Exemplarily, feature extraction is performed on the arranged sequence density information, including but not limited to calculating the mean value, variance, maximum value, and minimum value corresponding to the sequence density information, so as to obtain a label distribution result corresponding to the target label.

[0099] Exemplarily, all target labels and label distribution results corresponding to the target labels are determined as type distribution results corresponding to the target annotated image.

[0100] Step S106: classify and judge the radiation alveolar injury corresponding to the target patient using the type distribution result according to the injury recognition model to obtain a target judgment result corresponding to the target patient.

[0101] Exemplarily, the type distribution result is input into the injury recognition model, and the label distribution result corresponding to each target label in the type distribution result is spliced ​​in the feature splicing layer of the injury recognition model. The feature splicing layer is responsible for integrating the type distribution result into a unified target feature vector, and then inputting the generated target feature vector into the convolution layer, pooling layer, and fully connected layer of the injury recognition model. Finally, the final feature vector is converted into a target classification result through a classifier, and then the classification type corresponding to the maximum probability in the target classification result is obtained, so as to obtain the target judgment result corresponding to the target patient according to the classification type.

[0102] For example, the damage recognition model can determine whether there are grid-shaped areas, band-shaped areas, or areas with variable shapes, as well as normal areas in the target CT image, and then use the type distribution results through the damage recognition model to classify and judge the radiation alveolar damage corresponding to the target patient. When the classification type is any type of grid-shaped areas, band-shaped areas, or areas with variable shapes, the target judgment result corresponding to the target patient is determined to be the presence of radiation alveolar damage, and the treatment plan needs to be modified in time; when the classification type is the normal area type, the target judgment result corresponding to the target patient is determined to be the absence of radiation alveolar damage, and treatment can continue.

[0103] The embodiment of the present invention provides a method and system for judging radiation alveolar damage based on machine learning, the method comprising: obtaining a target CT image related to a target patient, then performing edge recognition on the target CT image, extracting key structures and contours in the target CT image to obtain a target edge image. Then, edge shape clustering is performed on the initial edge in the target edge image to obtain a target clustering result, so that through edge recognition and data clustering, the structural features in the target CT image can be more clearly extracted, providing support for subsequent damage judgment. After obtaining the target clustering result, type recognition is performed on each subclass cluster in the target clustering result to determine the target type corresponding to each subclass cluster. Subsequently, according to the identified target type, the target edge image is type-labeled to obtain a target labeled image, thereby not only making the distribution of different shapes in the target CT image clear at a glance, but also providing a basis for subsequent quantitative and qualitative analysis, and then performing density analysis on the target labeled image to obtain a type distribution result. The type distribution result can quantify the distribution of different target types in the target CT image, providing a basis for judging whether there is damage, and then classifying and judging the possible radiation alveolar damage of the target patient based on the damage recognition model using the type distribution result to obtain the target judgment result corresponding to the target patient. This method achieves efficient and accurate analysis from target CT image acquisition to injury judgment. It not only improves the accuracy and efficiency of diagnosis, but also provides a scientific basis for clinical decision-making, which helps to improve the treatment effect and prognosis of patients. It also solves the problem that related technologies cannot timely determine whether there is radiation alveolar damage, and thus cannot promptly switch to precise radiotherapy plans, which may cause unnecessary harm to patients.

[0104] See also Figure 2 , Figure 2A radiation alveolar injury judgment system 200 based on machine learning is provided in an embodiment of the present application. The radiation alveolar injury judgment system 200 based on machine learning includes a data processing module 201, a data clustering module 202, a type recognition module 203, a type labeling module 204, a density analysis module 205, and a damage judgment module 206. The data processing module 201 is used to obtain a target CT image corresponding to a target patient, and perform edge recognition on the target CT image to obtain a corresponding target edge image; the data clustering module 202 is used to perform data clustering on the corresponding initial edges in the target edge image to obtain a corresponding target The target clustering result is obtained by a type identification module 203, which is used to perform type identification on each sub-class cluster in the target clustering result to obtain the target type corresponding to the sub-class cluster; the type labeling module 204 is used to perform type labeling on the target edge image according to the target type to obtain the corresponding target labeling image; the density analysis module 205 is used to perform density analysis on the target labeling image to obtain the type distribution result corresponding to the target labeling image; the injury judgment module 206 is used to classify and judge the radiation alveolar injury corresponding to the target patient according to the injury identification model using the type distribution result to obtain the target judgment result corresponding to the target patient.

[0105] In some implementations, the data clustering module 202 performs, during the process of performing data clustering on the corresponding initial edges in the target edge image to obtain the corresponding target clustering result:

[0106] Obtaining a target coordinate set corresponding to the initial edge in a target coordinate system;

[0107] Determine, according to the target coordinate set, an initial characterization vector corresponding to the initial edge in the target coordinate system;

[0108] Performing dimensionality reduction processing on the initial representation vector to obtain a target representation vector corresponding to the initial edge;

[0109] Data clustering is performed on the initial edge according to the target characterization vector to obtain the target clustering result.

[0110] In some implementations, the data clustering module 202 performs, during the process of obtaining the target coordinate set corresponding to the initial edge in the target coordinate system:

[0111] Determine an initial coordinate system, and obtain a first coordinate set corresponding to the initial edge in the initial coordinate system;

[0112] Calculating the average value of all horizontal coordinates in the first coordinate set to determine a first horizontal center corresponding to the initial edge and calculating the average value of all vertical coordinates in the first coordinate set to determine a first vertical center corresponding to the initial edge;

[0113] Collect edge coordinates of the initial edge using a target angle according to the first horizontal center and the first vertical center to obtain a second coordinate set corresponding to the initial edge;

[0114] Calculating the area of ​​the initial edge according to the second coordinate set, the first horizontal center, and the first vertical center to obtain a target area corresponding to the initial edge;

[0115] Recalculate the target center corresponding to the initial edge according to the target area and the second coordinate set to obtain a second horizontal center and a second vertical center;

[0116] The origin corresponding to the initial coordinate system is translated to the second horizontal center and the second vertical center to obtain a target coordinate system;

[0117] Processing the second coordinate set according to the initial coordinate system and the target coordinate system to obtain a third coordinate set;

[0118] Performing distance calculation on any two coordinates in the third coordinate set to obtain a target major axis corresponding to the initial edge, and determining a maximum edge point corresponding to the initial edge according to the target major axis;

[0119] Determine the target rotation information corresponding to the initial edge according to the target major axis and the maximum edge point;

[0120] The third coordinate set is adjusted in the target coordinate system according to the target rotation information to obtain the target coordinate set corresponding to the initial edge in the target coordinate system.

[0121] In some implementations, the type identification module 203 performs, during the process of performing type identification on each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster:

[0122] Extracting features from each target edge in the subclass cluster to obtain initial color features, initial texture features, and initial morphological features;

[0123] Performing feature fusion according to the initial color feature, the initial texture feature and the initial morphological feature to obtain a target feature corresponding to the target edge;

[0124] Performing type recognition on the target edge using the target feature according to a type recognition model to obtain an initial type;

[0125] The initial types are summarized using a voting mechanism to determine the target type corresponding to the subclass cluster.

[0126] In some implementations, the type recognition module 203, in the process of obtaining the target feature corresponding to the target edge by performing feature fusion according to the initial color feature, the initial texture feature, and the initial morphological feature, performs:

[0127] Performing information analysis on all the initial color features to obtain a first entropy value corresponding to the subclass cluster in the color dimension;

[0128] Performing information analysis on all the initial texture features to obtain a second entropy value corresponding to the subclass cluster under the texture dimension;

[0129] Performing information analysis on all the initial morphological features to obtain the third entropy value corresponding to the subclass cluster in the morphological dimension;

[0130] Determine, according to the first entropy value, the second entropy value, and the third entropy value, a first weight corresponding to the subclass cluster in the color dimension, a second weight corresponding to the subclass cluster in the texture dimension, and a third weight corresponding to the subclass cluster in the morphology dimension;

[0131] Performing feature fusion on the initial color feature, the initial texture feature, and the initial morphological feature according to the first weight, the second weight, and the third weight to obtain the target feature corresponding to the target edge;

[0132] The first weight, the second weight, and the third weight are obtained according to the following formula:

[0133]

[0134] Among them, w1 represents the first weight, w2 represents the second weight, w3 represents the third weight, S1 represents the first entropy value, S2 represents the second entropy value, S3 represents the third entropy value, and m represents the number corresponding to the target edge in the subclass cluster.

[0135] In some implementations, the type labeling module 204, in the process of labeling the target edge image according to the target type to obtain the corresponding target labeling image, performs:

[0136] Obtaining a target label corresponding to the target type;

[0137] Setting an initial annotation image with the same size as the target edge image and a pixel value of a preset value;

[0138] Obtain edge information corresponding to the initial edge in the target edge image and internal information contained in the edge information;

[0139] The first pixel value corresponding to the edge information in the initial annotated image is set as the target label, and the second pixel value corresponding to the internal information is set as the target label, so as to obtain the target annotated image.

[0140] In some implementations, the density analysis module 205 performs, during the process of performing density analysis on the target annotated image to obtain a type distribution result corresponding to the target annotated image:

[0141] Determine a target window and a target step size, and determine a sliding direction corresponding to the sliding of the target window in the target annotated image;

[0142] Sliding the target window in the target annotated image according to the target step size and the sliding direction to obtain a window annotated image corresponding to the target window;

[0143] Performing density analysis on the target label in the window annotation image to obtain window density information corresponding to the target label under the target window;

[0144] Arranging the window density information in sequence according to the sliding direction to obtain sequence density information corresponding to the target tag;

[0145] Performing feature extraction on the sequence density information to obtain a label distribution result corresponding to the target label;

[0146] The type distribution result corresponding to the target annotated image is determined according to the target label and the label distribution result.

[0147] In some embodiments, the radiation alveolar damage judgment system 200 based on machine learning can be applied to a terminal device.

[0148] It should be noted that, those skilled in the art can clearly understand that, for the sake of convenience and brevity of description, the specific working process of the radiation alveolar damage judgment system 200 based on machine learning described above can refer to the corresponding process in the aforementioned embodiment of the radiation alveolar damage judgment method based on machine learning, and will not be repeated here.

[0149] See also Figure 3 , Figure 3 A schematic block diagram of the structure of a terminal device provided in an embodiment of the present invention.

[0150] like Figure 3As shown, the terminal device 300 includes a processor 301 and a memory 302, and the processor 301 and the memory 302 are connected via a bus 303, such as an I2C (Inter-integrated Circuit) bus.

[0151] Specifically, the processor 301 is used to provide computing and control capabilities to support the operation of the entire terminal device. The processor 301 can be a central processing unit (CPU), and the processor 301 can also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0152] Specifically, the memory 302 may be a Flash chip, a read-only memory (ROM) disk, an optical disk, a USB flash drive, or a mobile hard disk.

[0153] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the embodiment of the present invention, and does not constitute a limitation on the terminal device to which the embodiment of the present invention is applied. The specific server may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0154] The processor is used to run a computer program stored in the memory, and implement any one of the radiation alveolar damage judgment methods based on machine learning provided in the embodiments of the present invention when executing the computer program.

[0155] In one embodiment, the processor is used to run a computer program stored in the memory, and implements the following steps when executing the computer program:

[0156] Obtaining a target CT image corresponding to a target patient, and performing edge recognition on the target CT image to obtain a corresponding target edge image;

[0157] Performing data clustering on the corresponding initial edges in the target edge image to obtain corresponding target clustering results;

[0158] Performing type identification on each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster;

[0159] Performing type labeling on the target edge image according to the target type to obtain a corresponding target labeling image;

[0160] Performing density analysis on the target annotated image to obtain a type distribution result corresponding to the target annotated image;

[0161] The radiation alveolar injury corresponding to the target patient is classified and judged using the type distribution result according to the injury recognition model to obtain a target judgment result corresponding to the target patient.

[0162] In some implementations, the processor 301, in the process of performing data clustering on the corresponding initial edge in the target edge image to obtain the corresponding target clustering result, executes:

[0163] Obtaining a target coordinate set corresponding to the initial edge in a target coordinate system;

[0164] Determine, according to the target coordinate set, an initial characterization vector corresponding to the initial edge in the target coordinate system;

[0165] Performing dimensionality reduction processing on the initial representation vector to obtain a target representation vector corresponding to the initial edge;

[0166] Data clustering is performed on the initial edge according to the target characterization vector to obtain the target clustering result.

[0167] In some implementations, during the process of obtaining the target coordinate set corresponding to the initial edge in the target coordinate system, the processor 301 executes:

[0168] Determine an initial coordinate system, and obtain a first coordinate set corresponding to the initial edge in the initial coordinate system;

[0169] Calculating the average value of all horizontal coordinates in the first coordinate set to determine a first horizontal center corresponding to the initial edge and calculating the average value of all vertical coordinates in the first coordinate set to determine a first vertical center corresponding to the initial edge;

[0170] Collect edge coordinates of the initial edge using a target angle according to the first horizontal center and the first vertical center to obtain a second coordinate set corresponding to the initial edge;

[0171] Calculating the area of ​​the initial edge according to the second coordinate set, the first horizontal center, and the first vertical center to obtain a target area corresponding to the initial edge;

[0172] Recalculate the target center corresponding to the initial edge according to the target area and the second coordinate set to obtain a second horizontal center and a second vertical center;

[0173] The origin corresponding to the initial coordinate system is translated to the second horizontal center and the second vertical center to obtain a target coordinate system;

[0174] Processing the second coordinate set according to the initial coordinate system and the target coordinate system to obtain a third coordinate set;

[0175] Performing distance calculation on any two coordinates in the third coordinate set to obtain a target major axis corresponding to the initial edge, and determining a maximum edge point corresponding to the initial edge according to the target major axis;

[0176] Determine the target rotation information corresponding to the initial edge according to the target major axis and the maximum edge point;

[0177] The third coordinate set is adjusted in the target coordinate system according to the target rotation information to obtain the target coordinate set corresponding to the initial edge in the target coordinate system.

[0178] In some implementations, the processor 301, in the process of performing type identification on each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster, executes:

[0179] Extracting features from each target edge in the subclass cluster to obtain initial color features, initial texture features, and initial morphological features;

[0180] Performing feature fusion according to the initial color feature, the initial texture feature and the initial morphological feature to obtain a target feature corresponding to the target edge;

[0181] Performing type recognition on the target edge using the target feature according to a type recognition model to obtain an initial type;

[0182] The initial types are summarized using a voting mechanism to determine the target type corresponding to the subclass cluster.

[0183] In some implementations, the processor 301, in the process of obtaining the target feature corresponding to the target edge by performing feature fusion according to the initial color feature, the initial texture feature, and the initial morphological feature, executes:

[0184] Performing information analysis on all the initial color features to obtain a first entropy value corresponding to the subclass cluster in the color dimension;

[0185] Performing information analysis on all the initial texture features to obtain a second entropy value corresponding to the subclass cluster under the texture dimension;

[0186] Performing information analysis on all the initial morphological features to obtain the third entropy value corresponding to the subclass cluster in the morphological dimension;

[0187] Determine, according to the first entropy value, the second entropy value, and the third entropy value, a first weight corresponding to the subclass cluster in the color dimension, a second weight corresponding to the subclass cluster in the texture dimension, and a third weight corresponding to the subclass cluster in the morphology dimension;

[0188] Performing feature fusion on the initial color feature, the initial texture feature, and the initial morphological feature according to the first weight, the second weight, and the third weight to obtain the target feature corresponding to the target edge;

[0189] The first weight, the second weight, and the third weight are obtained according to the following formula:

[0190]

[0191]

[0192] Among them, w1 represents the first weight, w2 represents the second weight, w3 represents the third weight, S1 represents the first entropy value, S2 represents the second entropy value, S3 represents the third entropy value, and m represents the number corresponding to the target edge in the subclass cluster.

[0193] In some implementations, the processor 301, in the process of performing type labeling on the target edge image according to the target type to obtain a corresponding target labeled image, executes:

[0194] Obtaining a target label corresponding to the target type;

[0195] Setting an initial annotation image with the same size as the target edge image and a pixel value of a preset value;

[0196] Obtain edge information corresponding to the initial edge in the target edge image and internal information contained in the edge information;

[0197] The first pixel value corresponding to the edge information in the initial annotated image is set as the target label, and the second pixel value corresponding to the internal information is set as the target label, so as to obtain the target annotated image.

[0198] In some implementations, the processor 301, during the process of performing density analysis on the target annotated image to obtain a type distribution result corresponding to the target annotated image, executes:

[0199] Determine a target window and a target step size, and determine a sliding direction corresponding to the sliding of the target window in the target annotated image;

[0200] Sliding the target window in the target annotated image according to the target step size and the sliding direction to obtain a window annotated image corresponding to the target window;

[0201] Performing density analysis on the target label in the window annotation image to obtain window density information corresponding to the target label under the target window;

[0202] Arranging the window density information in sequence according to the sliding direction to obtain sequence density information corresponding to the target tag;

[0203] Performing feature extraction on the sequence density information to obtain a label distribution result corresponding to the target label;

[0204] The type distribution result corresponding to the target annotated image is determined according to the target label and the label distribution result.

[0205] It should be noted that those skilled in the art can clearly understand that, for the sake of convenience and brevity of description, the specific working process of the terminal device described above can refer to the corresponding process in the aforementioned embodiment of the radiation alveolar damage judgment method based on machine learning, and will not be repeated here.

[0206] An embodiment of the present invention also provides a storage medium for computer-readable storage, wherein the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any method for determining radiation alveolar damage based on machine learning as provided in the description of the embodiment of the present invention.

[0207] The storage medium may be an internal storage unit of the terminal device described in the foregoing embodiment, such as a hard disk or memory of the terminal device. The storage medium may also be an external storage device of the terminal device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc., equipped on the terminal device.

[0208] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In a hardware embodiment, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0209] It should be understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.

[0210] The serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for judging radiation alveolar damage based on machine learning, characterized in that: The method comprises: Obtaining a target CT image corresponding to a target patient, and performing edge recognition on the target CT image to obtain a corresponding target edge image; Performing data clustering on the corresponding initial edges in the target edge image to obtain corresponding target clustering results; Performing type identification on each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster; Performing type labeling on the target edge image according to the target type to obtain a corresponding target labeling image; Performing density analysis on the target annotated image to obtain a type distribution result corresponding to the target annotated image; The radiation alveolar injury corresponding to the target patient is classified and judged using the type distribution result according to the injury recognition model to obtain a target judgment result corresponding to the target patient.

2. The method according to claim 1, characterized in that The performing data clustering on the corresponding initial edges in the target edge image to obtain the corresponding target clustering result includes: Obtaining a target coordinate set corresponding to the initial edge in a target coordinate system; Determine, according to the target coordinate set, an initial characterization vector corresponding to the initial edge in the target coordinate system; Performing dimensionality reduction processing on the initial representation vector to obtain a target representation vector corresponding to the initial edge; Data clustering is performed on the initial edge according to the target characterization vector to obtain the target clustering result.

3. The method according to claim 2, characterized in that The obtaining of a target coordinate set corresponding to the initial edge in a target coordinate system includes: Determine an initial coordinate system, and obtain a first coordinate set corresponding to the initial edge in the initial coordinate system; Calculating the average value of all horizontal coordinates in the first coordinate set to determine a first horizontal center corresponding to the initial edge and calculating the average value of all vertical coordinates in the first coordinate set to determine a first vertical center corresponding to the initial edge; Collect edge coordinates of the initial edge using a target angle according to the first horizontal center and the first vertical center to obtain a second coordinate set corresponding to the initial edge; Calculating the area of ​​the initial edge according to the second coordinate set, the first horizontal center, and the first vertical center to obtain a target area corresponding to the initial edge; Recalculate the target center corresponding to the initial edge according to the target area and the second coordinate set to obtain a second horizontal center and a second vertical center; The origin corresponding to the initial coordinate system is translated to the second horizontal center and the second vertical center to obtain a target coordinate system; Processing the second coordinate set according to the initial coordinate system and the target coordinate system to obtain a third coordinate set; Performing distance calculation on any two coordinates in the third coordinate set to obtain a target major axis corresponding to the initial edge, and determining a maximum edge point corresponding to the initial edge according to the target major axis; Determine the target rotation information corresponding to the initial edge according to the target major axis and the maximum edge point; The third coordinate set is adjusted in the target coordinate system according to the target rotation information to obtain the target coordinate set corresponding to the initial edge in the target coordinate system.

4. The method according to claim 1, characterized in that: The step of performing type identification on each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster includes: Extracting features from each target edge in the subclass cluster to obtain initial color features, initial texture features, and initial morphological features; Performing feature fusion according to the initial color feature, the initial texture feature and the initial morphological feature to obtain a target feature corresponding to the target edge; Performing type recognition on the target edge using the target feature according to a type recognition model to obtain an initial type; The initial types are summarized using a voting mechanism to determine the target type corresponding to the subclass cluster.

5. The method according to claim 4, characterized in that The step of performing feature fusion according to the initial color feature, the initial texture feature and the initial morphological feature to obtain the target feature corresponding to the target edge includes: Performing information analysis on all the initial color features to obtain a first entropy value corresponding to the subclass cluster in the color dimension; Performing information analysis on all the initial texture features to obtain a second entropy value corresponding to the subclass cluster under the texture dimension; Performing information analysis on all the initial morphological features to obtain the third entropy value corresponding to the subclass cluster in the morphological dimension; Determine, according to the first entropy value, the second entropy value, and the third entropy value, a first weight corresponding to the subclass cluster in the color dimension, a second weight corresponding to the subclass cluster in the texture dimension, and a third weight corresponding to the subclass cluster in the morphology dimension; Performing feature fusion on the initial color feature, the initial texture feature, and the initial morphological feature according to the first weight, the second weight, and the third weight to obtain the target feature corresponding to the target edge; The first weight, the second weight, and the third weight are obtained according to the following formula: Among them, w1 represents the first weight, w2 represents the second weight, w3 represents the third weight, S1 represents the first entropy value, S2 represents the second entropy value, S3 represents the third entropy value, and m represents the number corresponding to the target edge in the subclass cluster.

6. The method according to claim 1, characterized in that The step of labeling the target edge image according to the target type to obtain a corresponding target labeled image includes: Obtaining a target label corresponding to the target type; Setting an initial annotation image with the same size as the target edge image and a pixel value of a preset value; Obtain edge information corresponding to the initial edge in the target edge image and internal information contained in the edge information; The first pixel value corresponding to the edge information in the initial annotated image is set as the target label, and the second pixel value corresponding to the internal information is set as the target label, so as to obtain the target annotated image.

7. The method according to claim 6, characterized in that The performing density analysis on the target annotated image to obtain a type distribution result corresponding to the target annotated image includes: Determine a target window and a target step size, and determine a sliding direction corresponding to the sliding of the target window in the target annotated image; Sliding the target window in the target annotated image according to the target step size and the sliding direction to obtain a window annotated image corresponding to the target window; Performing density analysis on the target label in the window annotation image to obtain window density information corresponding to the target label under the target window; Arranging the window density information in sequence according to the sliding direction to obtain sequence density information corresponding to the target tag; Performing feature extraction on the sequence density information to obtain a label distribution result corresponding to the target label; The type distribution result corresponding to the target annotated image is determined according to the target label and the label distribution result.

8. A radiation alveolar damage judgment system based on machine learning, characterized in that: include: A data processing module, used to obtain a target CT image corresponding to a target patient, and perform edge recognition on the target CT image to obtain a corresponding target edge image; A data clustering module, used for performing data clustering on the corresponding initial edges in the target edge image to obtain the corresponding target clustering results; A type identification module, used to perform type identification on each subclass cluster in the target clustering result to obtain the target type corresponding to the subclass cluster; A type labeling module, used for labeling the target edge image according to the target type to obtain a corresponding target labeling image; A density analysis module, used to perform density analysis on the target annotated image to obtain a type distribution result corresponding to the target annotated image; The injury judgment module is used to classify and judge the radiation alveolar injury corresponding to the target patient using the type distribution result according to the injury recognition model, and obtain the target judgment result corresponding to the target patient.

9. A terminal device, characterized in that: The terminal device includes a processor and a memory; The memory is used to store computer programs; The processor is used to execute the computer program and implement the method for determining radiation alveolar damage based on machine learning as described in any one of claims 1 to 7 when executing the computer program.

10. A computer storage medium for computer storage, characterized in that: The computer storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the method for determining radiation alveolar damage based on machine learning as described in any one of claims 1 to 7.

Images