Interstitial lung disease classification method and system based on multi-dimensional feature labels

Abstract

The application discloses an interstitial lung disease classification method based on a multidimensional feature label, and the method comprises the following steps: acquiring lung CT image information, wherein the CT image information comprises a CT image and associated clinical information; performing regional segmentation on the CT image to obtain three-dimensional space information of each region; performing target feature detection on the CT image based on the three-dimensional space information of each region to obtain three-dimensional morphological features of the target features; performing deep feature extraction and mathematical calculation on the target features to obtain high-order mathematical features; fusing the three-dimensional morphological features and the high-order mathematical features to generate high-dimensional fused image features; encoding the associated clinical information and combining the high-dimensional fused image features to generate a multidimensional feature label; and generating a classification result of interstitial lung disease based on the multidimensional feature label through a preset classification model. Through multidimensional feature analysis from the image level and the clinical level, various interstitial lung diseases can be quickly and accurately classified.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method and system for classifying interstitial lung diseases based on multidimensional feature labels. Background Art

[0002] Interstitial lung disease (ILD) is a group of diseases that primarily affect the alveolar wall and also include lesions of the tissue surrounding the alveoli and their adjacent supporting structures. There are several main types of interstitial lung diseases, such as UIP, NSIP, OP, and pUIP. How to quickly determine the type of interstitial lung disease helps to promptly determine treatment options and improve diagnostic efficiency.

[0003] Existing ILD classification methods mainly rely on doctors' empirical judgment of images or use visual models to retrieve medical images. However, since interstitial lung diseases have a variety of manifestations in imaging examinations, including honeycomb, reticular, ground-glass, etc., and the imaging manifestations of various types of interstitial lung diseases lack specificity and are prone to overlap, existing classification methods are difficult to quickly determine the category of interstitial lung diseases. Summary of the Invention

[0004] The purpose of this application is to provide an interstitial lung disease classification method and system based on multidimensional feature labels. In addition to in-depth analysis of the pathological characteristics of interstitial lung diseases in different dimensions, the relevant clinical information is also effectively integrated to form multidimensional feature labels. Using multidimensional feature labels as targets, interstitial lung diseases can be quickly and accurately classified.

[0005] In a first aspect, the present application provides a method for classifying interstitial lung diseases based on multidimensional feature labels, which adopts the following technical solutions:

[0006] Acquiring lung CT imaging information, wherein the CT imaging information includes a CT image and associated clinical information;

[0007] Perform regional segmentation on the CT image using a preset three-dimensional model to obtain three-dimensional spatial information of each region;

[0008] Based on the three-dimensional spatial information of each area, target feature detection is performed on the CT image to obtain the three-dimensional morphological features of the target features;

[0009] Perform deep feature extraction and mathematical calculation on target features to obtain high-order mathematical features;

[0010] Fuse three-dimensional morphological features and high-order mathematical features to generate high-dimensional fused image features;

[0011] Encode the relevant clinical information and combine it with high-dimensional fusion image features to generate multidimensional feature labels;

[0012] Based on multidimensional feature labels, the classification results of interstitial lung diseases are generated through a preset classification model.

[0013] Through the above technical solution, the three-dimensional morphological features of lesion characteristics can be integrated with the depth features and mathematical features, and the lesion characteristics can be analyzed and modeled from multiple dimensions, thereby enhancing the description of the overall to local correlation information of the lesion characteristics, which helps to quickly and accurately classify various interstitial lung diseases. At the same time, the integration of clinical information can further improve the classification efficiency.

[0014] Optionally, the three-dimensional morphological features include morphological feature information and spatial proportion information. The target feature detection is performed on the CT image based on the three-dimensional spatial information of each region to obtain the three-dimensional morphological features of the target features, including:

[0015] Perform target feature detection on CT images to obtain morphological feature information and three-dimensional spatial information of the target features;

[0016] Based on the three-dimensional spatial information of each area, match the spatial position information of the target feature in the associated area;

[0017] Based on the three-dimensional spatial information of each region and target feature, the volume of each region and the volume of the target feature are calculated and obtained respectively;

[0018] Based on the volume of each area and the volume of the target feature, the spatial proportion information of the target feature is calculated and obtained through the spatial position information of the target feature in the associated area.

[0019] Optionally, performing deep feature extraction on the target features to obtain high-order mathematical features includes:

[0020] Based on the target features, deep features are extracted through a preset deep neural network;

[0021] Perform mathematical calculations on the deep features to obtain mathematical features;

[0022] The deep features are combined with the mathematical features to generate high-order mathematical features.

[0023] Optionally, fusing the three-dimensional morphological features with the high-order mathematical features to generate high-dimensional fused image features includes:

[0024] Based on three-dimensional morphological features and high-order mathematical features, correlation features are obtained through a preset correlation model;

[0025] Generate fusion weights based on correlation features;

[0026] Based on the fusion weights, the three-dimensional morphological features and high-order mathematical features are fused to generate high-dimensional fused image features.

[0027] Optionally, the associated clinical information includes extra- and intra-table information, and encoding the associated clinical information and combining it with the high-dimensional fusion image features to generate a multi-dimensional feature label includes:

[0028] Through the preset data analysis model, the influence coefficient of off-balance sheet information on various types of interstitial lung diseases is obtained, and the first influence vector is generated;

[0029] Through the preset disease diagnosis library, the matching degree between the information in the table and various interstitial lung diseases is obtained, and the second influence vector is generated;

[0030] encoding the clinical information based on the first influence vector and the second influence vector to generate a clinical feature label;

[0031] Integrate clinical feature labels with high-dimensional fusion image features to generate multidimensional feature labels.

[0032] Optionally, after generating the classification results of interstitial lung disease based on the multidimensional feature labels by using a preset classification model, the method further includes:

[0033] According to the classification results, the dynamic evolution characteristics are obtained by presetting the dynamic evolution model;

[0034] Based on the dynamic evolution characteristics, the current CT image is reconstructed in three dimensions to generate a new CT image;

[0035] Add new CT images to the preset training database.

[0036] Optionally, obtaining dynamic evolution features based on the classification results by using a preset dynamic evolution model includes:

[0037] According to the classification results, a corresponding preset dynamic evolution model is matched from a preset dynamic evolution database;

[0038] Based on the current CT image and target features, variable features are generated through a preset dynamic evolution model;

[0039] Based on the associated clinical information, the variable features are modified to generate dynamically evolving features.

[0040] Optionally, after adding the newly added CT images to the preset training database, the method further includes:

[0041] Determine whether the amount of newly added CT image data in the current preset training database reaches a preset update threshold;

[0042] If so, the preset classification model is iteratively updated based on the current preset training database to generate a new preset classification model.

[0043] In a second aspect, the present application provides an interstitial lung disease classification system based on multidimensional feature labels, comprising:

[0044] The data acquisition module 101 is used to acquire lung CT imaging information, which includes CT images and associated clinical information;

[0045] The region segmentation module 102 is used to segment the CT image into regions using a preset three-dimensional model to obtain three-dimensional spatial information of each region;

[0046] The feature detection module 103 is used to perform target feature detection on the CT image based on the three-dimensional spatial information of each region to obtain the three-dimensional morphological features of the target features;

[0047] The multi-dimensional feature fusion module 104 is used to perform deep feature extraction and mathematical calculation on the target features to obtain high-order mathematical features, fuse the three-dimensional morphological features and the high-order mathematical features to generate high-dimensional fused image features, encode the associated clinical information, and combine the high-dimensional fused image features to generate multi-dimensional feature labels;

[0048] The classification result generating module 105 is configured to generate a classification result of interstitial lung disease based on the multi-dimensional feature labels and through a preset classification model.

[0049] In a third aspect, the present application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and execute the above-mentioned interstitial lung disease classification method based on multidimensional feature labels.

[0050] In summary, this application first integrates the three-dimensional morphological features of lesion characteristics with the depth features and mathematical features at the imaging level, and combines clinical information to construct multidimensional feature labels, analyzes and models the lesion characteristics from multiple dimensions, increases the differentiated expression of features, and helps to quickly and accurately classify various interstitial lung diseases; in addition, when integrating the morphological features and mathematical features of lesion characteristics, by constructing an association model, the intrinsic connection between lesion characteristics in different manifestations can be explored, the expression of lesion characteristics is further enhanced, and the classification efficiency can be improved; in addition, by constructing a dynamic evolution model, the data involved in the classification can be effectively expanded based on the evolution law of the lesion characteristics, which facilitates the subsequent iterative optimization of the classification model to enhance the generalization ability of the classification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of a method for classifying interstitial lung diseases based on multidimensional feature labels provided in an embodiment of the present application;

[0052] Figure 2 This is an example diagram of a lung CT image provided in an embodiment of the present application;

[0053] Figure 3 This is a flow chart of obtaining three-dimensional morphological features of target features provided by an embodiment of the present application;

[0054] Figure 4 This is a flowchart of performing deep feature extraction on target features and obtaining high-order mathematical features provided by an embodiment of the present application;

[0055] Figure 5 This is a flowchart of generating multi-dimensional feature labels provided by an embodiment of the present application;

[0056] Figure 6 This is an example diagram of the classification effect of interstitial lung disease provided by the embodiment of the present application;

[0057] Figure 7 This is a schematic diagram of an interstitial lung disease classification system based on multidimensional feature labels provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] The following is combined with Figure 1 -Attached Figure 7 , further details of this application are given.

[0059] This application provides a classification method for interstitial lung disease based on multidimensional feature labels, see Figure 1 , including the following steps:

[0060] S100: Acquire lung CT image information.

[0061] Among them, CT imaging information includes CT images and associated clinical information. CT images are three-dimensional DICOM images. DICCOM is a commonly used medical data imaging format. In addition to image pixel data, it also contains CT scanning information and image spatial positioning information, which can help to reconstruct the image in three dimensions so as to observe lung lesions from different angles and planes; associated clinical information includes data information such as height, weight, age and lung function.

[0062] The original lung CT images are shown in Figure 2 After obtaining the original CT image, the original CT image will be preprocessed accordingly. For example, the image is cropped and scaled according to the network model parameters to adjust the image size. That is, the CT image here is a preprocessed image.

[0063] S200 , performing regional segmentation on the CT image using a preset three-dimensional model to obtain three-dimensional spatial information of each region.

[0064] Because interstitial lung disease is often accompanied by changes in lung lobe morphology and abnormal lung parenchymal density, observing the characteristic changes in each lung lobe can help detect subtle lesions of the disease at an early stage. In addition, the symptoms of different types of interstitial lung disease also vary in different regions.

[0065] Therefore, in an embodiment of the present application, after obtaining the lung CT image, the CT image will first be segmented to obtain three-dimensional spatial information of the left lung, right lung, and left upper lobe, left lower lobe, right upper lobe, right middle lobe, and right lower lobe regions. The three-dimensional spatial information here represents the spatial distribution and morphological size of each region in the lung.

[0066] The preset three-dimensional model here uses the lung parenchyma, lung fissures, and each lung lobe area as annotation information. It uses a large amount of lung CT image data and adopts a three-dimensional deep model as the architecture, such as the commonly used 3D-UNet, VNet and other model architectures, to generate a lung area segmentation model through training.

[0067] S300 : Based on the three-dimensional spatial information of each region, target feature detection is performed on the CT image to obtain three-dimensional morphological features of the target feature.

[0068] Among them, the target features refer to the pathological characteristics of interstitial lung diseases, such as honeycomb, reticular, ground glass, consolidation, emphysema and other visual features; the three-dimensional morphological features represent the morphological and distribution characteristics of the lesion features, that is, the three-dimensional morphological features include morphological feature information and spatial proportion information. The morphological feature information refers to the shape, size, convexity, contour and edge information of the lesion features; the spatial proportion information refers to the spatial position, volume size and volume proportion of the lesion features in each associated area.

[0069] Because different types of interstitial lung diseases have different visual characteristics, clarifying the distribution of lesion characteristics in different areas of the lungs will help to accurately determine the type of interstitial lung disease.

[0070] In the embodiment of the present application, target feature detection is performed on the CT image based on the three-dimensional spatial information of each region to obtain the three-dimensional morphological features of the target features. Figure 3 , specifically including the following steps:

[0071] S310: Detect target features on the CT image to obtain morphological feature information and three-dimensional spatial information of the target features.

[0072] S320 : Based on the three-dimensional spatial information of each region, match the spatial position information of the target feature in the associated region.

[0073] S330 : Based on the three-dimensional spatial information of each region and the target feature, respectively calculate and obtain the volume of each region and the volume of the target feature.

[0074] S340 : Based on the volume of each region and the volume of the target feature, and through the spatial position information of the target feature in the associated region, calculate and obtain spatial proportion information of the target feature.

[0075] In the embodiment of the present application, target feature detection is first performed on the CT image to obtain morphological feature information and three-dimensional spatial information of the target feature.

[0076] The feature detection of CT images mentioned here is similar to the above-mentioned regional segmentation of CT images through a preset three-dimensional model to obtain three-dimensional spatial information of each region. Both use a preset three-dimensional model, but the detection targets are different. Regional segmentation is based on the detection of lung parenchyma and lung fissures as the benchmark for division and identification, while obtaining target features, that is, lesion features, uses visual features such as honeycomb, reticular, ground glass, consolidation, emphysema, etc. as detection targets for image detection, and the position of the detected targets is marked, that is, the three-dimensional spatial information of the target features can also be obtained. Morphological feature information is further morphologically processed on the target features based on the detection of target features. For example, instance segmentation algorithms such as MaskR-CNN are used to subdivide the target features and obtain morphological information such as the shape, size, convexity, contour and edge of the target features through mask operations. Refined detection of the morphology of the lesions also helps to classify and determine the type of interstitial lung disease.

[0077] In order to determine the distribution of lesion features in various areas of the lungs, spatial position relationship matching can be performed by combining the three-dimensional spatial information of each area and the three-dimensional spatial information of the target features to determine in which area or areas the target features are located, namely the so-called associated areas. In this way, the spatial position relationship between the target features and the associated areas can be obtained, that is, the spatial position information of the target features in the associated areas.

[0078] In addition, considering the volume of each lung lobe area and the volume proportion of the lesion characteristics can quantitatively assess the scope and severity of interstitial lung disease lesions, which is also helpful for the classification of interstitial lung diseases.

[0079] Therefore, after obtaining the three-dimensional spatial information of each region and the three-dimensional spatial information of the target feature, the volume size of each region and the volume size of the target feature can be calculated respectively. With the help of the spatial position information of the target region in the associated region, the spatial proportion of the target feature can be calculated to help evaluate the degree of lesion characteristics and subsequent lesion trends, and at the same time, it can also assist in determining the category of interstitial lung disease.

[0080] For example, for a lung CT image, through the above operations, the volume size of each region and the spatial proportion information of the target feature are obtained as shown in Table 1.

[0081] Table 1:

[0082]

[0083] Among them, "normal" refers to the normal tissues in the lung area except for the pathological features. The data in the table can intuitively reflect the main location distribution and spatial proportion information of each pathological feature in the entire lung area.

[0084] S400: Perform deep feature extraction and mathematical calculation on target features to obtain high-order mathematical features.

[0085] When it comes to the classification of interstitial lung diseases, both the morphological characteristics and spatial proportion information of the lesion characteristics are low-order dimensional references. In the face of the lack of specificity and easy overlap in the imaging manifestations of various types of interstitial lung diseases, they are not sufficient as the basis for classification.

[0086] Therefore, in the embodiment of the present application, deeper feature mining and analysis will be performed on the target features, that is, the lesion features, that is, deep feature extraction and mathematical calculations are performed on the target features to obtain high-order mathematical features.

[0087] Specifically, deep feature extraction is performed on the target features to obtain high-order mathematical features. Figure 4 , specifically including the following steps:

[0088] S410: Based on the target features, extract deep features through a preset deep neural network.

[0089] S420: Perform mathematical calculations on the depth features to obtain mathematical features.

[0090] S430: Fusing the deep features with the mathematical features to generate high-order mathematical features.

[0091] Deep features refer to feature representations learned through deep neural networks. They can capture complex, abstract feature expressions and semantic information in images, and have a good description of the overall structure and semantic content of the image. For example, in lung CT images, deep features can learn the comprehensive characteristics of texture, shape, density, and other aspects corresponding to different types of interstitial lung diseases, helping to distinguish different types of interstitial lung diseases from a macroscopic perspective.

[0092] High-order mathematical features are more statistical and structural features extracted through specific mathematical calculation methods based on deep features. For example, the gray-level co-occurrence matrix can describe the spatial correlation of pixel grayscale values ​​in an image and reflect information such as texture thickness and directionality. These high-order mathematical features can more finely characterize the local characteristics and microstructure of the image, are more sensitive to subtle characteristics and changes in the disease, and help to distinguish different types of interstitial lung diseases from a microscopic perspective.

[0093] First, the target features will be extracted through a preset deep neural network. That is, the data representation of the target features is used as input, and the preset deep neural network is used to extract the deep features to obtain the deep features. The preset deep neural network here is equivalent to a conventional convolutional neural network performing multi-layer convolution processing, such as using network structures such as GGNet and ResNet.

[0094] Then, mathematical calculations are performed on the deep features to obtain mathematical features. For example, statistical calculations are performed on the extracted deep features, such as calculating the mean, variance, standard deviation, etc., which can reflect the distribution of the deep features; high-order statistics such as skewness, kurtosis and covariance matrix are calculated to measure the degree of asymmetry of the feature distribution and the correlation between different deep features.

[0095] Finally, the deep features and mathematical features are integrated to generate high-order mathematical features, which will be used as a reference for subsequent judgment of the interstitial lung disease category.

[0096] The fusion here can adopt simple feature splicing, which splices the extracted deep features and the calculated mathematical features to form a new feature vector. It can also adopt feature weighted fusion, assigning different weights to the deep features and mathematical features, and then performing weighted fusion. Of course, the attention mechanism can also be introduced to allow the model to automatically learn the importance of deep features and mathematical features, and fuse the features according to the attention weight. This application does not make specific limitations.

[0097] S500: Fusing the three-dimensional morphological features and the high-order mathematical features to generate high-dimensional fused image features.

[0098] Since there may be a certain correlation between three-dimensional morphological features and high-order mathematical features, for example, morphological information such as edge roughness in the three-dimensional morphology of the lesion may be correlated with the texture complexity reflected by certain statistics in the high-order mathematical features. By establishing intrinsic connections, in addition to discovering the complementary parts of the two features in describing the lesions, the representation of the associated features can also be strengthened.

[0099] Therefore, in the embodiment of the present application, when the three-dimensional morphological features and the high-order mathematical features are integrated, a correlation analysis is also performed on the three-dimensional morphological features and the high-order mathematical features to further enhance the description of the lesion characteristics.

[0100] Specifically, the fusion of three-dimensional morphological features and high-order mathematical features to produce high-dimensional fused image features includes the following steps:

[0101] S510 , based on the three-dimensional morphological features and the high-order mathematical features, obtain correlation features through a preset correlation model.

[0102] S520: Generate dynamic fusion weights based on the associated features.

[0103] S530: Based on the fusion weight, the three-dimensional morphological features and the high-order mathematical features are fused to generate high-dimensional fused image features.

[0104] Among them, the preset association model is to generate association features by constructing a joint embedding space, mapping three-dimensional morphological features and high-order mathematical features into the same space, and then performing correlation statistics on the features and setting correlation thresholds.

[0105] First, by presetting the correlation model, correlation features can be extracted from 3D morphological features and high-order mathematical features. Fusion weights are then dynamically adjusted based on the correlation features. Finally, based on the generated fusion weights, the 3D morphological features and high-order mathematical features are fused to generate high-dimensional fused image features. Compared to simply adding or concatenating the two features, this fusion method can fully exploit the complementary and correlation information between the features, achieving a 1+1 greater than 2 effect.

[0106] S600: Encode the associated clinical information and generate multidimensional feature labels by combining high-dimensional fusion image features.

[0107] Among them, the associated clinical information includes external information and internal information. The external information represents the patient's body information such as height, weight, age, etc.; the internal information represents the patient's lung function data and clinical symptom information.

[0108] The pathology of interstitial lung disease is also related to the patient's body shape and physical condition. For example, UIP (usual interstitial pneumonia) typically develops in people over 50 years old, with the incidence gradually increasing with age. NSIP (nonspecific interstitial pneumonia) tends to develop in younger people, typically in their 40s and 50s. Furthermore, height, primarily through its influence on lung volume, and weight, through its influence on lung load, both influence the pathology of interstitial diseases. Lung function is particularly directly related to the pathology.

[0109] Therefore, in the embodiment of this application, related clinical information, namely the patient's height, weight, age, lung function, and other data information, is also incorporated into the classification considerations. That is, after obtaining the high-dimensional fusion image features, the related clinical information is also encoded and combined with the high-dimensional fusion image features to generate a multidimensional feature label, which is used as the final classification target.

[0110] Specifically, the associated clinical information is encoded and combined with high-dimensional fusion image features to generate multidimensional feature labels, see Figure 5 , including the following steps:

[0111] S610. Obtain the influence coefficient of off-balance sheet information on various types of interstitial lung diseases through a preset data analysis model, and generate a first influence vector.

[0112] S620: Obtain the matching degree between the information in the table and various types of interstitial lung diseases through a preset disease diagnosis library, and generate a second influence vector.

[0113] S630: Encode the clinical information based on the first influence vector and the second influence vector to generate a clinical feature label.

[0114] S640: Integrate the clinical feature labels with the high-dimensional fusion image features to generate multi-dimensional feature labels.

[0115] The first influence vector and the second influence vector respectively represent the influence of the external information and the internal information on the current determination of the interstitial lung disease category.

[0116] The preset data analysis model is primarily based on historical interstitial lung disease diagnostic data. This model, generated through data analysis and modeling, can be used to calculate the impact coefficient of off-table information on various interstitial lung diseases. The preset disease diagnosis database, constructed using historical interstitial lung disease diagnostic data, can be used to determine the degree of match between the current table and various interstitial lung diseases.

[0117] In an embodiment of the present application, first, through a preset data analysis model, the influence coefficient of off-balance sheet information on various types of interstitial lung diseases can be obtained, and a first influence vector can be generated.

[0118] Specifically, the patient's clinical information is input into a preset data analysis model. The model calculates the influence coefficient of the patient's external information on various types of interstitial lung diseases, arranges these influence coefficients in a certain order, and generates a first influence vector. This is equivalent to a possibility assessment of the category of interstitial lung disease based on the current patient's clinical information, and the first influence vector is a data representation of the assessment result.

[0119] Next, through the preset disease diagnosis library, the matching degree between the information in the table and various types of interstitial lung diseases is obtained, and a second influence vector is generated.

[0120] Specifically, the patient's clinical table information is matched with the data in the preset disease diagnosis library. By calculating the similarity or other relevant indicators, the matching degree between the current table information and various types of interstitial lung diseases is obtained. These matching degrees are sorted by category to generate a second influence vector, which is equivalent to conducting another dimension of possibility assessment of the category of interstitial lung disease based on the current patient's clinical table information, and the second influence vector is a data representation of another evaluation result.

[0121] Then, the clinical information is encoded based on the first influence vector and the second influence vector to generate a clinical feature label.

[0122] The first influence vector and the second influence vector here do not conflict with each other, so they will be combined to assist in the classification of interstitial lung diseases and to encode clinical information, that is, to encode extra- and intra-table information respectively. The first influence vector and the second influence vector can be regarded as weight parameters for extra- and intra-table information respectively, and the encoded information is configured with weight parameters to form clinical feature labels.

[0123] Finally, the clinical feature labels are integrated with the high-dimensional fused image features to generate multidimensional feature labels. This integration serves as a classification target for classifying interstitial lung diseases. Multidimensional feature labels combine clinical information and imaging features to more comprehensively describe the patient's condition and disease characteristics, providing more valuable information for classifying interstitial lung diseases.

[0124] The generated clinical feature labels are integrated with the high-dimensional fused image features. Methods such as splicing and weighted fusion can be used to combine the clinical feature labels and the high-dimensional fused image features into a multidimensional feature vector. For example, the clinical feature labels are used as a sub-vector and the high-dimensional fused image features are used as another sub-vector, and the splicing operation is performed to form a complete multidimensional feature vector.

[0125] S700. Based on the multidimensional feature labels, a classification result of interstitial lung disease is generated through a preset classification model.

[0126] Among them, the preset classification model uses multi-dimensional feature labels as classification targets, and uses a large amount of lung CT data and clinical diagnosis information as data to obtain an interstitial lung disease classification model through training.

[0127] By generating multidimensional feature labels based on the current lung CT images and associated clinical information, and with the help of a preset classification model, the classification results of interstitial lung diseases can be directly generated.

[0128] See also Figure 6 ,The upper part of the figure shows the maximum possible classification results of ,lung CT images, as well as the confidence of each category (UIP, NSIP, OP, pUIP), ,and the lower part shows the area where the lesion features are located.

[0129] In the embodiments of the present application, seven conventional machine learning models were used to generate a preset classification model, including SVM, random forest, MLP, XGBoost, KNN, logistic regression (LR), and decision tree. In addition, an integrated model was added. Classification experiments were carried out for each classification model, and the experimental results are shown in Table 2.

[0130] Table 2:

[0131]

[0132] According to the data in the table, it can be seen that the classification model constructed with multidimensional feature labels as classification targets can basically achieve an accuracy (Accuracy) and precision (Precision) of more than 85%, and the AUC is relatively high, indicating that the classification model has shown certain effectiveness and stability and has good potential for clinical auxiliary diagnosis, but there is still much room for improvement.

[0133] Because different interstitial lung diseases differ in clinical manifestations, imaging characteristics, pathophysiological mechanisms, etc., there is a problem of difficulty in generalizing the model.

[0134] Therefore, in an embodiment of the present application, after obtaining the classification results, the data diversity will be increased according to the classification results, and the preset classification model will be further optimized and trained to improve the generalization ability of the model.

[0135] Specifically, based on the multidimensional feature labels, after generating the classification results of interstitial lung disease through a preset classification model, the following steps are also included:

[0136] S810: According to the classification result, obtain dynamic evolution features through a preset dynamic evolution model.

[0137] S820: Based on the dynamic evolution characteristics, perform three-dimensional reconstruction on the current CT image to generate a new CT image.

[0138] S830: Add the newly added CT image to the preset training database.

[0139] Among them, the preset dynamic evolution model is because the changes in symptoms of each type of interstitial lung disease from the early to the middle and then to the late stages follow a certain pattern. By conducting big data analysis and modeling based on historical data, a model is constructed to simulate this pattern. Through this model, for various types of interstitial lung diseases, the pathological status of the current interstitial lung disease in the next period of time or the later stage can be inferred.

[0140] Dynamic evolution features represent the pathological features generated based on the subsequent changing state of the current disease.

[0141] In order to enhance the data diversity of the preset classification model, after determining the classification result of the current lung CT image, the preset dynamic evolution model can be used to expand the data, that is, the dynamic evolution features are obtained through the preset dynamic evolution model, and then the current lung CT image is three-dimensionally reconstructed according to the dynamic evolution features. In this way, a new lung CT image can be generated, and then the newly added lung CT image is added to the preset training database together with the current classification result and related clinical information, which can be used as subsequent expanded training data.

[0142] For example, if the current lung CT image is diagnosed as early-stage UIP, the disease state of mid-stage UIP can be inferred. Based on the inferred situation, the dynamic evolution characteristics can be obtained. Based on the current CT image, a new CT image can be generated through three-dimensional reconstruction. In this way, a new set of lung CT image data classified as mid-stage UIP is added.

[0143] It is worth noting that the use of preset dynamic evolution models for speculation cannot be reversed. For example, the early symptom status cannot be inferred from the middle or late stages. This is mainly because the development of the disease is an irreversible process, and the later symptom status may be affected by multiple factors, resulting in information loss or confusion. For example, the lung fibrosis changes that appear in the later stage may be the result of the gradual development of multiple early lesions. It is difficult to accurately infer the specific symptom status in the early stage based solely on the later fibrosis manifestations. Moreover, different types of interstitial lung diseases may have some similar manifestations in the middle or late stages, which also increases the difficulty of reverse inference.

[0144] In addition, cross-stage speculation is not possible. For example, the symptoms of early UIP cannot be directly inferred from the symptoms of late UIP. Because there are too many variable factors, cross-stage speculation is prone to large deviations.

[0145] This is because the evolution of the disease is greatly affected by the individual, which means it is closely related to clinical information.

[0146] Therefore, in the embodiment of the present application, clinical information will also be added. On the basis of the preset dynamic evolution model, clinical information will also be taken into consideration to further determine the evolution trend of the disease.

[0147] Specifically, according to the classification results, the dynamic evolution features are obtained by presetting the dynamic evolution model, including the following steps:

[0148] S811 . According to the classification result, a corresponding preset dynamic evolution model is matched from a preset dynamic evolution database.

[0149] S812: Generate variable features based on the current CT image and target features through a preset dynamic evolution model.

[0150] S813. Based on the associated clinical information, modify the variable features to generate dynamically evolving features.

[0151] Because different types of interstitial lung diseases have different pathological evolution patterns, there are multiple preset dynamic evolution models, which will be stored in a preset dynamic evolution database. Therefore, first, based on the current classification results, the corresponding preset dynamic evolution model will be matched from the preset dynamic evolution database, and then based on the current CT image and target features, variable features will be generated through the preset dynamic evolution model.

[0152] In addition to the possible changes in the current target features, the variable features here may also include newly added target features, that is, the variable features are equivalent to new lesion features.

[0153] Considering that the patient's clinical information will also affect the changes in the symptoms, after obtaining the variable features, the variable features will be corrected based on the associated clinical information. For example, the elasticity of lung tissue will be different at different age stages, and the corresponding pathological features such as honeycomb and reticular changes will also have different trends and amplitudes. Therefore, after obtaining the variable features, the rationality of the variable features can be judged with the help of associated clinical information, and unreasonable variable features can be corrected. The variable features obtained after correction are dynamically evolving features.

[0154] Through the dynamic evolution model, after new lung CT images are generated for the current lung CT images and classification results, the new lung CT images will be added to the preset database as a subsequent expanded training data set. When the subsequent expanded data set reaches a certain amount, iterative training can be performed based on the current preset classification model to achieve model optimization.

[0155] Specifically, after adding the new CT images to the preset training database, the following steps are also included:

[0156] S910: Determine whether the amount of newly added CT image data in the current preset training database reaches a preset update threshold.

[0157] S920: If yes, iteratively update the preset classification model based on the current preset training database to generate a new preset classification model.

[0158] When new lung CT images are stored in the preset training database, it will be determined whether the amount of new CT image data in the current preset training database reaches the preset update threshold. If it reaches the preset update threshold, the preset training database can be used as an expanded training data set to iteratively update the preset classification model to generate a new preset classification model, that is, a further optimized classification model. In this way, the model can be continuously improved to enhance the generalization ability of the classification model.

[0159] The present application also provides an interstitial lung disease classification system based on multi-dimensional feature labels, see Figure 7 The system includes: a data acquisition module 101, a region segmentation module 102, a feature detection module 103, a multi-dimensional feature fusion module 104, and a classification result generation module.

[0160] The data acquisition module 101 is used to acquire lung CT imaging information, which includes CT images and associated clinical information.

[0161] The region segmentation module 102 is used to perform region segmentation on the CT image using a preset three-dimensional model to obtain three-dimensional spatial information of each region.

[0162] The feature detection module 103 is used to perform target feature detection on the CT image based on the three-dimensional spatial information of each region to obtain the three-dimensional morphological features of the target features.

[0163] The multidimensional feature fusion module 104 is used to perform deep feature extraction and mathematical calculation on the target features, obtain high-order mathematical features, fuse the three-dimensional morphological features and the high-order mathematical features, generate high-dimensional fused image features, encode the associated clinical information, and combine the high-dimensional fused image features to generate multidimensional feature labels.

[0164] The classification result generating module 105 is configured to generate a classification result of interstitial lung disease based on the multi-dimensional feature labels and through a preset classification model.

[0165] In the embodiment of the present application, the data acquisition module 101 is specifically used to acquire lung CT image information, which includes CT images and associated clinical information, and perform corresponding preprocessing on the CT images.

[0166] The region segmentation module 102 is specifically used to perform region segmentation on the CT image obtained by the data acquisition module 101 through a preset three-dimensional model to obtain three-dimensional spatial information of the left lung, right lung, left upper lobe, left lower lobe, right upper lobe, right middle lobe, and right lower lobe regions.

[0167] The feature detection module 103 is specifically used to perform target feature detection on the CT image based on the three-dimensional spatial information of each region obtained by the region segmentation module 102, and obtain the target feature, that is, the three-dimensional morphological feature of the lesion feature including morphological feature information and spatial proportion information.

[0168] The multidimensional feature fusion module 104 is specifically used to perform deep feature extraction and mathematical calculations on the target features obtained by the feature detection module 103, obtain high-order mathematical features, fuse three-dimensional morphological features and high-order mathematical features, generate high-dimensional fused image features, encode related clinical information, and combine high-dimensional fused image features to generate multidimensional feature labels.

[0169] The classification result generating module 105 is specifically configured to generate a classification result of interstitial lung disease according to the multidimensional feature labels acquired by the multidimensional feature fusion module 104 through a preset classification model.

[0170] An embodiment of the present application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and execute any of the above-mentioned methods for classifying interstitial lung diseases based on multidimensional feature labels.

[0171] The embodiments of this specific implementation method are all preferred embodiments of the present application and are not intended to limit the scope of protection of the present application. Therefore, all equivalent changes made based on the principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for classifying interstitial lung diseases based on multidimensional feature labels, characterized in that: include: Acquiring lung CT imaging information, wherein the CT imaging information includes a CT image and associated clinical information; Performing regional segmentation on the CT image using a preset three-dimensional model to obtain three-dimensional spatial information of each region, wherein the regions are the left lung, right lung, left upper lobe, left lower lobe, right upper lobe, right middle lobe, and right lower lobe; Based on the three-dimensional spatial information of each region, target feature detection is performed on the CT image to obtain three-dimensional morphological features of the target features, wherein the target features are pathological features of interstitial lung disease; Perform deep feature extraction on target features to obtain deep features, perform mathematical calculations on the deep features, and fuse the mathematical calculation results with the deep features to obtain high-order mathematical features; Fuse three-dimensional morphological features and high-order mathematical features to generate high-dimensional fused image features; Encode the relevant clinical information and combine it with high-dimensional fusion image features to generate multidimensional feature labels; Based on multi-dimensional feature labels, the classification results of interstitial lung diseases are generated through a preset classification model; After generating the classification results of interstitial lung disease based on the multidimensional feature labels through a preset classification model, the method further includes: According to the classification results, the dynamic evolution characteristics are obtained by presetting the dynamic evolution model; Based on the dynamic evolution characteristics, the current CT image is reconstructed in three dimensions to generate a new CT image; Add the newly added CT images to the preset training database; The dynamic evolution characteristics are obtained by presetting a dynamic evolution model based on the classification results, including: According to the classification results, a corresponding preset dynamic evolution model is matched from a preset dynamic evolution database; Based on the current CT image and target features, variable features are generated through a preset dynamic evolution model; Based on the associated clinical information, the variable features are modified to generate dynamically evolving features.

2. The interstitial lung disease classification method based on multidimensional feature labels according to claim 1, characterized in that: The three-dimensional morphological features include morphological feature information and spatial proportion information. The target feature detection is performed on the CT image based on the three-dimensional spatial information of each region to obtain the three-dimensional morphological features of the target feature, including: Perform target feature detection on CT images to obtain morphological feature information and three-dimensional spatial information of the target features; Based on the three-dimensional spatial information of each area, match the spatial position information of the target feature in the associated area; Based on the three-dimensional spatial information of each region and target feature, the volume of each region and the volume of the target feature are calculated and obtained respectively; Based on the volume of each area and the volume of the target feature, the spatial proportion information of the target feature is calculated and obtained through the spatial position information of the target feature in the associated area.

3. The interstitial lung disease classification method based on multidimensional feature labels according to claim 1, characterized in that: The method of extracting deep features from target features to obtain deep features, performing mathematical calculations on the deep features, and fusing the mathematical calculation results with the deep features to obtain high-order mathematical features includes: Based on the target features, deep features are extracted through a preset deep neural network; Perform mathematical calculations on the deep features to obtain mathematical features; The deep features are combined with the mathematical features to generate high-order mathematical features.

4. The method for classifying interstitial lung diseases based on multidimensional feature labels according to claim 1, characterized in that: The fusing of the three-dimensional morphological features and the high-order mathematical features to generate high-dimensional fused image features includes: Based on three-dimensional morphological features and high-order mathematical features, correlation features are obtained through a preset correlation model; Generate fusion weights based on correlation features; Based on the fusion weights, the three-dimensional morphological features and high-order mathematical features are fused to generate high-dimensional fused image features.

5. The method for classifying interstitial lung diseases based on multidimensional feature labels according to claim 1, characterized in that: The associated clinical information includes extra-table information and intra-table information. The associated clinical information is encoded and combined with the high-dimensional fusion image features to generate a multi-dimensional feature label, including: Through the preset data analysis model, the influence coefficient of off-balance sheet information on various types of interstitial lung diseases is obtained, and the first influence vector is generated; Through the preset disease diagnosis library, the matching degree between the information in the table and various interstitial lung diseases is obtained, and the second influence vector is generated; encoding the clinical information based on the first influence vector and the second influence vector to generate a clinical feature label; Integrate clinical feature labels with high-dimensional fusion image features to generate multidimensional feature labels.

6. The method for classifying interstitial lung diseases based on multidimensional feature labels according to claim 1, characterized in that: After the newly added CT images are added to the preset training database, the method further includes: Determine whether the amount of newly added CT image data in the current preset training database reaches a preset update threshold; If so, the preset classification model is iteratively updated based on the current preset training database to generate a new preset classification model.

7. A classification system for interstitial lung diseases based on multidimensional feature labels, characterized by: include: A data acquisition module (101) is used to acquire lung CT imaging information, wherein the CT imaging information includes CT images and associated clinical information; A region segmentation module (102) is used to perform region segmentation on the CT image using a preset three-dimensional model to obtain three-dimensional spatial information of each region, wherein each region is the left lung, the right lung, and the left upper lobe, the left lower lobe, the right upper lobe, the right middle lobe, and the right lower lobe; A feature detection module (103) is used to perform target feature detection on the CT image based on the three-dimensional spatial information of each region, and obtain three-dimensional morphological features of the target features, wherein the target features are pathological features of interstitial lung disease; A multi-dimensional feature fusion module (104) is used to extract deep features from target features, obtain deep features, perform mathematical calculations on the deep features, fuse the mathematical calculation results with the deep features to obtain high-order mathematical features, fuse the three-dimensional morphological features and the high-order mathematical features to generate high-dimensional fused image features, encode associated clinical information, and generate multi-dimensional feature labels by combining the high-dimensional fused image features; The classification result generating module (105) is used to generate the classification result of interstitial lung disease based on the multidimensional feature labels and the preset classification model. After the classification result of interstitial lung disease is generated based on the multidimensional feature labels and the preset classification model, the module further includes: According to the classification results, the dynamic evolution characteristics are obtained by presetting the dynamic evolution model; Based on the dynamic evolution characteristics, the current CT image is reconstructed in three dimensions to generate a new CT image; Add the newly added CT images to the preset training database; The dynamic evolution characteristics are obtained by presetting a dynamic evolution model based on the classification results, including: According to the classification results, a corresponding preset dynamic evolution model is matched from a preset dynamic evolution database; Based on the current CT image and target features, variable features are generated through a preset dynamic evolution model; Based on the associated clinical information, the variable features are modified to generate dynamically evolving features.

8. A computer-readable storage medium storing a computer program capable of being loaded by a processor and executing the method for classifying interstitial lung diseases based on multidimensional feature labels as claimed in any one of claims 1 to 6.

Images