Method, device and system for constructing radiomics scoring model and method, device and system for classifying radiomics scoring model

Abstract

The invention discloses a radiomics scoring model construction method, a scoring method, a scoring device and a scoring system, and the method comprises the steps: carrying out the preprocessing of a CT image after the CT image is obtained, and obtaining a processed image; performing feature extraction on the processed image to obtain intensity feature data, morphological feature data and textural feature data, and performing scoring processing on the processed image to obtain a scoring image; and performing model training by adopting the scoring image and the feature data to obtain a radiomics scoring model. According to the method, model training can be carried out in combination with the texture, strength and morphological characteristics of the image, the model precision is effectively improved, and the image group of the patient can be graded by the model subsequently for reference by a doctor to help a clinician to make a decision.

Description

Background Art

[0002] Microvessel density (MVD) is an established parameter for quantifying tumor angiogenesis and is a candidate biomarker for accurately identifying gastric cancer patients at high risk of recurrence and metastasis. However, MVD is currently only obtained through postoperative pathological examination and cannot be used as a reference before treatment. In order to accurately evaluate MVD, radiomics can be used to convert patients' medical images into mineable high-dimensional data, and data processing of high-dimensional image data can be performed to determine the patient's MVD.

[0003] In the existing field of image processing, the commonly used method is: image labeling and recognition. First, the image of the patient's gastrointestinal tumor is collected through computed tomography (CT), and then the doctor manually identifies the texture of the image and labels the image. Then, the model is trained based on the texture of the image to obtain a scoring model to determine the patient's microvascular density according to the scoring model.

[0004] However, the commonly used methods currently have the following technical problems: the shape and color of the pathological area and the surrounding area in the image are similar. After using the texture training model, the recognition and scoring processing accuracy is low and prone to deviations, which makes it difficult to meet existing medical recognition needs. Summary of the invention

[0005] The present invention proposes a method for constructing an imaging omics scoring model, a scoring method, a device and a system. The method constructs an imaging omics scoring model by marking the intensity features, morphological features and grayscale texture features of an image to improve the accuracy of the model in identifying and scoring lesions.

[0006] A first aspect of an embodiment of the present invention provides a method for constructing a radiomics scoring model, the method comprising:

[0007] After acquiring the CT image, preprocessing the CT image to obtain a processed image;

[0008] Extracting features from the processed image to obtain feature data, and performing scoring processing on the processed image to obtain a scoring image, wherein the feature data includes intensity feature data, morphological feature data, and texture feature data, and the scoring processing is to score the microvessels of the pathological tissue in the processed image;

[0009] The scoring image and the feature data are used to perform model training to obtain a radiomics scoring model.

[0010] In a possible implementation manner of the first aspect, extracting features from the processed image to obtain feature data includes:

[0011] After converting the processed image into a grayscale image, calculating the average brightness value and histogram statistical information of each pixel of the grayscale image to obtain intensity feature data;

[0012] After the processed image is converted into an image in a binary format and subjected to morphological processing to obtain a morphological image, shape information of the morphological image is calculated to obtain morphological feature data;

[0013] The preset co-occurrence matrix is ​​called to perform texture analysis on the processed image to obtain texture feature data.

[0014] In a possible implementation manner of the first aspect, the performing scoring processing on the processed image to obtain a scoring image includes:

[0015] After displaying the processed image, a number of initial score values ​​are obtained, where the initial score value is the average score added by each pathologist after viewing microvessels at different positions in the gastric cancer tissue of the processed image;

[0016] According to the prognosis information of the patient corresponding to each processed image, a plurality of initial scoring values ​​corresponding to each processed image are used to calculate a microvessel density cutoff value, and the microvessel density cutoff value is added to the processed image to obtain a scoring image.

[0017] In a possible implementation manner of the first aspect, preprocessing the CT image to obtain a processed image includes:

[0018] Performing imaging processing on the CT image to obtain a portal venous phase image;

[0019] An image of a lesion region is extracted from the portal venous phase image to obtain a processed image, wherein the lesion region is a region with the largest cross-sectional area of ​​a gastric cancer lesion in the portal venous phase image.

[0020] In a possible implementation manner of the first aspect, the using the scoring image and the feature data to perform model training to obtain a radiomics scoring model includes:

[0021] Using the scoring image and the feature data to perform model training on a plurality of initial models to obtain a plurality of basic models;

[0022] The voting method based on ensemble learning integrates the several basic models to obtain the imaging omics scoring model.

[0023] In a possible implementation manner of the first aspect, after the step of using the scoring image and the feature data to perform model training to obtain a radiomics scoring model, the method further includes:

[0024] Calculating the sensitivity and specificity values ​​of the radiomics scoring model;

[0025] If the sensitivity value and the specificity value are respectively less than corresponding thresholds, the scoring image and the feature data are used to perform model optimization on the imaging omics scoring model.

[0026] A second aspect of an embodiment of the present invention provides an image group classification method for microvascular density, the method comprising:

[0027] Acquiring a plurality of image sets, each of the image sets comprising a CT image of microvascular density of a patient;

[0028] Calling the radiomics scoring model constructed by the radiomics scoring model construction method as described above to score each of the image groups to obtain a radiomics score cutoff value;

[0029] The imaging groups are classified according to the size of the imaging omics score cutoff value.

[0030] A third aspect of an embodiment of the present invention provides a device for constructing a radiomics scoring model, the device comprising:

[0031] An image processing module, used for preprocessing the CT image after acquiring the CT image to obtain a processed image;

[0032] An extraction and scoring module, used for performing feature extraction on the processed image to obtain feature data, and performing scoring processing on the processed image to obtain a scoring image, wherein the feature data includes intensity feature data, morphological feature data and texture feature data, and the scoring processing is to score the microvessels of the pathological tissue in the processed image;

[0033] A model module is constructed to use the scoring image and the feature data to perform model training to obtain a radiomics scoring model.

[0034] In a possible implementation manner of the third aspect, extracting features from the processed image to obtain feature data includes:

[0035] After converting the processed image into a grayscale image, calculating the average brightness value and histogram statistical information of each pixel of the grayscale image to obtain intensity feature data;

[0036] After the processed image is converted into an image in a binary format and subjected to morphological processing to obtain a morphological image, shape information of the morphological image is calculated to obtain morphological feature data;

[0037] The preset co-occurrence matrix is ​​called to perform texture analysis on the processed image to obtain texture feature data.

[0038] In a possible implementation manner of the third aspect, the performing scoring processing on the processed image to obtain a scoring image includes:

[0039] After displaying the processed image, a number of initial score values ​​are obtained, where the initial score value is the average score added by each pathologist after viewing microvessels at different positions in the gastric cancer tissue of the processed image;

[0040] According to the prognosis information of the patient corresponding to each processed image, a plurality of initial scoring values ​​corresponding to each processed image are used to calculate a microvessel density cutoff value, and the microvessel density cutoff value is added to the processed image to obtain a scoring image.

[0041] In a possible implementation manner of the third aspect, preprocessing the CT image to obtain a processed image includes:

[0042] Performing imaging processing on the CT image to obtain a portal venous phase image;

[0043] An image of a lesion region is extracted from the portal venous phase image to obtain a processed image, wherein the lesion region is a region with the largest cross-sectional area of ​​a gastric cancer lesion in the portal venous phase image.

[0044] In a possible implementation manner of the third aspect, the using the scoring image and the feature data to perform model training to obtain a radiomics scoring model includes:

[0045] Using the scoring image and the feature data to perform model training on a plurality of initial models to obtain a plurality of basic models;

[0046] The voting method based on ensemble learning integrates the several basic models to obtain the imaging omics scoring model.

[0047] In a possible implementation manner of the third aspect, the device further includes:

[0048] A parameter calculation module, used for calculating the sensitivity value and specificity value of the radiomics scoring model after the step of using the scoring image and the feature data to perform model training to obtain the radiomics scoring model;

[0049] An optimization module is used to optimize the imaging omics scoring model using the scoring image and the feature data if the sensitivity value and the specificity value are respectively less than corresponding thresholds.

[0050] A fourth aspect of the embodiments of the present invention provides an image group classification device for microvessel density, the device comprising:

[0051] An acquisition module, used for acquiring a plurality of image groups, each of which includes a CT image of microvascular density of a patient;

[0052] A scoring module, used for calling the radiomics scoring model constructed by the radiomics scoring model construction method as described above to score each of the image groups and obtain a radiomics scoring cutoff value;

[0053] A classification module is used to classify the imaging group according to the size of the imaging omics score cutoff value.

[0054] A fifth aspect of the embodiments of the present invention provides a microvessel density assessment system, the system comprising: a medical image processing module, an artificial intelligence data analysis module, and a result determination module;

[0055] The medical image processing module is used to extract radiomics features from the input enhanced CT image using radiomics methods;

[0056] The artificial intelligence data analysis module screens radiomics features, builds models, and calculates scores;

[0057] The result determination module is used to make a determination based on the score to obtain an evaluation result of the microvessel density.

[0058] Compared with the prior art, the embodiments of the present invention provide a method for constructing a radiomics scoring model, a scoring method, a device and a system, which have the following beneficial effects: the present invention can preprocess the acquired CT image to obtain a processed image; extract intensity feature data, morphological feature data and texture feature data from the processed image, and score the processed image to obtain a scored image; use the scored image and feature data to perform model training to obtain a radiomics scoring model; the present invention can combine the texture, intensity and morphological features of the image for model training, which can effectively improve the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flowchart of a method for constructing an imaging omics scoring model provided by one embodiment of the present invention;

[0060] Figure 2 is a graph of operating characteristic curve results in a training group provided by an embodiment of the present invention;

[0061] Figure 3 It is a graph of operating characteristic curve results in a validation group provided by an embodiment of the present invention;

[0062] Figure 4 is a graph of the working characteristic curve results of an external verification group provided by an embodiment of the present invention;

[0063] Figure 5It is a flowchart of a method for classifying image groups of microvessel density provided by an embodiment of the present invention;

[0064] Figure 6 is an operation flow chart of an image group classification method for microvessel density provided by an embodiment of the present invention;

[0065] Figure 7 It is a structural schematic diagram of a device for constructing an imaging omics scoring model provided by an embodiment of the present invention;

[0066] Figure 8 It is a structural schematic diagram of an image group classification device for microvessel density provided by an embodiment of the present invention;

[0067] Fig. 9 It is a structural schematic diagram of a microvessel density evaluation system provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0068] 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 only 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.

[0069] Microvessel density (MVD) is an established parameter for quantifying tumor angiogenesis and is a candidate biomarker for accurately identifying gastric cancer patients at high risk of recurrence and metastasis. However, MVD is currently only obtained through postoperative pathological examination and cannot be used as a reference before treatment. In order to accurately evaluate MVD, radiomics can be used to convert patients' medical images into mineable high-dimensional data, and data processing of high-dimensional image data can be performed to determine the patient's MVD.

[0070] In the existing field of image processing, the commonly used method is: image labeling and recognition. First, the image of the patient's gastrointestinal tumor is collected through computed tomography (CT), and then the doctor manually identifies the texture of the image and labels the image. Then, the model is trained based on the texture of the image to obtain a scoring model to determine the patient's microvascular density according to the scoring model.

[0071] However, the commonly used methods currently have the following technical problems: the shape and color of the pathological area and the surrounding area in the image are similar. After using the texture training model, the recognition and scoring processing accuracy is low and prone to deviations, which makes it difficult to meet existing medical recognition needs.

[0072] In order to solve the above problems, the following specific embodiments will be used to introduce and explain in detail a method for constructing an imaging omics scoring model, a scoring method, a device and a system provided in the embodiments of the present application.

[0073] Reference Figure 1 , showing a flow chart of a method for constructing an imaging omics scoring model provided by an embodiment of the present invention.

[0074] In one embodiment, the method is applicable to a computer system, which may also be a smart terminal.

[0075] As an example, the method for constructing the radiomics scoring model may include:

[0076] S11. After acquiring the CT image, preprocess the CT image to obtain a processed image.

[0077] In one operation mode, the CT image may be an enhanced CT image of the abdomen of a gastric cancer patient. Then, the CT image may be preprocessed to improve the clarity of the image.

[0078] The patient's pathological area and normal area may be included. In subsequent training, the normal area may affect the recognition accuracy. In order to avoid the above situation, as an example, the preprocessing of the CT image to obtain the processed image may include the following sub-steps:

[0079] S111, performing imaging processing on the CT image to obtain a portal venous phase image.

[0080] S112, extracting an image of a lesion region from the portal venous phase image to obtain a processed image, wherein the lesion region is a region with the largest cross-sectional area of ​​a gastric cancer lesion in the portal venous phase image.

[0081] Specifically, the input CT image can be processed to obtain a portal venous phase image, which can be in a dicom format, and then image preprocessing software (eg, ITK-SNAP (v3.4)) is called to preprocess the portal venous phase image, including clarity adjustment.

[0082] Next, the processed image can be shown to the medical staff, who can view the processed portal venous phase image. The medical staff then selects the region with the largest cross-sectional area of ​​the gastric cancer lesion (all gastric cancer lesion regions, ROI, Region of Interest) in the processed portal venous phase image, intercepts the portal venous phase image according to the region selected by the medical staff, and obtains a processed image (referred to as ROI image).

[0083] Optionally, medical personnel may select a region of interest in the processed portal venous phase image, and then intercept the portal venous phase image according to the region to obtain a processed image.

[0084] It should be noted that, in order to facilitate subsequent processing, the output format of the processed image can be .nii.

[0085] S12, performing feature extraction on the processed image to obtain feature data, and performing scoring processing on the processed image to obtain a scoring image, wherein the feature data includes intensity feature data, morphological feature data and texture feature data, and the scoring processing is to score the microvessels of the pathological tissue in the processed image.

[0086] After obtaining the processed image, feature extraction can be performed on the processed image to obtain feature data. At the same time, scoring processing can be performed on the processed image to obtain a scoring image.

[0087] The extracted feature data includes intensity feature data, morphological feature data and texture feature data. The intensity feature data may be various numerical values ​​of the image, such as intensity value, grayscale value, maximum value, etc. The morphological feature data may be various data of the pathological morphology of the image, such as size, area, length, width, height, etc. The texture feature data may be the color, arrangement, distribution position, etc. of the texture of the image.

[0088] In an optional embodiment, there are multiple extracted feature data. In order to perform corresponding extraction processing on different feature data, as an example, the feature extraction of the processed image to obtain the feature data may include the following sub-steps:

[0089] S21, after converting the processed image into a grayscale image, calculating the average brightness value and histogram statistical information of each pixel of the grayscale image to obtain intensity feature data.

[0090] S22, after converting the processed image into a binary format image and performing morphological processing to obtain a morphological image, shape information of the morphological image is calculated to obtain morphological feature data.

[0091] S23, calling a preset co-occurrence matrix to perform texture analysis on the processed image to obtain texture feature data.

[0092] In one embodiment, the computer software Matlab may be called to perform feature extraction on the processed image, and the extraction may be performed using the automatic extraction function image processing tools and algorithms of the computer software Matlab.

[0093] In an actual operation mode, 292 feature data can be extracted from the processed images of each patient, including three types of features: intensity feature data, morphological feature data, and grayscale texture feature data.

[0094] In a specific operation mode, after reading the processed image, the processed image can be converted into a grayscale image. The intensity feature data can be obtained by calculating the average brightness value or histogram statistical information of each pixel of the grayscale image. Specifically, the intensity feature data can be used to describe the grayscale value of the tumor area of ​​the enhanced CT image, including the average value, maximum value, minimum value, etc.

[0095] In a specific operation mode, the processed image can be binarized, and the binarized processed image can be morphologically processed to obtain a morphological image. The morphological processing can include corrosion, expansion, opening operation, closing operation and other processing. The morphological feature data can be obtained by calculating the shape information such as the target contour, area, perimeter, aspect ratio and the like of the morphological image. Specifically, the shape feature is used to describe the size, shape and other information of the gastric cancer tumor lesion.

[0096] In a specific operation mode, texture analysis can be performed on the processed image, for example, using a gray level co-occurrence matrix (GLCM) or a gray level difference co-occurrence matrix (GLDM) for texture analysis. Then, texture features such as contrast, angle, energy, entropy, etc. can be extracted to describe the texture features of the image. Specifically, the grayscale texture features are used to describe the spatial arrangement information of the color or intensity of gastric cancer lesions on the CT image.

[0097] It should be noted that the extracted feature data can be stored in a feature vector. The above process is repeated for each patient's image to extract 292 feature data. The final feature vector includes intensity, morphology and texture features, which can be used for tasks such as training and prediction of machine learning models.

[0098] In one embodiment, in order to accurately score the processed image to determine whether the image has a disease risk, as an example, the scoring process of the processed image to obtain a scoring image may include the following sub-steps:

[0099] S31. After displaying the processed image, a number of initial score values ​​are obtained. The initial score value is an average score added by each pathologist after viewing microvessels at different positions in the gastric cancer tissue of the processed image.

[0100] S32, according to the prognosis information of the patient corresponding to each processed image, using a number of initial scoring values ​​corresponding to each processed image to calculate a microvessel density cutoff value, and adding the microvessel density cutoff value to the processed image to obtain a scoring image.

[0101] Specifically, the processed images may be shown to medical staff, who may then be asked to rate the processed images.

[0102] For example, the processed images can be shown to two pathologists. The two pathologists can use CD34 as a marker for vascular endothelium, mark the microvessels in gastric cancer tissue by immunohistochemistry, and observe the pathological sections using an upright microscope. Four positions distributed in different directions and the middle position are selected, and the staining results of each area are evaluated at a magnification of 200 times, and then each position is scored. Correspondingly, each doctor can add 5 initial scoring values, and the two doctors can get 10 initial scoring values.

[0103] It should be noted that a single CD34-positive cell or discrete cluster was considered as one microvessel. The average number of microvessels at five locations was defined as MVD.

[0104] After obtaining several initial scoring values, the relationship between the image and the patient's prognosis can be obtained, and the X-tile software can be called to determine the optimal cutoff score of MVD based on the relationship between the image and the patient's prognosis.

[0105] Specifically, several initial scoring values ​​and corresponding prognostic information corresponding to the processed images of each gastric cancer patient can be collected first. The collected data needs to ensure that the data is input in a suitable format, including the microvascular score and the patient's prognostic information. Then, the input several initial scoring values ​​and corresponding prognostic information can be processed, and the processing can include deleting abnormal values, processing missing data, etc. to ensure data integrity and accuracy. Then, the input data can be grouped, and a grouping strategy can be set to group patients according to several initial scoring values ​​and prognostic information. Then, the data of each group is statistically analyzed, and specifically, KM survival analysis can be performed on the patients in each group to determine the relationship between microvascular density and prognostic information.

[0106] After the analysis is completed, the cutoff value can be calculated. Specifically, the X-tile software can be called to calculate the optimal microvascular density cutoff value based on the results of the statistical analysis. This cutoff value can help distinguish individuals with different prognoses in the patient population. The optimal microvascular density cutoff value is used as the score value for the processed image, and it is added to the image to obtain a scored image.

[0107] In actual operation, X-tile software can find the optimal cutoff value of MVD based on the exhaustive method, and divide all data into two groups through this optimal cutoff value. At this time, KM survival analysis is performed on the two groups, and the P value obtained is the smallest.

[0108] In an optional embodiment, after calculating the optimal microvessel density cutoff value, the result of the microvessel density cutoff value can be displayed, and a corresponding chart or report can also be generated. The above method can help doctors determine the optimal microvessel density cutoff value based on the microvessel score and prognostic information of each gastric cancer patient, so as to help predict the patient's survival and provide more scientific guidance for clinical treatment.

[0109] S13, using the scoring image and the feature data to perform model training to obtain a radiomics scoring model.

[0110] In one embodiment, the scoring images and feature data may be divided into a training set and a validation set, and the radiomics scoring model may be trained based on the training set.

[0111] In one embodiment, in order to further improve the accuracy of the model, as an example, the model training using the scoring image and the feature data to obtain the radiomics scoring model may include the following sub-steps:

[0112] S131. Use the scoring image and the feature data to perform model training on several initial models to obtain several basic models.

[0113] S132. Integrate the several basic models by a voting method based on ensemble learning to obtain an imaging omics scoring model.

[0114] Specifically, four initial models including support vector regression (SVR), k-nearest neighbor algorithm (KNN), decision tree, and random forest can be first trained using a training set of scoring images and feature data to obtain four basic models.

[0115] Then, the four base models were integrated using the voting method of ensemble learning to obtain the imaging omics scoring model.

[0116] In specific operations, each initial model can be trained and optimized using a training data set, including tuning the parameters of each initial model to achieve the best performance on the training data.

[0117] Next, each basic model can be used to predict the test data set and obtain the prediction results of each model. After obtaining the prediction results of each basic model, the voting method can be used to combine these results. That is, the category label with the most votes is selected as the final prediction result. The voting method is an ensemble learning model that follows the principle of majority rule. It reduces variance by integrating multiple models, thereby improving the robustness of the model. It should be noted that the training of the initial model and the ensemble learning model is based on the python machine learning library scikit-learn.

[0118] Through the above method, the texture, intensity and morphological features of the image can be combined for model training, which can effectively improve the accuracy of the model.

[0119] In an optional embodiment, after the model is trained, the accuracy of the model needs to be verified to determine whether the model meets the requirements, wherein, as an example, the method may further include the following steps:

[0120] S14. Calculate the sensitivity value and specificity value of the imaging omics scoring model.

[0121] S15. If the sensitivity value and the specificity value are respectively less than corresponding thresholds, optimizing the imaging omics scoring model using the scoring image and the feature data.

[0122] Specifically, the radiomics scoring model can be verified based on the validation set. The verification method can draw a receiver operating characteristic (ROC) curve based on the validation set, calculate the area under the curve (AUC), and evaluate the model performance accuracy based on the size of the area.

[0123] Reference Figure 2-4 , respectively showing a working characteristic curve result diagram of a training group provided by an embodiment of the present invention, a working characteristic curve result diagram of a validation group provided by an embodiment of the present invention, and a working characteristic curve result diagram of an external verification group provided by an embodiment of the present invention.

[0124] Specifically, the roc_curve function of the python machine learning library scikit-learn is called to draw the receiver operating characteristic curve (ROC), and then calculate the area under the curve (AUC). The operating characteristic curve is a comprehensive indicator that reflects the sensitivity and specificity of continuous variables. It uses the composition method to reveal the relationship between sensitivity and specificity. It calculates a series of sensitivity and specificity values ​​by setting multiple different critical values ​​for continuous variables. It should be noted that the ROC curve is a curve drawn based on a series of different binary classification methods (cutoff values ​​or decision thresholds), with the true positive rate (TPR, that is, sensitivity) as the vertical coordinate and the false positive rate (FPR, 1-specificity) as the horizontal coordinate.

[0125] For different ROC curves, the greater the curvature / curvature, that is, the farther away from the 45-degree diagonal of the accidental event (the diagonal represents the subject's discrimination is 0), the better the model performance. After the operating characteristic curve is drawn, the area under the curve (AUC) can be calculated using the sklearn.metrics.roc_curve function. The area under the curve is defined as the area under the ROC curve and the coordinate axis, so the value range of AUC is between 0.5 and 1. The closer the AUC is to 1.0, the higher the accuracy of the model prediction; when it is equal to 0.5, the prediction accuracy is the lowest and there is no application value.

[0126] In this embodiment, an embodiment of the present invention provides a method for constructing an imaging omics scoring model, and its beneficial effects are: the present invention can preprocess the acquired CT image to obtain a processed image; extract intensity feature data, morphological feature data and texture feature data from the processed image, and score the processed image to obtain a scored image; use the scored image and feature data to perform model training to obtain an imaging omics scoring model; the present invention can combine the texture, intensity and morphological features of the image for model training, which can effectively improve the accuracy of the model.

[0127] Reference Figure 5 , showing a flow chart of an image group classification method for microvessel density provided by an embodiment of the present invention.

[0128] Wherein, as an example, the image group classification method for microvascular density may include:

[0129] S41, acquiring a plurality of image groups, each of which includes a CT image of microvascular density of a patient.

[0130] S42, calling the radiomics scoring model constructed by the radiomics scoring model construction method as described above to score each of the image groups to obtain a radiomics score cutoff value.

[0131] S43, classifying the imaging groups according to the size of the imaging omics score cutoff value.

[0132] In specific operations, the microvascular density grade and the corresponding imaging group of gastric cancer patients can be obtained, and then the microvascular density grade and the corresponding imaging group score can be matched to obtain the score cutoff value of the imaging group, which can be called at the same time.

[0133] Specifically, in Python, you can use the roc_curve function of the scikit-learn library to draw the ROC curve, and the sklearn.metrics.roc_curve function calculates the Youden index and finds the optimal ROC curve threshold. Using the Youden index to find the optimal ROC curve threshold is a common method to determine the optimal threshold of the classifier to achieve the best classification effect.

[0134] Specifically, we can first draw the ROC curve of the training model. The ROC curve is a curve with the true positive rate (TPR) as the vertical axis and the false positive rate (FPR) as the horizontal axis, which is used to evaluate the performance of the classifier. Then calculate the Youden index. The definition of the Youden index is: Youden index = TPR-FPR, that is, the true positive rate minus the false positive rate. The Youden index can measure the performance of the classifier under different thresholds. Find the point with the largest Youden index on the ROC curve, and the corresponding horizontal axis is the optimal threshold. The point corresponding to the maximum Youden index indicates that the difference between the true positive rate and the false positive rate is the largest, that is, the classification effect is the best.

[0135] Finally, the imaging groups were separated according to the cutoff value of the imaging omics score to determine the grading of vascular density corresponding to the imaging group.

[0136] Specifically, the imaging group's imaging omics score value can be obtained. If the imaging group's imaging omics score value is less than the imaging group's imaging omics score cutoff value, the imaging group is a low microvessel density group; if the imaging group's imaging omics score value is greater than or equal to the imaging group's imaging omics score cutoff value, the imaging group is a high microvessel density group.

[0137] Reference Figure 6 , shows an operational flow chart of an image group classification method for microvessel density provided by an embodiment of the present invention.

[0138] The first step is to enhance CT image acquisition and processing.

[0139] The second step is imaging feature extraction.

[0140] The third step is to obtain the microvascular score through pathological testing of the patient's gastric cancer tissue.

[0141] The fourth step is to construct a microvascular scoring model for predicting gastric cancer based on radiomics characteristics.

[0142] Step 5: Result determination and output.

[0143] For further explanation, in an actual operation mode, the operation flow of the above method may include the following steps:

[0144] In the first step, 34x gastric cancer patients who underwent gastrectomy between x month 20xx and x month 20xx were included as the training group, and 13x gastric cancer patients who underwent surgery between x month 20xx and x month 20xx were selected as the internal validation group. 10x gastric cancer patients who underwent surgery from other tumor centers from x month 20xx to x month 20xx were selected as the external validation group. The preoperative enhanced CT images of these patients were selected. The external validation group was used to verify whether the model was reliable. The input enhanced CT images were processed into images; specifically, the input enhanced CT images were processed into images to obtain portal venous phase images (in dicom format), and the obtained patient portal venous phase images were imported into the computer software ITK-SNAP (v3.4) for image preprocessing. Experienced radiologists selected the portal venous phase images with the largest cross-sectional area of ​​gastric cancer lesions and outlined the region of interest on them. The output format was .nii.

[0145] In the second step, the enhanced CT images of the abdomen of gastric cancer patients were delineated after the first step, and the image features of the region of interest were extracted. 292 feature data were extracted from each patient's CT image, including three types of features: intensity feature data, morphological feature data, and grayscale texture feature data. The intensity feature data is used to reflect the image intensity information of gastric cancer tumor lesions on CT images; the morphological feature data is used to reflect the morphological information of gastric cancer tumor lesions; the grayscale texture feature data is used to reflect the voxel spatial distribution intensity level information of gastric cancer lesions on CT images and present the feature information of the surface and internal tissues corresponding to gastric cancer lesions.

[0146] In the third step, the microvascular score was obtained by pathological examination of the patient's gastric cancer tissue, and the gastric cancer patients were divided into two groups according to the prognosis: high microvascular density group and low microvascular density group. Gastric cancer tissue specimens, clinical information and prognostic information were collected. CD34 was used as a marker of vascular endothelium, and the microvessels in gastric cancer tissue were marked by immunohistochemistry. Two pathologists evaluated the microvascular score, and the microvascular score of each gastric cancer patient was combined with the prognostic information for analysis using MedCalc software to obtain the microvascular density cutoff value. Gastric cancer patients with a microvascular score greater than the cutoff value were in the high microvascular density group, and conversely, gastric cancer patients with a microvascular score less than the cutoff value were in the low microvascular density group.

[0147] In the fourth step, based on the training group, a model for evaluating the microvascular density of gastric cancer based on imaging omics features was trained. The imaging omics features obtained in the second step were used as input, and the microvascular density grouping obtained in the third step was used as output. Four base models including support vector regression, k-nearest neighbor algorithm, decision tree, and random forest were trained, and the four base models were further integrated using the voting method of ensemble learning. The base models were fitted using the entire data set. The prediction results of each base model were averaged by voting to generate the final prediction results. The voting method is an ensemble learning model that follows the principle of minority obeys majority. It reduces variance by integrating multiple models, thereby improving the robustness of the model. Subsequently, the model was verified in the validation group and the external validation group. The receiver operating characteristic curves of the model for evaluating the microvascular density of gastric cancer based on imaging omics features in the training group, validation group, and external validation group are shown in ( Figure 2-Figure 4 ), the areas under the curve of the training group, validation group and external validation group were 0.925, 0.893 and 0.874, respectively.

[0148] Step 5: Result determination and output. The microvascular density of each gastric cancer patient is matched with the corresponding radiomics score value, and the matching results are analyzed and a cutoff value is generated. Specifically, according to the numerical value of the gastric cancer radiomics score, qualitative prediction analysis is performed on the individuals with specific image information. The specific qualitative analysis process is as follows: using MedCalc V12.7 software, the actual microvascular density of each patient in the training group (high microvascular density group and low microvascular density group) is matched with the corresponding specific radiomics score value. The cutoff value is calculated by MedCalc software to be 0.51. When the patient's Rad-score ≥ 0.51, the qualitative patient is in the high microvascular density group. When the patient's Rad-score < 0.51, the qualitative patient is in the low microvascular density group, providing reliable prediction analysis results before surgery.

[0149] In this embodiment, an embodiment of the present invention provides a method for constructing a radiomics scoring model, and its beneficial effects are: after obtaining the patient's image group, the present invention can call the radiomics scoring model to score each image group to obtain the radiomics score cutoff value; classify the image groups according to the size of the radiomics score cutoff value to improve the accuracy of recognition and classification, for doctors' reference, which can help clinicians make decisions and choose more reasonable treatment plans, so that patients benefit from survival, reduce treatment toxicity and side effects, and reduce social and economic burdens.

[0150] The present invention also provides a device for constructing a radiomics scoring model, see Figure 7 , showing a schematic structural diagram of a device for constructing an imaging omics scoring model provided by an embodiment of the present invention.

[0151] Wherein, as an example, the apparatus for constructing the radiomics scoring model may include:

[0152] An image processing module 201 is used to pre-process the CT image after acquiring the CT image to obtain a processed image;

[0153] An extraction and scoring module 202 is used to extract features from the processed image to obtain feature data, and to perform scoring processing on the processed image to obtain a scoring image, wherein the feature data includes intensity feature data, morphological feature data and texture feature data, and the scoring processing is to score the microvessels of the pathological tissue in the processed image;

[0154] The model building module 203 is used to use the scoring image and the feature data to perform model training to obtain a radiomics scoring model.

[0155] Optionally, extracting features from the processed image to obtain feature data includes:

[0156] After converting the processed image into a grayscale image, calculating the average brightness value and histogram statistical information of each pixel of the grayscale image to obtain intensity feature data;

[0157] After the processed image is converted into an image in a binary format and subjected to morphological processing to obtain a morphological image, shape information of the morphological image is calculated to obtain morphological feature data;

[0158] The preset co-occurrence matrix is ​​called to perform texture analysis on the processed image to obtain texture feature data.

[0159] Optionally, performing scoring processing on the processed image to obtain a scoring image includes:

[0160] After displaying the processed image, a number of initial score values ​​are obtained, where the initial score value is the average score added by each pathologist after viewing microvessels at different positions in the gastric cancer tissue of the processed image;

[0161] According to the prognosis information of the patient corresponding to each processed image, a plurality of initial scoring values ​​corresponding to each processed image are used to calculate a microvessel density cutoff value, and the microvessel density cutoff value is added to the processed image to obtain a scoring image.

[0162] Optionally, preprocessing the CT image to obtain a processed image includes:

[0163] Performing imaging processing on the CT image to obtain a portal venous phase image;

[0164] An image of a lesion region is extracted from the portal venous phase image to obtain a processed image, wherein the lesion region is a region with the largest cross-sectional area of ​​a gastric cancer lesion in the portal venous phase image.

[0165] Optionally, the using the scoring image and the feature data to perform model training to obtain a radiomics scoring model includes:

[0166] Using the scoring image and the feature data to perform model training on a plurality of initial models to obtain a plurality of basic models;

[0167] The voting method based on ensemble learning integrates the several basic models to obtain the imaging omics scoring model.

[0168] Optionally, the device further comprises:

[0169] A parameter calculation module, used for calculating the sensitivity value and specificity value of the radiomics scoring model after the step of using the scoring image and the feature data to perform model training to obtain the radiomics scoring model;

[0170] An optimization module is used to optimize the imaging omics scoring model using the scoring image and the feature data if the sensitivity value and the specificity value are respectively less than corresponding thresholds.

[0171] The present invention also provides an image group classification device for microvascular density, see Figure 8 , showing a schematic structural diagram of an image group classification device for microvessel density provided by an embodiment of the present invention.

[0172] Wherein, as an example, the image group classification device for microvascular density may include:

[0173] An acquisition module 301 is used to acquire a plurality of image groups, each of which includes a CT image of microvascular density of a patient;

[0174] A scoring module 302 is used to call the radiomics scoring model constructed by the radiomics scoring model construction method as described above to score each of the image groups to obtain a radiomics scoring cutoff value;

[0175] The classification module 303 is used to classify the image group according to the size of the image group score cutoff value.

[0176] The present invention also provides a microvessel density evaluation system. Fig. 9 , showing a schematic structural diagram of a microvessel density assessment system provided by an embodiment of the present invention.

[0177] Wherein, as an example, the microvessel density assessment system may include: a medical image processing module, an artificial intelligence data analysis module, and a result determination module;

[0178] The medical image processing module is used to extract radiomics features from the input enhanced CT image using radiomics methods;

[0179] The artificial intelligence data analysis module screens radiomics features, builds models, and calculates scores;

[0180] The result determination module is used to make a determination based on the score to obtain an evaluation result of the microvessel density.

[0181] Specifically, the medical image processing module uses radiomics methods to extract radiomics features from the input enhanced CT image, and transmits the radiomics feature data to the artificial intelligence data analysis module; the artificial intelligence data analysis module screens the radiomics features, builds a model, calculates the score, and transmits the score value to the result judgment module; the result judgment module judges the obtained score and draws a conclusion.

[0182] The medical image processing module includes an enhanced CT image input submodule, a CT image preprocessing submodule and an imaging genomics feature extraction submodule. The enhanced CT image input and preprocessing submodule is used to input enhanced CT images of gastric cancer patients before surgery; the CT image preprocessing submodule is used for CT image standardization processing and region of interest (ROI) delineation; the imaging genomics feature extraction submodule is used to extract three groups of feature data groups from the identified image features, wherein the first group of feature data groups includes a number of intensity feature data, the second group of feature data groups includes a number of morphological feature data, and the third group of feature data groups includes a number of grayscale texture feature data; the intensity feature data is used to reflect the image intensity information of gastric cancer tumor lesions on CT images; the morphological feature data is used to reflect the morphological information of gastric cancer tumor lesions; the grayscale texture feature data is used to reflect the voxel spatial distribution intensity level information of gastric cancer lesions on CT images and present the feature information of the surface and interior of the corresponding tissue of the gastric cancer lesions;

[0183] The artificial intelligence data analysis module includes a model building submodule and a result output module. The model building submodule divides the enrolled patients into a training set and a validation set, and trains a model for evaluating gastric cancer microvascular density based on imaging genomics features based on the training set. Specifically, four base models including support vector regression (SVR), k-nearest neighbor algorithm (KNN), decision tree, and random forest are first trained, and the four base models are further integrated using the voting method of ensemble learning. The base model is fitted using the entire data set. The prediction results of each base model are averaged by voting to generate the final prediction result. The voting method is an ensemble learning model that follows the principle of minority obeys the majority. It reduces variance by integrating multiple models, thereby improving the robustness of the model. Subsequently, a model for evaluating gastric cancer microvascular density based on imaging genomics features is trained based on the training set and verified in the validation set. The receiver operating characteristic curve (ROC) was drawn, and the area under the curve (AUC) was calculated to evaluate the accuracy of the model performance.

[0184] The result output module inputs the imaging features of each patient into the above-mentioned model for evaluating gastric cancer microvessel density based on imaging features to obtain the imaging score of each patient.

[0185] The result determination module includes a microvessel density grading submodule, an imaging genomics score-microvessel density grading matching submodule, a cutoff value determination submodule, and a result output submodule. The microvessel density grading submodule obtains a microvessel score by pathologically detecting the patient's gastric cancer tissue, and divides the gastric cancer patients into two categories: a high microvessel density group and a low microvessel density group in combination with the prognosis; the imaging genomics score-microvessel density grading matching submodule matches the microvessel density and clinical data of gastric cancer patients with the corresponding imaging genomics score, and transmits the matching result to the cutoff value determination submodule; the cutoff determination submodule analyzes the matching data and generates a cutoff value, which is used as a judgment threshold for the probability of peritoneal metastasis in the patient; the probability classification submodule is used to compare the score value of each patient with the cutoff value, and use the comparison result as a judgment result of high or low probability of disease.

[0186] Furthermore, the four base models of support vector regression, k-nearest neighbor algorithm, decision tree, and random forest and the integrated learning model are completed through the sklearn package of the computer programming tool python.

[0187] Furthermore, the patient score matching submodule calls the MedCalc medical statistical analysis software to match each patient's microvascular density grade with the corresponding imaging genomics score value, and transmits the matching result to the cutoff value acquisition submodule; the cutoff value acquisition submodule calls the MedCalc medical statistical analysis software to analyze the matching result and generate a cutoff value, and the cutoff value serves as a judgment threshold for the probability of the patient being judged as a high microvascular density group.

[0188] Furthermore, when the patient's imaging omics score was < the cutoff value, the patient was classified into the low microvessel density group; when the patient's imaging omics score was ≥ the cutoff value, the patient was classified into the high microvessel density group.

[0189] Those skilled in the art can clearly understand that, for the sake of convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0190] Furthermore, an embodiment of the present application also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for constructing an imaging genomics scoring model or the imaging group classification method for microvascular density as described in the above embodiments is implemented.

[0191] Furthermore, an embodiment of the present application also provides a computer-readable storage medium, which stores a computer-executable program, and the computer-executable program is used to enable a computer to execute the method for constructing an imaging genomics scoring model or the imaging group classification method for microvascular density as described in the above embodiments.

[0192] It should be noted that the orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the embodiments of the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. When an element such as a layer, region or substrate is referred to as being "on" or "above" another element, it can be directly on the other element, or there can also be an intermediate element. On the contrary, when an element is referred to as "directly on" or "above" another element, there is no intermediate element. It should also be understood that when an element is referred to as being "under" or "below" another element, it can be directly under or below the other element, or there can also be an intermediate element. On the contrary, when an element is referred to as being "directly under" or "below" another element, there is no intermediate element. Unless otherwise clearly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0193] Those skilled in the art will appreciate that the embodiments of the present application may also provide computer program products. Therefore, the present application may adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program codes.

[0194] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), apparatuses, and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0195] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0196] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0197] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for constructing a radiomics scoring model, characterized in that: The method comprises: After acquiring the CT image, preprocessing the CT image to obtain a processed image; Extracting features from the processed image to obtain feature data, and performing scoring processing on the processed image to obtain a scoring image, wherein the feature data includes intensity feature data, morphological feature data, and texture feature data, and the scoring processing is to score the microvessels of the pathological tissue in the processed image; The scoring image and the feature data are used to perform model training to obtain a radiomics scoring model.

2. The method for constructing a radiomics scoring model according to claim 1, characterized in that: The step of extracting features from the processed image to obtain feature data comprises: After converting the processed image into a grayscale image, calculating the average brightness value and histogram statistical information of each pixel of the grayscale image to obtain intensity feature data; After the processed image is converted into an image in a binary format and subjected to morphological processing to obtain a morphological image, shape information of the morphological image is calculated to obtain morphological feature data; The preset co-occurrence matrix is ​​called to perform texture analysis on the processed image to obtain texture feature data.

3. The method for constructing a radiomics scoring model according to claim 1, characterized in that: The step of performing scoring processing on the processed image to obtain a scoring image comprises: After displaying the processed image, a number of initial score values ​​are obtained, where the initial score value is the average score added by each pathologist after viewing microvessels at different positions in the gastric cancer tissue of the processed image; According to the prognosis information of the patient corresponding to each processed image, a plurality of initial scoring values ​​corresponding to each processed image are used to calculate a microvessel density cutoff value, and the microvessel density cutoff value is added to the processed image to obtain a scoring image.

4. The method for constructing a radiomics scoring model according to claim 1, characterized in that: The preprocessing of the CT image to obtain a processed image includes: Performing imaging processing on the CT image to obtain a portal venous phase image; An image of a lesion region is extracted from the portal venous phase image to obtain a processed image, wherein the lesion region is a region with the largest cross-sectional area of ​​a gastric cancer lesion in the portal venous phase image.

5. The method for constructing a radiomics scoring model according to claim 1, characterized in that: The step of using the scoring image and the feature data to perform model training to obtain a radiomics scoring model comprises: Using the scoring image and the feature data to perform model training on a plurality of initial models to obtain a plurality of basic models; The voting method based on ensemble learning integrates the several basic models to obtain the imaging omics scoring model.

6. The method for constructing a radiomics scoring model according to any one of claims 1 to 5, characterized in that: After the step of using the scoring image and the feature data to perform model training to obtain a radiomics scoring model, the method further includes: Calculating the sensitivity and specificity values ​​of the radiomics scoring model; If the sensitivity value and the specificity value are respectively less than corresponding thresholds, the scoring image and the feature data are used to perform model optimization on the imaging omics scoring model.

7. A method for classifying image groups based on microvessel density, characterized in that: The method comprises: Acquiring a plurality of image sets, each of the image sets comprising a CT image of microvascular density of a patient; Calling the radiomics scoring model constructed by the method for constructing a radiomics scoring model according to any one of claims 1 to 6 to score each of the image groups to obtain a radiomics score cutoff value; The imaging groups are classified according to the size of the imaging omics score cutoff value.

8. A device for constructing a radiomics scoring model, characterized in that: The device comprises: An image processing module, used for preprocessing the CT image after acquiring the CT image to obtain a processed image; An extraction and scoring module, used for performing feature extraction on the processed image to obtain feature data, and performing scoring processing on the processed image to obtain a scoring image, wherein the feature data includes intensity feature data, morphological feature data and texture feature data, and the scoring processing is to score the microvessels of the pathological tissue in the processed image; A model module is constructed to use the scoring image and the feature data to perform model training to obtain a radiomics scoring model.

9. An image group classification device for microvascular density, characterized in that: The device comprises: An acquisition module, used for acquiring a plurality of image groups, each of which includes a CT image of microvascular density of a patient; A scoring module, configured to call the radiomics scoring model constructed by the method for constructing a radiomics scoring model according to any one of claims 1 to 6 to score each of the image groups to obtain a radiomics scoring cutoff value; A classification module is used to classify the imaging group according to the size of the imaging omics score cutoff value.

10. A microvessel density assessment system, characterized in that: The system includes: a medical image processing module, an artificial intelligence data analysis module, and a result determination module; The medical image processing module is used to extract radiomics features from the input enhanced CT image using radiomics methods; The artificial intelligence data analysis module screens radiomics features, builds models, and calculates scores; The result determination module is used to make a determination based on the score to obtain an evaluation result of the microvessel density.

Images