A Method for Extracting Color Doppler Ultrasound Radiomics Features and Its Application

Abstract

The present invention discloses a method for extracting color Doppler ultrasound imaging omics features and its application, belonging to the technical field of medical image processing, including: determining the tumor solid region image from within the color sampling frame of the color Doppler ultrasound image, and determining the vascular distribution density based on this as the blood flow distribution; after processing the tumor solid region image into RGB three-channel data, finding the obvious red pixel positions and blue pixel positions from the tumor solid region image, and determining the blood flow direction and blood flow velocity based on this, to obtain a processed tumor real-time region image containing the blood flow direction, blood flow velocity, and blood flow distribution; extracting color Doppler imaging omics features from the processed tumor real-time region image. The imaging omics features extracted in this way have more comprehensive and rich tissue structure information, and at the same time, the application of the color Doppler imaging omics features in clinical medical diagnosis is also provided, combining clinical data to improve the comprehensive ability of disease assessment.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and particularly relates to a method for extracting color Doppler ultrasound imaging omics features and its application. Background Art

[0002] Color Doppler ultrasound uses Doppler technology to describe the relative motion characteristics of tissues and body fluids (usually blood) with an ultrasound transducer. The imaging range of its signal represents the blood flow distribution in the region of interest. The red and blue colors represent the blood flow direction, and the brightness represents the flow velocity, so as to evaluate the characteristics of vascular blood flow, neovascularization of solid tumors, etc. Taking ovarian tumors as an example, according to the expert consensus definition of the International Ovarian Tumor Analysis (IOTA) of ultrasound, the description of the vascular richness (imaging range of the signal) is semi-quantitatively scored, which are no blood flow, small amount, medium amount, and large amount respectively, but to a certain extent, it has the subjectivity of ultrasound doctors. Spectral Doppler technology quantifies the blood flow velocity at the sampling point. However, due to factors such as the blood flow pattern in blood vessels (laminar flow, eddy current, turbulent flow, etc.) and the heterogeneity within the tumor, it is difficult for the blood flow velocity at the sampling point to reflect the overall flow velocity characteristics.

[0003] Imaging omics is a quantitative image analysis technology that uses computer image processing to extract medical image features in a high-throughput manner. A large number of papers emerging in recent years have proved that imaging omics can be applied to the early screening and accurate identification of tumors, and its diagnostic efficacy approaches the gold standard of pathological examination. However, at present, ultrasound imaging omics technology mainly focuses on gray-scale ultrasound images or single-color red coding of Doppler signals. For example, Chen Hui and Feng Weiwei from Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine encoded the blood flow signals in the lesion area as a single color red, and the rest of the lesion area was encoded as white. This single-color processing only shows the distribution of blood vessels and loses the blood flow direction and some flow velocity information. So far, there has been no imaging omics analysis method for extracting the signal features of the three colors and brightness of color Doppler. Summary of the Invention

[0004] In view of the above, the purpose of the present invention is to provide a method for extracting color Doppler ultrasound imaging omics features, aiming to quantitatively describe blood flow information in a high-throughput manner, including features such as blood flow distribution, direction, and flow velocity shown by color Doppler ultrasound, and to extract imaging omics features based on these features describing blood flow, and to give clinical applications based on the imaging omics features.

[0005] To achieve the above-mentioned invention purpose, an embodiment provides a method for extracting color Doppler ultrasound imaging omics features, including the following steps:

[0006] After standardizing the color Doppler ultrasound image, determine the image of the solid tumor area from within the color sampling frame, and divide the color area within the color sampling frame by the image of the solid tumor area to obtain the vascular distribution density as the blood flow distribution;

[0007] After processing the image of the solid tumor area into RGB three-channel data, crop the image of the solid tumor area, and then find the positions of obvious red pixels and blue pixels from the cropped image. Based on the positions of the blue pixels and red pixels, determine the blood flow direction and blood flow velocity, and obtain the processed real-time tumor area image containing the blood flow direction, blood flow velocity, and blood flow distribution;

[0008] Extract the radiomics features from the processed real-time tumor area image, and calculate the average layer features of the corresponding radiomics features of multiple frames of images as the candidate radiomics features. Then, perform standardization processing and feature screening on the candidate radiomics features to obtain the color Doppler radiomics features.

[0009] Preferably, cropping the image of the solid tumor area includes:

[0010] According to the position rule of the color scale in the image of the solid tumor area, crop the image to ensure that the image contains the color scale and the flow velocity text, and extract the flow velocity text from the cropped image, where the flow velocity text represents the maximum blood flow velocity.

[0011] Preferably, determining the blood flow direction and blood flow velocity based on the positions of the blue pixels and red pixels includes:

[0012] Extract and quantify the red boundary information and blue boundary information of the color scale based on the positions of the blue pixels and red pixels. These boundary information represent the RGB values of the pixel points corresponding to the maximum blood flow velocity in the red direction and blue direction of each image. Based on this RGB value and the maximum blood flow velocity, calculate the relative blood flow velocity of each pixel point of the blood flow signal in the solid tumor area to obtain the normalized blood flow velocity;

[0013] Among them, the blood flow direction corresponding to the red direction is the direction towards the probe, and the blood flow direction corresponding to the blue direction is the direction away from the probe.

[0014] Preferably, the method further includes: processing the color Doppler ultrasound image into RGB three-channel data, and simultaneously extracting the radiomics features from the processed real-time tumor area image and the RGB three-channel color Doppler ultrasound image to participate in the calculation of the subsequent average layer features.

[0015] Preferably, performing standardization processing on the candidate radiomics features includes:

[0016] Using the Z-score normalization method, after subtracting the feature mean from each candidate radiomics feature, divide by the standard deviation of the feature to obtain the normalization result.

[0017] Preferably, feature screening includes:

[0018] Use the univariate logistic regression method to analyze the correlation between each candidate radiomics feature corresponding to the normalized color Doppler image and the disease label, retain the candidate radiomics features with a correlation less than 0.1, and then use the maximum correlation and minimum redundancy feature selection method to rank the importance of the candidate radiomics features, and determine the color Doppler radiomics features based on the ranking.

[0019] An embodiment of the present invention also provides a target classification device based on color Doppler ultrasound radiomics features, including:

[0020] The first color Doppler radiomics feature module is used to extract color Doppler radiomics features based on the above color Doppler ultrasound radiomics feature extraction method;

[0021] The first grayscale ultrasound radiomics feature module is used to extract grayscale ultrasound radiomics features from grayscale ultrasound images;

[0022] The target classification module is used to perform target classification by separately combining color Doppler radiomics features, or simultaneously fusing color Doppler radiomics features and grayscale ultrasound radiomics features to obtain a target classification result, and the target classification result includes ovarian borderline or malignant tumors.

[0023] An embodiment of the present invention also provides a serous ovarian malignant tumor grade classification device based on color Doppler ultrasound radiomics features, including:

[0024] The second color Doppler radiomics feature module is used to extract color Doppler radiomics features based on the above color Doppler ultrasound radiomics feature extraction method;

[0025] The second grayscale ultrasound radiomics feature module is used to extract grayscale ultrasound radiomics features from grayscale ultrasound images;

[0026] The grade classification module is used to perform serous ovarian malignant tumor grade classification by separately combining color Doppler radiomics features, or simultaneously fusing color Doppler radiomics features and grayscale ultrasound radiomics features to obtain a grade classification result, and the grade classification result includes high grade or non-high grade.

[0027] An embodiment of the present invention also provides an ovarian malignant tumor clinical stage prediction device based on color Doppler ultrasound radiomics features, including:

[0028] The third color Doppler imaging omics feature module, which is used to extract color Doppler imaging omics features based on the above-mentioned color Doppler ultrasound imaging omics feature extraction method;

[0029] The third gray-scale ultrasound imaging omics feature module, which is used to extract gray-scale ultrasound imaging omics features from gray-scale ultrasound images;

[0030] The stage classification module, which is used to predict the clinical stage of ovarian malignant tumors by separately combining color Doppler imaging omics features, or simultaneously fusing color Doppler imaging omics features and gray-scale ultrasound imaging omics features, to obtain a stage prediction result, and the stage prediction result includes I-IV.

[0031] Compared with the prior art, the beneficial effects of the present invention at least include:

[0032] The present invention extracts RGB three-channel information from the tumor solid region image of the color Doppler ultrasound image to analyze the color Doppler blood flow distribution, blood flow direction, and blood flow velocity characteristics, and extracts color Doppler imaging omics features based on these features. The imaging omics features extracted in this way have more comprehensive and rich organizational structure information, and through the verification of the application of imaging omics features in clinical practice, it is shown that these imaging omics features can be more accurately used in tumor diseases such as the differential diagnosis of tumor benign and malignant, neovascular evaluation, pathological typing, heterogeneity, and clinical staging. Combining clinical data can improve the comprehensive ability of disease assessment. Description of the Drawings

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0034] Figure 1 It is a flowchart of the color Doppler ultrasound imaging omics feature extraction method provided by the embodiment;

[0035] Figure 2 It is a structural schematic diagram of the target classification device provided by the embodiment;

[0036] Figure 3 It is a working flowchart of the target classification device provided by the embodiment;

[0037] Figure 4 It is a structural schematic diagram of the serous ovarian malignant tumor stage classification device provided by the embodiment;

[0038] Figure 5It is a schematic structural diagram of the ovarian malignant tumor clinical staging prediction device provided by the embodiment. Detailed implementation manners

[0039] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific implementation manners described herein are only used to explain the present invention and do not limit the protection scope of the present invention.

[0040] The inventive concept of the present invention is as follows: Aiming at the technical problem that the existing radiomics features mainly come from the analysis of grayscale images and monochromatic-encoded color Doppler images, lacking important information on the blood flow velocity and direction of tumor blood vessels, the present invention provides a method for extracting color Doppler ultrasound radiomics features, which extracts blood flow distribution, blood flow direction, and blood flow velocity information from the color Doppler images of the RGB three channels, and based on this, extracts radiomics features. The radiomics features extracted in this way contain more comprehensive and accurate tissue structure information, and also perform clinical verification through the radiomics features provided by the present invention to illustrate that the extracted radiomics features can be used for classification, recognition, and evaluation results in clinical applications.

[0041] As Figure 1 shown, the method for extracting color Doppler ultrasound radiomics features provided by the embodiment includes the following steps:

[0042] S1. After performing standardization processing on the color Doppler ultrasound image, determine the tumor solid region image from within the color sampling frame, and divide the color region within the color sampling frame by the tumor solid region image to obtain the blood vessel distribution density as the blood flow distribution.

[0043] In the embodiment, a color Doppler ultrasound image is obtained, and each frame of the color Doppler ultrasound image in the image is subjected to standardization processing, specifically including size adjustment, etc., as the basic image for subsequent feature extraction, and is stored in the DICOM (Digital Imaging and Communications in Medicine) format. Then, a color sampling frame in the standardized color Doppler ultrasound image is determined, and the tumor solid region is determined as the region of interest (ROI region) based on the color sampling frame. The tumor solid region is the intratumoral papilla, cyst wall, septum, etc., and the tumor parenchymal region is saved, and other regions are set to blank, obtaining a tumor parenchymal region image with the same size as the initial color Doppler ultrasound image, and storing it as an Nrrd (Nearly Raw Raster Data) file. The color region within the color sampling frame is divided by the tumor solid region image to obtain the vascular distribution density as the blood flow distribution. For multiple color Doppler ultrasound images, the vascular distribution density of each image is calculated, and the average is calculated and recorded as the average vascular density value to exclude the influence of the cardiac cycle on blood flow display.

[0044] S2. After processing the tumor solid region image into RGB three-channel data, the tumor solid region image is cropped, and the obvious red pixel positions and blue pixel positions are found from the cropped image. Based on the blue pixel positions and red pixel positions, the blood flow direction and blood flow velocity are determined, obtaining a processed tumor real-time region image containing the blood flow direction, blood flow velocity, and blood flow distribution.

[0045] In the embodiment, Matlab can be used to unify the DICOM file of the color Doppler image and the Nrrd file of the tumor parenchymal region image into R / G / B two-dimensional three-channel data, and the pixel values of each channel are sampled and normalized to 1 to 64. Then, the tumor solid region image is cropped. Specifically, according to the position rule of the color scale bar in the tumor solid region image, the image is cropped to ensure that the image contains the color scale bar and the flow velocity text, and an OCR tool is used to extract the flow velocity text in the cropped image. The flow velocity text represents the maximum blood flow velocity and is used for the calculation of the flow velocity in other regions later.

[0046] In the embodiment, the obvious red pixel positions and blue pixel positions are found from the cropped image. For example, the upper limit of the red pixel is set as [25, 255, 255], and the lower limit is set as [0, 30, 30]; the upper limit of the blue pixel is set as [140, 255, 255], and the lower limit is set as [100, 50, 50]. Since there are certain impurities in the figure, such as flow velocity values and images outside the color scale bar, the extraction of the largest connected region is performed. Since the shape of the largest connected region is irregular, the smaller linear structures with small areas are removed, and the structures as close to rectangles as possible are retained.

[0047] Then, based on the blue pixel positions and red pixel positions, the blood flow direction and blood flow velocity are determined. Specifically, the red boundary information and blue boundary information of the color scale are extracted and quantified according to the blue pixel positions and red pixel positions. These boundary information represent the RGB values of the pixel points corresponding to the maximum blood flow velocities in the red direction and blue direction of each image. Based on this RGB value and the maximum blood flow velocity, the relative blood flow velocity of each pixel point of the blood flow signal in the solid region of the tumor is calculated to obtain the normalized blood flow velocity. At the same time, it is determined that the blood flow direction corresponding to the red direction is the direction towards the probe, and the blood flow direction corresponding to the blue direction is the direction away from the probe. Based on this, the blood flow direction is determined, where the positive sign '+' represents red and the negative sign '-' represents blue, and then a processed real-time tumor region image including the blood flow direction, blood flow velocity, and blood flow distribution is obtained.

[0048] S3. Extract radiomics features from the processed real-time tumor region image and calculate the average layer features of the corresponding radiomics features of multiple frames of images as candidate radiomics features, and then perform standardization processing and feature screening on the candidate radiomics features to obtain color Doppler radiomics features.

[0049] In the embodiment, open-source Matlab can be used to extract radiomics features from the processed real-time tumor region image, including shape features, first-order statistical features, texture features, and high-order features. For wavelet radiomics features, considering all possible combinations of applying high-pass filters (H) or low-pass filters (L) on the x, y, and z axes, eight types of wavelet features can be generated, including: '-LLL', '-LLH', '-LHL', '-LHH', '-HLL', '-HLH', '-HHL', and '-HHH' features. The texture features include gray level cooccurence matrix (GLCM), graylevel run length matrix (GLRLM), gray level size zone matrix (GLSZM), neighbouring gray tone difference matrix (NGTDM), and gray level dependence matrix (GLDM) features.

[0050] Then, the average values of the corresponding radiomics features of multiple color Doppler ultrasound images in the color Doppler ultrasound image are calculated to obtain the average layer features as candidate radiomics features to exclude the influence of the cardiac cycle on blood flow display.

[0051] In the embodiment, the candidate radiomics features are also subjected to standardization processing and feature screening to obtain color Doppler radiomics features. Specifically, using the Z-score standardization method, the candidate radiomics features are respectively subtracted by the feature mean value and then divided by the standard deviation value of the features to obtain the standardization processing result. Then, when performing feature screening, first, the univariate logistic regression method is used to analyze the correlation between the candidate radiomics features corresponding to each standardized color Doppler image and the disease label, and the candidate radiomics features with a correlation less than 0.1 are retained. Then, the feature selection method of maximum relevance and minimum redundancy (mRMR) is used to rank the importance of the candidate radiomics features, and the appropriate color Doppler radiomics features are determined based on the ranking. Among them, the maximum correlation constraint is to select the feature with the largest mutual information value between a single feature and the label. The minimum redundancy constraint is to remove the redundant features that highly depend on other features.

[0052] In the embodiment, a disease prediction model based on color Doppler ultrasound radiomics features is also constructed, and the ROC (Receiver Operating Characteristic Curve) and AUC (Area Under Curve) are used to evaluate the model performance, and the corresponding sensitivity, specificity, PPV (positive predictive value), NPV (negative predictive value), and accuracy are calculated to verify the model efficacy. At the same time, the gray-scale ultrasound radiomics features combined with gray-scale images are used to improve the ultrasound diagnosis ability from two dimensions of ultrasound gray-scale and color.

[0053] Next, taking ovarian cancer as an example, it is proved that the above-mentioned color Doppler ultrasound radiomics feature extraction method can be applied to, including but not limited to: differentiating between borderline and malignant ovarian tumors, differentiating between high-grade and non-high-grade serous ovarian malignant tumors, and predicting the clinical stage of ovarian malignant tumors.

[0054] For the task of differentiating between borderline and malignant ovarian tumors, accurate diagnosis of borderline ovarian tumors is crucial for optimizing the surgical plan for fertility preservation. Therefore, currently, intraoperative frozen section analysis (FSA) is initially recommended for borderline ovarian tumors, but the accuracy is only 44.8 - 81.0%. Therefore, as Figure 2 and Figure 3As shown in the figure, an object classification device 20 based on color Doppler ultrasound imaging omics features is constructed in an embodiment of the present invention, including: a first color Doppler imaging omics feature module 31, a first grayscale ultrasound imaging omics feature module 22, and an object classification module 23.

[0055] Among them, the first color Doppler imaging omics feature module 21 is used to extract color Doppler imaging omics features by the above color Doppler ultrasound imaging omics feature extraction method. For the specific extraction process, please refer to the above extraction method.

[0056] The first grayscale ultrasound imaging omics feature module 22 is used to extract grayscale ultrasound imaging omics features from grayscale ultrasound images. Specifically, extracting grayscale ultrasound imaging omics features from grayscale ultrasound images includes shape features, first-order statistical features, texture features, and high-order features, and the extraction method is not limited.

[0057] The object classification module 33 is used to perform object classification by separately combining grayscale ultrasound imaging omics features, separately combining color Doppler imaging omics features, or simultaneously fusing color Doppler imaging omics features and grayscale ultrasound imaging omics features to obtain an object classification result, and the object classification result includes ovarian borderline tumors or malignant tumors. Specifically, in the object classification module 23, a grayscale ultrasound imaging omics model (Gray-scale Radomics, GS1), a color Doppler imaging omics model (Color-Doppler Radomics, CD1), and a fusion model (Gray-scale Color-Doppler Radiomics, GSCD1) are respectively constructed and compared for their diagnostic performance in differentiating ovarian borderline and malignant tumors. When fusing color Doppler imaging omics features and grayscale ultrasound imaging omics features, the color Doppler imaging omics features and grayscale ultrasound imaging omics features are combined and then features with the same number as the color Doppler imaging omics features are selected from them and input into the fusion model for object classification to ensure that the feature input sizes of the three models are consistent. Each model uses the random forest algorithm.

[0058] In the dataset, the AUCs of the three models GS1, CD1, and GSCD1 for classifying ovarian borderline tumors and malignant tumors are 0.825, 0.899, and 0.908 respectively. It can be seen that compared with GS1, the performance of CD1 and GSCD1 is enhanced, and the comparison between the two models GSCD1 and GS1 PThe value is 0.005. The smaller this value is, the more obvious the statistical difference between the two models is. Therefore, the difference between the two models is statistically significant. Moreover, the diagnostic accuracy of GSCD1 (90.8%) is better than that of the current clinical preoperative FSA for diagnosing borderline ovarian tumors (44.8% - 81%), indicating that color Doppler blood flow signals can effectively improve the diagnostic performance of the model (Table 1).

[0059] Table 1 Diagnostic performance of GS1, CD1, and GSCD1

[0060]

[0061] The number of cases included in the model is 209. The values in parentheses for each model index are their corresponding 95% confidence intervals, which are used to estimate the interval range of the corresponding population parameters. For example, AUC 0.825(0.767, 0.874) means that when repeatedly sampling from the same population data and calculating the interval, approximately 95% of the intervals will contain the true population AUC value.

[0062] For the task of differentiating high-grade and non-high-grade serous ovarian malignancies, the precise treatment of ovarian cancer is closely related to the histological subtypes of tumors. Currently, in clinical practice, puncture biopsy and pathological examination are used to clarify the histological type of ovarian epithelial malignancies before surgery. However, domestic and foreign guidelines indicate that fine needle puncture is prohibited for early-stage and cystic lesions to avoid artificially increasing the clinical stage of the tumor. Therefore, in the embodiments of the present invention, preoperative ultrasound image analysis is used, such as Figure 4 shown, to construct a serous ovarian malignancy grade classification device 40 based on color Doppler ultrasound radiomics features, including: a second color Doppler radiomics feature module 41, a second grayscale ultrasound radiomics feature module 42, and a grade classification module 43.

[0063] Among them, the second color Doppler radiomics feature module 41 is used to extract color Doppler radiomics features based on the above color Doppler ultrasound radiomics feature extraction method; the second grayscale ultrasound radiomics feature module 42 is used to extract grayscale ultrasound radiomics features from grayscale ultrasound images. For the specific process, please refer to the above color Doppler ultrasound radiomics feature extraction method, which will not be elaborated here.

[0064] The level classification module 43 is used to classify the grade of serous ovarian malignancies by separately combining gray-scale ultrasound imaging omics features, separately combining color Doppler imaging omics features, or simultaneously fusing color Doppler imaging omics features and gray-scale ultrasound imaging omics features, to obtain a level classification result, which includes high-grade or non-high-grade. Specifically, in the level classification module 43, three models GS2, CD2, and GSCD2 are also constructed to evaluate the diagnostic performance in differentiating high-grade from non-high-grade serous ovarian malignancies. The AUC values of the three models in the dataset are 0.760, 0.792, and 0.861 respectively. Thus, compared with the GS2 model, the performance of GSCD2 is significantly enhanced ( P <0.0001), and it maintains a stable and high sensitivity, as shown in Table 2.

[0065] Table 2 Diagnostic performance of GS2, CD2, and GSCD2

[0066]

[0067] The number of cases included in the model is 428. The values in parentheses for each model index are their corresponding 95% confidence intervals.

[0068] For the task of predicting the clinical stage of ovarian malignancies, the commonly used ADNEX model has a certain predictive value for the FIGO clinical stage of ovarian malignancies (International Federation of Gynecology and Obstetrics), but the predictive accuracy of borderline ovarian tumors and FIGO stage I is relatively low (AUC 0.75). Therefore, as Figure 5 shown, an embodiment of the present invention constructs a clinical stage prediction device 50 for ovarian malignancies based on color Doppler ultrasound imaging omics features, including: a third color Doppler imaging omics feature module 51, a third gray-scale ultrasound imaging omics feature module 52, and a staging module 53.

[0069] Among them, the third color Doppler imaging omics feature module 51 is used to extract color Doppler imaging omics features based on the above-mentioned color Doppler ultrasound imaging omics feature extraction method; the third gray-scale ultrasound imaging omics feature module 52 is used to extract gray-scale ultrasound imaging omics features from gray-scale ultrasound images. The specific process is detailed in the above-mentioned color Doppler ultrasound imaging omics feature extraction method and will not be elaborated here.

[0070] The staging module 53 is used to predict the clinical stage of ovarian malignant tumors by separately combining gray-scale ultrasound imaging features, separately combining color Doppler imaging features, or simultaneously fusing color Doppler imaging features and gray-scale ultrasound imaging features, and obtain the staging prediction result. Specifically, in the staging module 53, three models GS3, CD3, and GSCD3 are also constructed to identify the diagnostic performance of borderline ovarian tumors and FIGO stage I epithelial malignant tumors, and FIGO stage II-IV epithelial malignant tumors. In the task of differentiating borderline ovarian tumors and FIGO stage I epithelial malignant tumors, the AUC values of the three models are 0.803, 0.857, and 0.871 respectively, as shown in Table 3.

[0071] Table 3 Diagnostic efficacy of GS3, CD3, and GSCD3

[0072]

[0073] The number of cases included in the model is 106, and the values in parentheses of each model index are their corresponding 95% confidence intervals.

[0074] In the task of differentiating borderline ovarian tumors and FIGO stage II-IV epithelial malignant tumors, the AUC values of the three models are 0.853, 0.918, and 0.924 respectively, as shown in Table 4.

[0075] Table 4 Diagnostic efficacy of GS4, CD4, and GSCD4

[0076]

[0077] The number of cases included in the model is 170, and the values in parentheses of each model index are their corresponding 95% confidence intervals.

[0078] GSCD has the discriminative value for predicting the differentiation between borderline ovarian tumors and FIGO clinical stages. Especially in the task of differentiating borderline ovarian tumors and FIGO stage I epithelial malignant tumors, the AUC value is higher than that of the ADNEX model, which is commonly used for the staging of ovarian malignant tumors.

[0079] After the above clinical application comparison, it is verified that the fusion model GSCD that fuses gray-scale ultrasound images and color Doppler images has a significant improvement in the ability to predict and differentially diagnose the pathological nature of preoperative ovarian masses compared with the model GS constructed only based on gray-scale images, indicating the feasibility and practicality of the color Doppler ultrasound imaging features of the present invention.

[0080] The specific embodiments described above have elaborated on the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements, equivalent replacements, etc. made within the scope of the principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for extracting color Doppler ultrasound imaging features, characterized in that: The following steps are involved: After the color Doppler ultrasound image is standardized, the solid area image of the tumor is determined from the color sampling frame, and the color area in the color sampling frame is divided by the solid area image of the tumor to obtain the vascular distribution density as the blood flow distribution; After processing the solid area image of the tumor into RGB three-channel data, the solid area image of the tumor is cropped, and then obvious red pixel positions and blue pixel positions are found from the cropped image, and the blood flow direction and blood flow velocity are determined based on the blue pixel position and the red pixel position, so as to obtain a processed real-time area image of the tumor including the blood flow direction, blood flow velocity, and blood flow distribution; Radiomic features were extracted from the real-time regional images of the processed tumors, and the average layer features of the radiomic features corresponding to the multi-frame images were calculated as candidate radiomic features. The candidate radiomic features were then standardized and screened to obtain color Doppler radiomic features.

2. The color Doppler ultrasound imaging feature extraction method according to claim 1, characterized in that: Crop the solid area image of the tumor, including: In the solid area image of the tumor, according to the position rule of the color scale, the image is cropped to ensure that the image contains the color scale and the flow rate text, and the flow rate text in the cropped image is extracted, wherein the flow rate text represents the maximum blood flow velocity.

3. The color Doppler ultrasound imaging feature extraction method according to claim 2, characterized in that: The blood flow direction and blood flow velocity are determined based on the blue pixel position and the red pixel position, including: The red boundary information and blue boundary information of the color scale are extracted and quantified according to the blue pixel position and the red pixel position. These boundary information represent the RGB values ​​of the pixels corresponding to the maximum blood flow velocity in the red direction and the blue direction of each image. The relative blood flow velocity of each pixel of the blood flow signal in the solid area of ​​the tumor is calculated based on the RGB value and the maximum blood flow velocity to obtain the normalized blood flow velocity. The red direction corresponds to the blood flow direction towards the probe, and the blue reverse direction corresponds to the blood flow direction away from the probe.

4. The color Doppler ultrasound imaging feature extraction method according to claim 2, characterized in that: Also includes: The color Doppler ultrasound image is processed into RGB three-channel data, and the imaging omics features are simultaneously extracted from the processed real-time tumor regional image and the RGB three-channel color Doppler ultrasound image to participate in the subsequent calculation of the average layer features.

5. The color Doppler ultrasound imaging feature extraction method according to claim 2, characterized in that: Standardization of candidate radiomics features including: Using the Z-score standardization method, the candidate imaging features were subtracted from the feature mean and then divided by the standard deviation of the feature to obtain the standardized processing results.

6. The method for extracting color Doppler ultrasound imaging features according to claim 2, characterized in that: Feature screening, including: The univariate logistic regression method was used to analyze the correlation between the candidate radiomics features and disease labels corresponding to each standardized color Doppler image. The candidate radiomics features with correlation less than 0.1 were retained. The maximum correlation and minimum redundancy feature selection method was then used to rank the importance of the candidate radiomics features, and the color Doppler radiomics features were determined based on the ranking.

7. A target classification device based on color Doppler ultrasound imaging features, characterized in that: include: A first color Doppler imaging omics feature module, which is used to extract color Doppler imaging omics features based on the color Doppler ultrasound imaging omics feature extraction method according to any one of claims 1 to 6; A first grayscale ultrasound imaging omics feature module, which is used to extract grayscale ultrasound imaging omics features from grayscale ultrasound images; The target classification module is used to classify the target by combining the color Doppler imaging genomics features alone or by fusing the color Doppler imaging genomics features and the grayscale ultrasound imaging genomics features simultaneously to obtain the target classification results, which include ovarian borderline tumors or malignant tumors.

8. A device for classifying the grade of serous ovarian malignant tumors based on color Doppler ultrasound imaging features, characterized in that: include: A second color Doppler imaging omics feature module, which is used to extract color Doppler imaging omics features based on the color Doppler ultrasound imaging omics feature extraction method according to any one of claims 1 to 6; A second grayscale ultrasound imaging omics feature module, which is used to extract grayscale ultrasound imaging omics features from grayscale ultrasound images; The grade classification module is used to classify the grade of serous ovarian malignant tumors by combining color Doppler imaging genomics features alone, or by fusing color Doppler imaging genomics features and grayscale ultrasound imaging genomics features simultaneously, to obtain a grade classification result, which includes high grade or non-high grade.

9. A device for predicting clinical staging of ovarian malignant tumors based on color Doppler ultrasound imaging features, characterized in that: include: A third color Doppler imaging omics feature module, which is used to extract color Doppler imaging omics features based on the color Doppler ultrasound imaging omics feature extraction method according to any one of claims 1 to 6; A third grayscale ultrasound imaging omics feature module is used to extract grayscale ultrasound imaging omics features from grayscale ultrasound images; The staging module is used to predict the clinical staging of ovarian malignant tumors by combining color Doppler imaging genomics features alone or by simultaneously fusing color Doppler imaging genomics features and grayscale ultrasound imaging genomics features to obtain staging prediction results, which include I-IV.

Images