Method and System for Quantifying Coronary Vascular Hemodynamics Based on IVUS Images

Abstract

The present invention relates to the technical field of medical image processing, and specifically relates to a method and system for quantifying coronary artery hemodynamics based on IVUS images, including the following steps: in the IVUS image, enhance the target respectively to obtain a snowflake target enhanced image and an interference target enhanced image; use the Mask R-CNN model to perform target detection on the enhanced images respectively to obtain a snowflake target mask image and an interference target mask image; process the IVUS image with the fusion mask of the snowflake target mask image and the interference target mask image to obtain a snowflake segmentation result image; quantify the hemodynamics based on the snowflake segmentation result image. The present invention uses a neural network model to accurately segment the snowflake region in the IVUS image, realizes the accurate extraction of the information of the hypoechoic region caused by blood flow in the intima reflected by the IVUS image, and then realizes the accurate quantitative evaluation of the hemodynamic parameters of the coronary artery.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and particularly to a method and system for quantifying coronary artery hemodynamics based on IVUS images. Background Art

[0002] In the field of cardiovascular medicine, the IVUS (Intravascular Ultrasound) technology has developed over several decades and has increasingly become the mainstream of precise treatment. Especially for the treatment of complex coronary artery lesions, such as left main trunk lesions, bifurcation lesions, calcified lesions, CTO (Chronic Total Occlusion), etc., it can fully evaluate the coronary artery lesions and has great guiding significance and optimization strategies for PCI (Percutaneous Coronary Intervention).

[0003] Currently, the clinical determination significance of IVUS images is only limited to the analysis of evaluating vascular structure, plaque characteristics, and postoperative evaluation of stent conditions, etc. There has been no characteristic analysis of the hypoechoic region, namely "snowflake points", caused by blood flow in the intima reflected by IVUS images. Since the images of this region are mainly caused by the emission and reception of blood flow at the front end of the ultrasound catheter, if this part of the data can be extracted and further analyzed, the hemodynamic parameters of the coronary artery can be evaluated, and thus important information can be provided for evaluating cardiovascular health, diagnosing coronary artery lesions, and formulating individualized treatment strategies.

[0004] Therefore, in the prior art, there is a lack of characteristic analysis of the hypoechoic region, namely "snowflake points", caused by blood flow in the intima reflected by IVUS images, making it difficult to quantitatively evaluate the hemodynamic parameters of the coronary artery. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for quantifying coronary artery hemodynamics based on IVUS images, so as to solve the technical problem in the prior art that there is a lack of characteristic analysis of the hypoechoic region, namely "snowflake points", caused by blood flow in the intima reflected by IVUS images, making it difficult to quantitatively evaluate the hemodynamic parameters of the coronary artery.

[0006] To solve the above technical problem, the present invention specifically provides the following technical solutions:

[0007] A method for quantifying coronary artery hemodynamics based on IVUS images, comprising the following steps:

[0008] Simultaneously acquire IVUS images of the coronary artery while the patient undergoes coronary angiography;

[0009] In the IVUS image, the snowflake point target and the interference target are enhanced respectively to obtain the snowflake point target enhanced image and the interference target enhanced image;

[0010] The Mask R-CNN model is used to detect the snowflake point target and the interference target on the snowflake point target enhanced image and the interference target enhanced image respectively, and the snowflake point target mask image and the interference target mask image are obtained;

[0011] The IVUS image is processed with the fusion mask of the snowflake point target mask image and the interference target mask image to obtain the snowflake point segmentation result image;

[0012] Based on the snowflake point segmentation result image, the hemodynamics is quantified to obtain the coronary artery hemodynamic parameters.

[0013] As a preferred solution of the present invention, the interference targets include the lumen of the IVUS catheter, the diameter of the IVUS catheter, and plaque.

[0014] As a preferred solution of the present invention, the method for obtaining the snowflake point target enhanced image includes:

[0015] The Sobel operator is used to perform horizontal edge detection and vertical edge detection on the IVUS image to obtain the IVUS horizontal edge detection image and the IVUS vertical edge detection image;

[0016] The combined image of the IVUS horizontal edge detection image and the IVUS vertical edge detection image is added and fused with the IVUS image to obtain the snowflake point target enhanced image;

[0017] The enhancement operation process of the snowflake point target enhanced image is:

[0018] ;

[0019] In the formula, is the snowflake point target enhanced image, is the IVUS image, is the horizontal detection template of the Sobel operator, is the vertical detection template of the Sobel operator, is the enhancement coefficient, and are both absolute values;

[0020] Among them, ;

[0021] In the formula, is the L2 norm formula.

[0022] As a preferred embodiment of the present invention, the method for obtaining the interference target enhanced image includes:

[0023] Subtracting and fusing the combined image of the IVUS horizontal edge detection image and the IVUS vertical edge detection image with the IVUS image to obtain the interference target enhanced image;

[0024] The enhancement operation process of the interference target enhanced image is:

[0025] ;

[0026] In the formula, is the interference target enhanced image, is the IVUS image, is the horizontal detection template of the Sobel operator, is the vertical detection template of the Sobel operator, is the enhancement coefficient, and are both absolute values;

[0027] Among them, ;

[0028] In the formula, is the L2 norm formula.

[0029] As a preferred embodiment of the present invention, the method for obtaining the snowflake point target mask image includes:

[0030] Select multiple snowflake point target enhanced images, and mark the snowflake point area mask image on the snowflake point target enhanced image;

[0031] Form a data set with multiple snowflake point target enhanced images and snowflake point area mask images, and train a Mask R-CNN model based on the data set to pre-establish a snowflake point target detection model for snowflake point target detection on the snowflake point target enhanced image;

[0032] Use the snowflake point target detection model to detect the snowflake point target on the snowflake point target enhanced image to obtain the snowflake point target mask image;

[0033] The snowflake point target detection model is:

[0034] ;

[0035] In the formula, is the snowflake point target mask image, is the snowflake point target enhanced image, and Mask R-CNN is the Mask R-CNN model.

[0036] As a preferred embodiment of the present invention, the method for obtaining the interference target mask image includes:

[0037] Select multiple interference target enhanced images, and mark the interference region mask image on the interference target enhanced images;

[0038] Form a data set with multiple interference target enhanced images and interference region mask images, and train a Mask R-CNN model based on the data set to pre-establish an interference target detection model for detecting interference targets in the interference target enhanced images;

[0039] Use the interference target detection model to detect interference targets on the interference target enhanced images to obtain the interference target mask image;

[0040] The interference target detection model is:

[0041] ;

[0042] In the formula, is the interference target mask image, is the interference target enhanced image, and Mask R-CNN is the Mask R-CNN model.

[0043] As a preferred embodiment of the present invention, the method for obtaining the snowflake segmentation result image includes:

[0044] In the snowflake target mask image, remove the overlapping part of the snowflake target and the interference target between the snowflake target mask image and the interference target mask image to obtain the fusion mask;

[0045] Based on the fusion mask, perform pixel extraction on the IVUS image to obtain the snowflake segmentation result image;

[0046] The operation process of the fusion mask is:

[0047] ;

[0048] In the formula, is the fusion mask, is the snowflake target mask image, is the interference target mask image, mix is the overlapping operator, is the overlapping part of the snowflake target and the interference target between the snowflake target mask image and the interference target mask image.

[0049] As a preferred embodiment of the present invention, the method for quantifying the coronary artery hemodynamic parameters includes:

[0050] In the snowflake segmentation result image of the IVUS image of each coronary artery vessel cross-section, count the number of blood flow snowflakes;

[0051] Count the pixel number of each blood flow snowflake and sum them to obtain the blood flow snowflake area of the IVUS image of each coronary artery vessel cross-section;

[0052] Multiply the blood flow snowflake area of the IVUS image of each coronary artery vessel cross-section by the thickness of each coronary artery vessel cross-section to obtain the blood flow snowflake volume;

[0053] In the snowflake segmentation result image of the IVUS image of each coronary artery vessel cross-section, calculate the cross-sectional area of the lumen of the coronary artery vessel;

[0054] Calculate the ratio of the difference between the cross-sectional area of the lumen of the coronary artery vessel in the IVUS image of each coronary artery vessel cross-section and the normal cross-sectional area of the coronary artery vessel to the normal cross-sectional area of the coronary artery vessel to obtain the stenosis degree of the coronary artery vessel;

[0055] Multiply the ratio of the cross-sectional area of the lumen of the coronary artery vessel in the IVUS image to the normal cross-sectional area of the coronary artery vessel by the normal blood flow velocity to obtain the blood flow velocity at the coronary artery vessel;

[0056] Subtract the square of the normal blood flow velocity multiplied by the coefficient 4 from the square of the blood flow velocity at the coronary artery vessel to obtain the pressure gradient at the coronary artery vessel;

[0057] Multiply the blood flow velocity at the coronary artery vessel by the blood flow snowflake area to obtain the blood flow volume at the coronary artery vessel.

[0058] As a preferred solution of the present invention, the loss function for training the Mask R-CNN model is the segmentation mask loss, and the segmentation mask loss is:

[0059]

[0060] In the formula, is the segmentation mask loss, is the fusion mask, is the ground truth of the snowflake target mask image, is the total number of pixels of, is the overlapping part of the snowflake target and the interference target between the snowflake target mask image and the interference target mask image, is the snowflake target mask image, is the interference target mask image, and mix is the overlapping operator.

[0061] As a preferred embodiment of the present invention, the present invention provides a coronary artery hemodynamic quantification system based on IVUS images, which is applied to a coronary artery hemodynamic quantification method based on IVUS images. The system includes:

[0062] A data acquisition unit for acquiring IVUS images of coronary arteries while performing coronary angiography on a patient;

[0063] An image enhancement unit for enhancing the snowflake point target and the interference target in the IVUS image respectively to obtain a snowflake point target enhanced image and an interference target enhanced image;

[0064] A mask making unit for detecting the snowflake point target and the interference target on the snowflake point target enhanced image and the interference target enhanced image respectively by using the Mask R-CNN model to obtain a snowflake point target mask image and an interference target mask image;

[0065] A target segmentation unit for processing the IVUS image with the fusion mask of the snowflake point target mask image and the interference target mask image to obtain a snowflake point segmentation result image;

[0066] A hemodynamic quantification unit for quantifying the hemodynamics based on the snowflake point segmentation result image to obtain coronary artery hemodynamic parameters.

[0067] The present invention has the following beneficial effects compared with the prior art:

[0068] The present invention accurately segments the snowflake point region in the IVUS image by using a neural network model, realizes the accurate extraction of the information of the hypoechoic region caused by blood flow in the intima reflected by the IVUS image, and then analyzes the characteristics of the hypoechoic region caused by blood flow in the intima reflected by the IVUS image, that is, the "snowflake point", to realize the accurate quantitative evaluation of the hemodynamic parameters of the coronary artery. Description of the Drawings

[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in 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 exemplary, and for those of ordinary skill in the art, without creative efforts, other implementation drawings can be obtained according to the provided drawings.

[0070] Figure 1 It is a flowchart of the coronary artery hemodynamic quantification method based on IVUS images provided by the embodiments of the present invention;

[0071] Figure 2Block diagram of a coronary artery hemodynamic quantification system based on IVUS images provided by an embodiment of the present invention;

[0072] Figure 3 Schematic diagram of interference target annotation provided by an embodiment of the present invention;

[0073] Figure 4 Mask image of interference target provided by an embodiment of the present invention;

[0074] Figure 5 Mask image of snowflake target provided by an embodiment of the present invention;

[0075] Figure 6 Fusion mask image provided by an embodiment of the present invention;

[0076] Figure 7 Snowflake segmentation result image provided by an embodiment of the present invention. Detailed implementation manners

[0077] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0078] As Figure 1 shown, the present invention provides a method for quantifying coronary artery hemodynamics based on IVUS images, including the following steps:

[0079] Acquire IVUS images of coronary arteries while the patient undergoes coronary angiography;

[0080] In the IVUS images, enhance the snowflake targets and interference targets respectively to obtain enhanced snowflake target images and enhanced interference target images;

[0081] Use the Mask R-CNN model to detect snowflake targets and interference targets on the enhanced snowflake target images and enhanced interference target images respectively to obtain snowflake target mask images and interference target mask images;

[0082] Process the IVUS images with the fusion mask of the snowflake target mask image and the interference target mask image to obtain snowflake segmentation result images;

[0083] Quantify the hemodynamics based on the snowflake segmentation result images to obtain coronary artery hemodynamic parameters.

[0084] In order to quantitatively obtain accurate hemodynamic parameters of coronary artery vessels, the present invention precisely detects targets in IVUS images by using a neural network model, and realizes the accurate identification of the hypoechoic region caused by blood flow in the intima, that is, the snowflake point target, in the IVUS images of coronary artery vessels.

[0085] Specifically, the present invention combines two methods, direct detection of snowflake point targets and indirect detection of snowflake point targets, to achieve accurate image detection of snowflake point targets. Among them, the direct detection of snowflake point targets includes enhancing the snowflake point targets in the IVUS image and segmenting the enhanced snowflake point targets in the IVUS image (that is, in the IVUS image, enhancing the snowflake point targets to obtain an enhanced image of snowflake point targets; using the Mask R-CNN model to detect snowflake point targets and interfering targets on the enhanced image of snowflake point targets to obtain a mask image of snowflake point targets), and obtaining a mask image for extracting snowflake point targets in the IVUS image. Since the mask image can be directly mapped onto the original IVUS image to segment snowflake point targets from it, the detection of snowflake point targets can achieve the segmentation of snowflake point targets on the IVUS image and directly complete the detection of snowflake point targets.

[0086] The indirect detection of snowflake points includes enhancing interfering targets (that is, other targets except snowflake point targets, which will interfere with the recognition of snowflake point targets) in the IVUS image and segmenting the enhanced interfering targets in the IVUS image (in the IVUS image, enhancing the interfering targets to obtain an enhanced image of interfering targets; using the Mask R-CNN model to detect snowflake point targets and interfering targets on the enhanced image of interfering targets to obtain a mask image of interfering targets), and obtaining a mask image for extracting interfering targets in the IVUS image. Since the mask image can be directly mapped onto the original IVUS image to segment interfering targets from it, the accurate segmentation of interfering targets will leave snowflake point targets, which can avoid the mis-segmentation of snowflake point targets into interfering targets. Therefore, the detection of interfering targets can achieve the segmentation of interfering targets on the IVUS image, eliminate the interference noise during the segmentation of snowflake point targets, and indirectly complete the detection of snowflake point targets.

[0087] Therefore, the overlapping part between the mask image of snowflake point targets obtained by the direct detection of snowflake point targets and the mask image of interfering targets is removed, so as to remove the interference noise in the mask image of snowflake point targets, and thus obtain an accurate segmentation result with only snowflake point targets, providing a guarantee for quantitatively obtaining accurate hemodynamic parameters of coronary artery vessels.

[0088] Such as Figure 3 shown, the interfering targets include the lumen of the IVUS catheter, the diameter of the IVUS catheter, and plaques.

[0089] The method for obtaining a snowflake target enhanced image includes:

[0090] Performing horizontal edge detection and vertical edge detection on the IVUS image using the Sobel operator to obtain an IVUS horizontal edge detection image and an IVUS vertical edge detection image;

[0091] Adding and fusing the combined image of the IVUS horizontal edge detection image and the IVUS vertical edge detection image with the IVUS image to obtain a snowflake target enhanced image;

[0092] The enhancement operation process of the snowflake target enhanced image is:

[0093] ;

[0094] In the formula, is the snowflake target enhanced image, is the IVUS image, is the horizontal detection template of the Sobel operator, is the vertical detection template of the Sobel operator, is the enhancement coefficient, and are both absolute values;

[0095] Among them, ;

[0096] In the formula, is the L2 norm formula.

[0097] In the present invention, the edge information of the IVUS image is detected by the Sobel operator for edge detection. Since the snowflake targets are discrete and numerous, the number of edges of the snowflake targets is also large. By using the Sobel operator to detect the edges of the snowflake targets, an edge detection image is formed, and the edge detection image is superimposed on the original IVUS image, so that the edge information of the snowflake targets in the original IVUS image is superimposed and enhanced.

[0098] During this edge enhancement process, not only the edge information of the snowflake targets is superimposed and enhanced, but the edge information of the interference targets will also be enhanced to a certain extent. However, generally, the number of edges of the interference targets is small, and the difference from the snowflake targets is relatively large. After enhancement, the difference will also be magnified. Therefore, when the edge information of both the snowflake targets and the interference targets is enhanced, all the edge information of the discrete snowflake targets can be obtained, and at the same time, the difference information between the snowflake targets and the interference targets is magnified. Comparatively speaking, the expectation of snowflake target enhancement is achieved.

[0099] Moreover, an enhancement coefficient is added. The enhancement coefficient measures the edge detection effect between the edge detection image and the original IVUS image. If the difference in edge information between the edge detection image and the original IVUS image is large, the edge detection effect is better, the edge detection credibility is high, and a high enhancement coefficient is given. Enabling the edge detection image to be highly superimposed on the original IVUS image can enhance the enhancement effect of the snowflake target. Correspondingly, if the edge detection credibility is low, a low enhancement coefficient is given, so that the edge detection image is superimposed on the original IVUS image to a low degree, avoiding false edge detection and reducing the enhancement of the snowflake target.

[0100] The method for obtaining the interference target enhancement image includes:

[0101] Subtracting and fusing the combined image of the IVUS horizontal edge detection image and the IVUS vertical edge detection image with the IVUS image to obtain the interference target enhancement image;

[0102] The enhancement operation process of the interference target enhancement image is:

[0103] ;

[0104] In the formula, is the interference target enhancement image, is the IVUS image, is the horizontal detection template of the Sobel operator, is the vertical detection template of the Sobel operator, is the enhancement coefficient, and are both absolute values;

[0105] Among them, ;

[0106] In the formula, is the L2 norm formula.

[0107] In the present invention, the Sobel operator for edge detection is used to detect the edge information of the IVUS image. Since the snowflake targets are discrete and numerous, the number of edges possessed by the snowflake targets is also large. By using the Sobel operator, the edges of the snowflake targets are detected to form an edge detection image, and the original IVUS image subtracts the edge information of the edge detection image, so that the edge information of the snowflake targets in the original IVUS image is superimposed and weakened, and the edge information intensity of the interference targets will also be superimposed and weakened during the subtraction process.

[0108] Due to the setting of the enhancement coefficient, the intensity of the edge information in the edge detection image is attenuated. Subtracting in this way avoids subtracting the edge information of the original IVUS image from the edge detection image, which would cause the edge information of the snowflake target and the infected target to be subtracted from themselves to 0. Moreover, since the intensity of the edge information of the snowflake target is small and the intensity of the edge information of the interference target is large, after superposition and weakening, the edge information of the snowflake target is almost gone. Compared with the almost disappeared edge information of the snowflake target, the differential information between the two is still amplified.

[0109] Therefore, it can make the edge information of all discrete snowflake targets disappear, while amplifying the differential information between the snowflake target and the interference target. Comparatively speaking, the expectation of enhancing the interference target is achieved.

[0110] The method for obtaining the snowflake target mask image includes:

[0111] Select multiple enhanced snowflake target images, and mark the snowflake region mask image on the enhanced snowflake target image;

[0112] Form a data set with multiple enhanced snowflake target images and snowflake region mask images, and train a Mask R-CNN model based on the data set to pre-establish a snowflake target detection model for detecting snowflake targets in the enhanced snowflake target image;

[0113] Use the snowflake target detection model to detect snowflake targets on the enhanced snowflake target image to obtain the snowflake target mask image, as Figure 5 shown;

[0114] The snowflake target detection model is:

[0115] ;

[0116] In the formula, is the snowflake target mask image, is the enhanced snowflake target image, and Mask R-CNN is the Mask R-CNN model.

[0117] The method for obtaining the interference target mask image includes:

[0118] Select multiple enhanced interference target images, and mark the interference region mask image on the enhanced interference target image;

[0119] Form a data set with multiple enhanced interference target images and interference region mask images, and train a Mask R-CNN model based on the data set to pre-establish an interference target detection model for detecting interference targets in the enhanced interference target image;

[0120] Using the interference target detection model, detect the interference target on the interference target enhanced image to obtain the interference target mask image, as Figure 4 shown;

[0121] The interference target detection model is:

[0122] ;

[0123] In the formula, is the interference target mask image, is the interference target enhanced image, and Mask R-CNN is the Mask R-CNN model.

[0124] The method for obtaining the speckle segmentation result image includes:

[0125] In the speckle target mask image, remove the overlapping part of the speckle target and the interference target between the speckle target mask image and the interference target mask image to obtain the fusion mask, as Figure 6 shown;

[0126] Based on the fusion mask, perform pixel extraction on the IVUS image to obtain the speckle segmentation result image, as Figure 7 shown;

[0127] The operation process of the fusion mask is:

[0128] ;

[0129] In the formula, is the fusion mask, is the speckle target mask image, is the interference target mask image, mix is the overlapping operator, is the overlapping part of the speckle target and the interference target between the speckle target mask image and the interference target mask image.

[0130] The method for quantifying the coronary artery hemodynamic parameters includes:

[0131] In the speckle segmentation result image of the IVUS image of each coronary artery cross-section (i.e., the Figure 6 speckle target in), count the number of blood flow speckles;

[0132] Count the pixel number of each blood flow speckle and sum to obtain the blood flow speckle area of the IVUS image of each coronary artery cross-section;

[0133] By multiplying the blood flow speckle area of the IVUS image of each coronary artery cross-section by the thickness of each coronary artery cross-section, obtain the blood flow speckle volume;

[0134] In the speckle segmentation result image of the IVUS image of each cross-section of the coronary artery vessel, calculate the cross-sectional area of the lumen of the coronary artery vessel;

[0135] Calculate the ratio of the difference between the cross-sectional area of the lumen of the coronary artery vessel and the normal cross-sectional area of the coronary artery vessel in the IVUS image of each cross-section of the coronary artery vessel to the normal cross-sectional area of the coronary artery vessel to obtain the degree of stenosis of the coronary artery vessel;

[0136] Multiply the ratio of the cross-sectional area of the lumen of the coronary artery vessel to the normal cross-sectional area of the coronary artery vessel in the IVUS image by the normal blood flow velocity to obtain the blood flow velocity at the coronary artery vessel;

[0137] The square of the blood flow velocity at the coronary artery vessel minus the square of the normal blood flow velocity multiplied by the coefficient 4 to obtain the pressure gradient at the coronary artery vessel, which can be estimated by the Bernoulli equation, especially in the case of significant stenosis.

[0138] Among them, the normal cross-sectional area and normal blood flow velocity of the coronary artery vessel are within a general range due to individual differences. Left main coronary artery (LMCA): Usually about 3.5 - 4.5 mm in diameter, which means a cross-sectional area of about 10 - 16 square millimeters. Left anterior descending branch (LAD) and left circumflex branch (LCx): Usually about 2.5 - 4 mm in diameter, and the cross-sectional area is between 5 - 12 square millimeters. Right coronary artery (RCA): Usually about 2.5 - 3.5 mm in diameter, and the cross-sectional area is between 5 - 10 square millimeters. In this way, the final calculation may lead to deviations. To eliminate such deviations as much as possible, calculate the speckle values of multiple coronary artery vessels within a cycle through one or more cardiac cycles shown by electrocardiogram, so as to obtain the mean value of the hemodynamic parameters of the coronary artery vessel, which is more accurate.

[0139] Multiply the blood flow velocity at the coronary artery vessel by the blood flow speckle area to obtain the blood flow volume at the coronary artery vessel.

[0140] The loss function for training the Mask R-CNN model is the segmentation mask loss, and the segmentation mask loss is:

[0141]

[0142] In the formula, is the segmentation mask loss, is the fusion mask, is the ground truth of the speckle target mask image, is the total number of pixels of, is the overlapping part of the snowflake target and the interference target between the snowflake target mask image and the interference target mask image, is the snowflake target mask image, is the interference target mask image, and mix is the overlapping operator.

[0143] The present invention uses the segmentation mask loss as the loss function for training the Mask R-CNN model, which measures the fusion mask between the snowflake target mask image and the interference target mask image, that is, the difference degree between the target segmentation result combined by direct detection and indirect detection and the real result. Taking this as the loss function can enable the trained Mask R-CNN model to detect the snowflake target mask image and the interference target mask image to achieve the best segmentation result and the most accurate target detection result. which measures the overlapping part of the snowflake target and the interference target between the snowflake target mask image and the interference target mask image. Taking this as the loss function can enable the trained Mask R-CNN model to detect the snowflake target mask image and the interference target mask image to achieve the minimum overlap, ensuring that the segmented snowflake target and interference target have the minimum overlap and obtaining accurate results in both snowflake target detection and interference target detection.

[0144] As Figure 2 shown, the present invention provides a coronary artery vascular hemodynamic quantification system based on IVUS images, which is applied to a coronary artery vascular hemodynamic quantification method based on IVUS images. The system includes:

[0145] A data acquisition unit for acquiring IVUS images of coronary artery vessels while the patient undergoes coronary angiography;

[0146] An image enhancement unit for enhancing the snowflake target and the interference target respectively in the IVUS image to obtain a snowflake target enhanced image and an interference target enhanced image;

[0147] A mask making unit for using the Mask R-CNN model to detect the snowflake target and the interference target on the snowflake target enhanced image and the interference target enhanced image respectively to obtain a snowflake target mask image and an interference target mask image;

[0148] A target segmentation unit for processing the IVUS image with the fusion mask of the snowflake target mask image and the interference target mask image to obtain a snowflake segmentation result image;

[0149] A hemodynamic quantification unit for quantifying the hemodynamics based on the snowflake segmentation result image to obtain coronary artery vascular hemodynamic parameters.

[0150] Different from the FFR (Fractional Flow Reserve) and QFR (Quantitative Flow Ratio) in the coronary artery function evaluation method, the present invention can quantitatively evaluate the hemodynamics of coronary blood vessels without additionally increasing the catheter and guide wire, and further evaluate the impact of coronary artery stenosis on hemodynamics. In actual operation, reading IVUS data requires professional interventional technicians and has a very high threshold for image reading. The automated detection system of the present invention can not only quickly read images and detect relevant current technologies such as the position and characteristics of low echo regions, but also detect the hemodynamic parameters of the patient's coronary blood vessels, quantify the blood supply before and after stent implantation, reduce the threshold for image reading, and save various costs and time. Secondly, it avoids the side effects caused by injecting vasodilator drugs to be detected, which may cause discomfort to the patient, and the guide wire is prone to damage the blood vessels to be detected during the intervention process. This method provides more detailed and quantitative data for clinically evaluating the severity of coronary artery disease, especially for pre-operative hemodynamic evaluation, intraoperative plaque characteristics and stability, personalized treatment, and postoperative comprehensive evaluation of patients, and fills the gaps in some traditional evaluation methods.

[0151] The present invention uses a neural network model to accurately segment the snowflake point area in the IVUS image, realizes the accurate extraction of the information of the low echo area caused by blood flow in the intima reflected by the IVUS image, and then analyzes the characteristics of the low echo area caused by blood flow in the intima reflected by the IVUS image, that is, the "snowflake point", to realize the accurate quantitative evaluation of the hemodynamic parameters of the coronary blood vessels.

[0152] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the present application within the essence and protection scope of the present application, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present application.

Claims

1. A method for quantifying coronary artery hemodynamics based on IVUS images, characterized in that: The following steps are involved: IVUS images of the coronary arteries are obtained while the patient undergoes coronary angiography; In the IVUS image, the snowflake point target and the interference target are enhanced respectively to obtain the snowflake point target enhanced image and the interference target enhanced image; The Mask R-CNN model is used to detect snowflake point targets and interference targets on the snowflake point target enhanced image and the interference target enhanced image respectively, and the snowflake point target mask image and the interference target mask image are obtained; The IVUS image is processed by a fusion mask of the snowflake point target mask image and the interference target mask image to obtain a snowflake point segmentation result image; The hemodynamics is quantified based on the snowflake point segmentation result image to obtain the coronary artery hemodynamic parameters; The method for acquiring the snowflake point target enhanced image includes: The Sobel operator is used to perform transverse edge detection and longitudinal edge detection on the IVUS image to obtain an IVUS transverse edge detection image and an IVUS longitudinal edge detection image; The combined image of IVUS transverse edge detection image and IVUS longitudinal edge detection image is added and fused with the IVUS image to obtain the snowflake point target enhanced image; The enhancement operation process of the snowflake target enhanced image is as follows: ; In the formula, Enhance the image for snowflake targets, For IVUS images, is the horizontal detection template of the Sobel operator, is the vertical detection template of the Sobel operator, is the enhancement factor, and All are absolute values; in, ; In the formula, is the L2 norm.

2. The method for quantifying coronary artery hemodynamics based on IVUS images according to claim 1, characterized in that: The interference targets include the lumen of the IVUS catheter, the caliber of the IVUS catheter, and plaques.

3. The method for quantifying coronary artery hemodynamics based on IVUS images according to claim 2, characterized in that: The method for acquiring the interference target enhanced image comprises: Subtract and fuse the combined image of IVUS transverse edge detection image and IVUS longitudinal edge detection image with the IVUS image to obtain an interference target enhanced image; The enhancement operation process of the interference target enhanced image is as follows: ; In the formula, Enhance the image for the interference target, For IVUS images, is the horizontal detection template of the Sobel operator, is the vertical detection template of the Sobel operator, is the enhancement factor, and are all absolute values; ; In the formula, is the L2 norm.

4. The method for quantifying coronary artery hemodynamics based on IVUS images according to claim 3, characterized in that: The method for acquiring the snowflake point target mask image comprises: Selecting multiple snowflake point target enhanced images, and marking snowflake point area mask images on the snowflake point target enhanced images; A data set is formed by combining multiple snowflake target enhanced images and snowflake region mask images, and a Mask R-CNN model is trained based on the data set to pre-establish a snowflake target detection model for detecting snowflake targets in snowflake target enhanced images; Using the snowflake point target detection model, the snowflake point target is detected on the snowflake point target enhanced image to obtain the snowflake point target mask image; The snowflake target detection model is: ; In the formula, is the snowflake target mask image, The image is enhanced for snowflake targets, and Mask R-CNN is the Mask R-CNN model.

5. The method for quantifying coronary artery hemodynamics based on IVUS images according to claim 4, characterized in that: The method for acquiring the interference target mask image includes: Selecting multiple interference target enhanced images, and marking interference area mask images on the interference target enhanced images; A plurality of interference target enhanced images and interference area mask images are combined into a data set, and a Mask R-CNN model is trained based on the data set to pre-establish an interference target detection model for detecting interference targets in interference target enhanced images; Using the interference target detection model, the interference target is detected on the interference target enhanced image to obtain the interference target mask image; The interference target detection model is: ; In the formula, To interfere with the target mask image, To enhance the image for distracting targets, Mask R-CNN is the Mask R-CNN model.

6. The method for quantifying coronary artery hemodynamics based on IVUS images according to claim 5, characterized in that: The method for obtaining the snowflake point segmentation result image comprises: In the snowflake point target mask image, the overlapping parts of the snowflake point target and the interference target mask image are removed to obtain the fused mask; Perform pixel extraction on the IVUS image based on the fusion mask to obtain the snowflake point segmentation result image; The calculation process of the fusion mask is as follows: ; In the formula, is the fusion mask, is the snowflake target mask image, To interfere with the target mask image, To take the coincidence operator, It is the overlapping part of the snowflake point target and the interference target in the snowflake point target mask image and the interference target mask image.

7. The method for quantifying coronary artery hemodynamics based on IVUS images according to claim 6, characterized in that: The method for quantifying the coronary artery hemodynamic parameters comprises: In the snowflake point segmentation result image of the IVUS image of each coronary artery cross-section, the number of blood flow snowflake points is counted; The number of pixels of each blood flow snowflake point is counted, and the sum is calculated to obtain the blood flow snowflake point area of ​​each coronary artery cross-section IVUS image; The blood flow snowflake volume is obtained by multiplying the blood flow snowflake area of ​​the IVUS image of each coronary artery cross section by the thickness of each coronary artery cross section. In the snowflake point segmentation result image of the IVUS image of each coronary artery cross-section, the lumen cross-sectional area of ​​the coronary artery is calculated; The difference between the lumen cross-sectional area of ​​the coronary artery in the IVUS image of each coronary artery cross-section and the normal lumen cross-sectional area of ​​the coronary artery is calculated by ratio with the normal lumen cross-sectional area of ​​the coronary artery to obtain the degree of coronary artery stenosis. The blood flow velocity at the coronary artery is obtained by multiplying the ratio of the lumen cross-sectional area of ​​the coronary artery in the IVUS image to the normal lumen cross-sectional area of ​​the coronary artery by the normal blood flow velocity; The square of the blood flow velocity in the coronary artery minus the square of the normal blood flow velocity multiplied by a coefficient of 4 gives the pressure gradient in the coronary artery. The blood flow velocity in the coronary arteries is multiplied by the blood flow snowflake area to obtain the blood flow in the coronary arteries.

8. The method for quantifying coronary artery hemodynamics based on IVUS images according to claim 6, characterized in that: The loss function for training the Mask R-CNN model is the segmentation mask loss, which is: ; In the formula, is the segmentation mask loss, is the fusion mask, is the true value of the snowflake target mask image, for The total number of pixels, is the overlapped part of the snowflake point target mask image and the interference target mask image. is the snowflake target mask image, is to interfere with the target mask image, and mix is ​​the coincidence operator.

9. A coronary artery hemodynamic quantification system based on IVUS images, characterized in that: A method for quantifying coronary artery hemodynamics based on IVUS images as described in any one of claims 1 to 8, the system comprising: A data acquisition unit, used for simultaneously acquiring IVUS images of coronary arteries based on coronary angiography of the patient; An image enhancement unit is used to enhance the snowflake point target and the interference target in the IVUS image, respectively, to obtain a snowflake point target enhanced image and an interference target enhanced image; A mask making unit is used to detect snowflake point targets and interference targets on the snowflake point target enhanced image and the interference target enhanced image respectively by using the Mask R-CNN model to obtain a snowflake point target mask image and an interference target mask image; The target segmentation unit is used to process the IVUS image with a fusion mask of the snowflake point target mask image and the interference target mask image to obtain a snowflake point segmentation result image; The hemodynamic quantification unit is used to quantify the hemodynamics based on the snowflake point segmentation result image to obtain the coronary artery vascular hemodynamic parameters.

Images