Blood vessel segmentation method and device, medical imaging apparatus, and storage medium

Abstract

Embodiments of the present application disclose a blood vessel segmentation method and device, a medical imaging device and a storage medium. The method comprises: acquiring a blood vessel image to be segmented and a blood vessel distribution map corresponding to the blood vessel image to be segmented; inputting the blood vessel image to be segmented and the blood vessel distribution map into a preset blood vessel segmentation model to obtain a blood vessel segmentation result. The above technical solution solves the problem that the blood vessel segmentation method in the prior art is not applicable to the segmentation of relatively complex blood vessel images such as coronary images, and improves the blood vessel segmentation precision.

Description

[0001] The present application is a divisional application of the case with the application date of "December 28, 2018", the application number of "201811627428.3", and the invention name of "Vessel Segmentation Method, Device, Medical Imaging Equipment and Storage Medium". TECHNICAL FIELD

[0002] Embodiments of the present application relate to the technical field of image processing, in particular to a vessel segmentation method, device, medical imaging equipment and storage medium. BACKGROUND

[0003] Angiography is an interventional treatment method, which infuses a contrast agent into a blood vessel to observe the final imaging characteristics to infer the location and extent of blood vessel lesions. Angiography has a very important significance in the diagnosis and treatment of heart and great vessel diseases, peripheral vascular diseases, and tumors. For example, a doctor can judge the malignancy of a patient's liver cancer through hepatic vein angiography, diagnose cerebral vascular diseases by observing cerebral vein angiography, and observe coronary angiography, which has become an important basis for the diagnosis and treatment of coronary heart disease.

[0004] In the prior art, a threshold segmentation method is usually used to segment and extract a blood vessel image. Specifically, the blood vessel structure is first enhanced, and then a filter with a specific threshold is designed by hand to further enhance the tubular structure and suppress the non-tubular structure, so as to finally realize the segmentation and extraction of the blood vessel.

[0005] However, the threshold segmentation method is only suitable for application scenarios of segmenting simple blood vessel images. For more complex images, such as coronary images, the contrast of the blood vessels is low, and the influence of factors such as uneven distribution of contrast agents makes it difficult for the threshold segmentation method to distinguish the coronary structure and the background area, so the segmentation of the coronary artery still has great challenges. SUMMARY

[0006] The present application provides a vessel segmentation method, device, medical imaging equipment and storage medium to improve the segmentation accuracy of blood vessels.

[0007] In a first aspect, the embodiments of the present application provide a vessel segmentation method, comprising:

[0008] obtaining a to-be-segmented blood vessel image and a blood vessel distribution map corresponding to the to-be-segmented blood vessel image;

[0009] inputting the to-be-segmented blood vessel image and the blood vessel distribution map into a preset vessel segmentation model to obtain a vessel segmentation result.

[0010] In a second aspect, the embodiments of the present application further provide a vessel segmentation device, comprising:

[0011] The acquisition module is configured to acquire a to-be-segmented blood vessel image and a blood vessel distribution map corresponding to the to-be-segmented blood vessel image.

[0012] The segmentation module is configured to input the to-be-segmented blood vessel image and the blood vessel distribution map into a preset blood vessel segmentation model to obtain a blood vessel segmentation result.

[0013] In a third aspect, an embodiment of the present application further provides a medical imaging device, which comprises an input device and an output device, and further comprises:

[0014] one or more processors;

[0015] a memory configured to store one or more programs;

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the blood vessel segmentation method provided in the first aspect.

[0017] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the blood vessel segmentation method provided in the first aspect.

[0018] The embodiment of the present application acquires a to-be-segmented blood vessel image and a blood vessel distribution map corresponding to the to-be-segmented blood vessel image, inputs the to-be-segmented blood vessel image and the blood vessel distribution map into a preset blood vessel segmentation model, and obtains a blood vessel segmentation result. The above technical solution solves the problem that the blood vessel segmentation method in the prior art is not applicable to the segmentation of relatively complex blood vessel images such as coronary images, and improves the blood vessel segmentation precision. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of a blood vessel segmentation method in the first embodiment of the present application;

[0020] Figure 2 is a flowchart of a blood vessel segmentation method in the second embodiment of the present application;

[0021] Figure 3A is a flowchart of a blood vessel segmentation method in the third embodiment of the present application;

[0022] Figure 3B is a flowchart of a blood vessel segmentation model training method in the third embodiment of the present application;

[0023] Figure 4A is a flowchart of a blood vessel segmentation method in the fourth embodiment of the present application;

[0024] Figure 4B is a coronary angiogram in the fourth embodiment of the present application;

[0025] Figure 4C This is a blood vessel extraction diagram from Embodiment 4 of the present invention;

[0026] Figure 4D This is a schematic diagram of the model testing process in Embodiment 4 of the present invention;

[0027] Figure 4E This is the result of blood vessel segmentation in Embodiment 4 of the present invention;

[0028] Figure 4F This is the result of vessel segmentation using only coronary angiography images in Embodiment 4 of the present invention;

[0029] Figure 4G This is the medical gold standard diagram in Embodiment 4 of the present invention;

[0030] Figure 4H This is the target segmentation result in Embodiment 4 of the present invention;

[0031] Figure 5 This is a structural diagram of a blood vessel segmentation device according to Embodiment 5 of the present invention;

[0032] Figure 6 This is a schematic diagram of the structure of a medical imaging device according to Embodiment Six of the present invention. Detailed Implementation

[0033] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0034] Example 1

[0035] Figure 1 This is a flowchart of a blood vessel segmentation method according to Embodiment 1 of the present invention. This embodiment is applicable to situations where blood vessel tissue in a blood vessel image is separated from the background region. The method is executed by a blood vessel segmentation device, which is implemented by software and / or hardware and specifically configured in a medical imaging device. The medical imaging device can be a CT (Computed Tomography) device, an MRI (Magnetic Resonance Imaging) device, or a DSA (Digital Subtraction Angiography) device, etc.

[0036] like Figure 1 The illustrated method for segmenting blood vessels includes:

[0037] S110, acquire a to-be-segmented blood vessel image and a blood vessel distribution image corresponding to the to-be-segmented blood vessel image.

[0038] The to-be-segmented blood vessel image can be at least one of a CT image, an MRI image, and a DSA image obtained by scanning machine data reconstruction through a medical imaging device. The blood vessel distribution image is a relatively rough blood vessel image obtained after blood vessel extraction on the to-be-segmented blood vessel image.

[0039] For example, the blood vessel extraction on the to-be-segmented blood vessel image can be performed by a threshold segmentation method, or the to-be-segmented image can be processed using at least one filtering algorithm to obtain the corresponding blood vessel distribution image. The filtering algorithm can be at least one of a Gaussian filtering algorithm, a linear filtering algorithm, a Wiener filtering algorithm, and a threshold filtering algorithm.

[0040] It should be noted that the to-be-segmented blood vessel image can be directly acquired from a local storage of a medical imaging device, or from other storage devices or cloud storage associated with the medical imaging device. The to-be-segmented blood vessel image can also be obtained by controlling the medical imaging device to perform a scanning operation to obtain scanning data, and then reconstructing the scanning data to obtain the to-be-segmented blood vessel image.

[0041] Correspondingly, the blood vessel distribution image corresponding to the to-be-segmented blood vessel image can be directly acquired from a local storage of a medical imaging device, or from other storage devices or cloud storage associated with the medical imaging device. The blood vessel distribution image corresponding to the to-be-segmented blood vessel image can also be obtained by performing blood vessel extraction on the acquired to-be-segmented blood vessel image to obtain the corresponding blood vessel distribution image.

[0042] S120, input the to-be-segmented blood vessel image and the blood vessel distribution image into a preset blood vessel segmentation model to obtain a blood vessel segmentation result.

[0043] The preset blood vessel segmentation model is obtained by training an existing model using a large amount of learning data. The existing model can be a convolutional neural network model, such as a U-Net convolutional neural network or a V-Net convolutional neural network.

[0044] After acquiring the to-be-segmented blood vessel image and the blood vessel distribution image corresponding to the to-be-segmented blood vessel image, the to-be-segmented blood vessel image and the blood vessel distribution image are input into the preset blood vessel segmentation model as prediction samples to obtain a blood vessel segmentation result.

[0045] It can be understood that, in order to preserve the complete image of the blood vessel tissue and filter out irrelevant images other than the blood vessel tissue, thereby improving the segmentation accuracy of the obtained blood vessels, the largest blood vessel connected graph in the blood vessel segmentation result can be extracted after obtaining the blood vessel segmentation result to obtain a target blood vessel segmentation image.

[0046] The embodiment of the present application obtains a to-be-segmented blood vessel image and a blood vessel distribution map corresponding to the to-be-segmented blood vessel image; inputs the to-be-segmented blood vessel image and the blood vessel distribution map into a preset blood vessel segmentation model to obtain a blood vessel segmentation result. The above technical solution solves the problem that the blood vessel segmentation method in the prior art is not applicable to the segmentation of relatively complex blood vessel images such as coronary images, and improves the blood vessel segmentation precision.

[0047] Embodiment two

[0048] Figure 2 is a flowchart of a blood vessel segmentation method in the second embodiment of the present application. The embodiment of the present application further optimizes the technical solutions of the above embodiments.

[0049] Further, the operation of inputting the to-be-segmented blood vessel image and the blood vessel distribution map into the preset blood vessel segmentation model is refined into the operation of fusing the to-be-segmented blood vessel image and the blood vessel distribution map and inputting the fused image into the preset blood vessel segmentation model, so as to improve the use mechanism of the preset blood vessel segmentation model.

[0050] Further, the operation of obtaining the blood vessel distribution map corresponding to the to-be-segmented blood vessel image is refined into the operation of processing the to-be-segmented blood vessel image using at least one filtering algorithm to obtain the blood vessel distribution map, so as to improve the determination mechanism of the blood vessel distribution map.

[0051] As shown in a blood vessel segmentation method in the first embodiment of the present application, the method comprises the following steps: Figure 2 S210, a to-be-segmented blood vessel image is obtained.

[0052] S220, at least one filtering algorithm is used to process the to-be-segmented blood vessel image to obtain a blood vessel distribution map.

[0053]

[0054] The filtering algorithm includes at least one of a Gaussian filtering algorithm, a tubular filtering algorithm, a linear filtering algorithm, and a Wiener filtering algorithm.

[0055] For example, the processing of the to-be-segmented blood vessel image using at least one filtering algorithm to obtain the blood vessel distribution map can be: calculating the eigenvalues of the Hessian matrix of each pixel point in the to-be-segmented blood vessel image; processing the eigenvalues using a tubular filter to obtain response values; and combining the response values corresponding to the pixel points to obtain the blood vessel distribution map.

[0056] ​Since the eigenvalue of the Hessian matrix of a pixel point can be associated with the shape feature corresponding to the pixel point, the blood vessel image to be segmented can be associatedly mapped by calculating the eigenvalue of the Hessian matrix of each pixel point, and each eigenvalue can be enhanced by a tubular filter to enhance the image of the tubular morphological blood vessel tissue and suppress the image of other non-blood vessel tissue. After the eigenvalue is enhanced by the tubular filter, the response value corresponding to each pixel point is obtained, and the blood vessel distribution map is obtained by combining the response values through the arrangement order of each pixel point.

[0057] It can be understood that, in order to improve the accuracy of the obtained blood vessel distribution map and further improve the blood vessel segmentation accuracy of the trained blood vessel segmentation model, at least one target filter can be used to smooth the blood vessel image to be segmented before the eigenvalue of the Hessian matrix of each pixel point in the blood vessel image to be segmented is calculated, and the smoothed image is used to replace the blood vessel image to be segmented; wherein the target filter includes at least one of a Gaussian filter, a linear filter and a Wiener filter.

[0058] Of course, in order to further improve the accuracy of the obtained blood vessel distribution map, the response value corresponding to each pixel point can be combined to obtain the blood vessel distribution map, which can be to compare the blood vessel response values corresponding to the same pixel point based on different target filters, and to obtain the maximum response value as the target response value; the target response value corresponding to each pixel point is combined to form the blood vessel distribution map.

[0059] Alternatively, the response value corresponding to each pixel point can be combined to obtain the blood vessel distribution map, which includes: obtaining the blood vessel distribution map corresponding to different target filters as a candidate distribution map; counting the number of pixel points with a response value greater than a set threshold in each candidate distribution map; and determining the candidate distribution map with the most pixel points as the final blood vessel distribution map.

[0060] Alternatively, the response value corresponding to each pixel point can be combined to obtain the blood vessel distribution map, which includes: comparing the blood vessel response values corresponding to the same pixel point based on different target filters, and obtaining the maximum response value as the target response value; and combining the target response value corresponding to each pixel point to form the blood vessel distribution map.

[0061] S230, the blood vessel image to be segmented and the blood vessel distribution map are fused, and the fused image is input into a preset blood vessel segmentation model to obtain a blood vessel segmentation result.

[0062] Specifically, a prediction sample is formed based on the blood vessel image to be segmented and the blood vessel distribution map corresponding to the blood vessel image to be segmented, the prediction is input into a preset blood vessel segmentation model, and a blood vessel segmentation result is obtained.

[0063] Optionally, the to-be-segmented blood vessel image and the blood vessel distribution map are fused, and the fused image is input into the preset blood vessel segmentation model, including: the to-be-segmented blood vessel image and the blood vessel distribution map are input into different channels of the preset blood vessel segmentation model input in parallel, and the images of each channel are input into the preset blood vessel segmentation model respectively.

[0064] Or optionally, the to-be-segmented blood vessel image and the blood vessel distribution map are fused, and the fused image is input into the preset blood vessel segmentation model, including: each pixel point of the to-be-segmented blood vessel image and the blood vessel distribution map is multiplied or added to obtain a fused image, and the fused image is input into the preset blood vessel segmentation model.

[0065] The embodiment of the application optimizes the use process of the preset blood vessel segmentation model, specifically fuses the to-be-segmented blood vessel image and the blood vessel distribution map, and inputs the fused image into the preset blood vessel segmentation model for use, thereby improving the use mechanism of the preset blood vessel segmentation model; the acquisition process of the blood vessel distribution map is refined to process the to-be-segmented blood vessel image using at least one filtering algorithm to obtain the blood vessel distribution map, thereby improving the determination mechanism of the blood vessel distribution map, so that the segmentation of the blood vessel can be realized by acquiring the to-be-segmented blood vessel image, reducing the acquisition process of the input data, and improving the user experience.

[0066] Embodiment three

[0067] Figure 3A It is a flowchart of a blood vessel segmentation method in the third embodiment of the application, and the embodiment of the application further optimizes and subdivides the technical solutions of the above-mentioned embodiments.

[0068] Further, before the operation of "inputting the to-be-segmented blood vessel image and the blood vessel distribution map into the preset blood vessel segmentation model to obtain a blood vessel segmentation result", "training the blood vessel segmentation model" is added; further, "training the blood vessel segmentation model" is subdivided into "determining a corresponding historical blood vessel distribution map according to at least one historical blood vessel image", "inputting the historical blood vessel image and the historical blood vessel distribution map into the blood vessel segmentation model to be trained to obtain a current blood vessel segmentation result corresponding to each historical blood vessel image", and "adjusting the prediction parameters of the blood vessel segmentation model according to the error between the current blood vessel segmentation result and the expected segmentation result", so as to improve the training mechanism of the blood vessel segmentation model.

[0069] As shown in a blood vessel segmentation method in Figure 3A , including:

[0070] S310, training the blood vessel segmentation model.

[0071] S320, acquire a to-be-segmented blood vessel image and a blood vessel distribution map corresponding to the to-be-segmented blood vessel image.

[0072] S330, input the to-be-segmented blood vessel image and the blood vessel distribution map into a preset blood vessel segmentation model to obtain a blood vessel segmentation result.

[0073] It should be noted that S310 can be executed before S320 or after S320, and the execution order of S310 and S320 is not limited herein.

[0074] Of course, after S330, the to-be-segmented blood vessel image and the blood vessel distribution map can also be input into the blood vessel segmentation model as training samples for training, so as to optimize the used blood vessel segmentation model.

[0075] Referring to a blood vessel segmentation model training method in Figure 3B The method comprises the following steps:

[0076] S311, determining a corresponding historical blood vessel distribution map according to at least one historical blood vessel image.

[0077] The historical blood vessel image can be at least one of a CT image, an MRI image, and a DSA image.

[0078] It can be understood that, in order to improve the blood vessel segmentation accuracy of the trained blood vessel segmentation model, it is preferred to train a blood vessel segmentation model corresponding to a blood vessel image of the same category using a blood vessel image of the same category, and to perform blood vessel segmentation on a to-be-segmented blood vessel image of the same category using the trained blood vessel segmentation model. For example, training a coronary artery segmentation model using a coronary artery image, training a choroid blood vessel segmentation model using a choroid blood vessel image, and training a hepatic vein segmentation model using a hepatic vein angiography blood vessel image.

[0079] Optionally, determining a corresponding historical blood vessel distribution map according to at least one historical blood vessel image can be: processing each of the historical blood vessel images using at least one filtering algorithm to obtain the historical blood vessel distribution map. The filtering algorithm comprises at least one of a Gaussian filtering algorithm, a tubular filtering algorithm, a linear filtering algorithm, and a Wiener filtering algorithm.

[0080] For example, processing each of the historical blood vessel images using at least one filtering algorithm to obtain the historical blood vessel distribution map can be: calculating the eigenvalues of the Hessian matrix of each pixel point in the historical blood vessel image; enhancing the eigenvalues using a tubular filter to obtain response values; and combining the response values corresponding to each pixel point to obtain the historical blood vessel distribution map.

[0081] Since the eigenvalue of the Hessian matrix of a pixel point can be associated with the shape feature corresponding to the pixel point, the historical blood vessel image can be associatedly mapped by calculating the eigenvalue of the Hessian matrix of each pixel point, and the eigenvalue is enhanced by a tubular filter to enhance the image of the tubular shape of the blood vessel tissue, while suppressing the image of other non-blood vessel tissues. After the eigenvalue is enhanced by the tubular filter, the response value corresponding to each pixel point is obtained, and the historical blood vessel distribution map is obtained by combining the response values of each pixel point in the order of the arrangement of each pixel point.

[0082] It can be understood that, in order to improve the accuracy of the obtained historical blood vessel distribution map, and further improve the blood vessel segmentation accuracy of the trained blood vessel segmentation model, at least one target filter can be used to smooth the historical blood vessel image before calculating the eigenvalue of the Hessian matrix of each pixel point in the historical blood vessel image, and the smoothed image is used to replace the historical blood vessel image; wherein the target filter includes at least one of a Gaussian filter, a linear filter and a Wiener filter.

[0083] Of course, in order to further improve the accuracy of the obtained historical blood vessel distribution map, the response values corresponding to each pixel point are combined to obtain the historical blood vessel distribution map, which can be obtained as a candidate distribution map corresponding to different target filters; the response mean value corresponding to the candidate distribution map is determined by the response values corresponding to each pixel point; the candidate distribution map with the largest response mean value is selected as the final historical blood vessel distribution map.

[0084] Alternatively, the response values corresponding to each pixel point are combined to obtain the historical blood vessel distribution map, which includes: obtaining the historical blood vessel distribution map corresponding to different target filters as a candidate distribution map; counting the number of pixel points with response values greater than a set threshold in each candidate distribution map; and determining the candidate distribution map with the most pixel points as the final historical blood vessel distribution map.

[0085] Alternatively, the response values corresponding to each pixel point are combined to obtain the historical blood vessel distribution map, which includes: comparing the blood vessel response values corresponding to the same pixel point based on different target filters, and obtaining the maximum response value as the target response value; and combining the target response values corresponding to each pixel point to form the historical blood vessel distribution map.

[0086] It can be understood that, in order to improve the segmentation accuracy when using the blood vessel segmentation model to segment the blood vessel image to be segmented, preferably, the determination process of the blood vessel distribution map corresponding to the blood vessel image to be segmented is completely consistent with the determination process of the historical blood vessel distribution map corresponding to the historical blood vessel image in the blood vessel segmentation model training process.

[0087] S312, input the historical blood vessel image and the historical blood vessel distribution map into the blood vessel segmentation model to be trained to obtain a current blood vessel segmentation result corresponding to each historical blood vessel image.

[0088] Specifically, a training sample set is formed based on at least one historical blood vessel image and a historical blood vessel distribution map corresponding to each historical blood vessel image, the training sample set is input into the blood vessel segmentation model to be trained for training, and a current blood vessel segmentation result corresponding to each historical blood vessel image is obtained.

[0089] Exemplarily, the historical blood vessel image and the historical blood vessel distribution map are input into the blood vessel segmentation model to be trained for training, which can be fusing the historical blood vessel image and the historical blood vessel distribution map, and inputting the fused image into the blood vessel segmentation model to be trained to obtain a final preset blood vessel segmentation model.

[0090] Optionally, the historical blood vessel image and the historical blood vessel distribution map are fused, and the fused image is input into the blood vessel segmentation model to be trained, which includes connecting the historical blood vessel image and the historical blood vessel distribution map in parallel into different channels of the blood vessel segmentation model to be trained, and inputting the images of each channel into the blood vessel segmentation model to be trained respectively; or multiplying or adding each pixel point of the historical blood vessel image and the historical blood vessel distribution map to obtain a fused image, and inputting the fused image into the blood vessel segmentation model to be trained.

[0091] It can be understood that, in order to improve the segmentation accuracy when using the blood vessel segmentation model to segment a blood vessel image to be segmented, preferably, the fusion process of the blood vessel image to be segmented and the corresponding blood vessel distribution map is completely consistent with the fusion process of the historical blood vessel image and the corresponding historical blood vessel distribution map in the training process of the blood vessel segmentation model.

[0092] S313, adjusting a prejudgment parameter of the blood vessel segmentation model according to an error between the current blood vessel segmentation result and an expected segmentation result.

[0093] An expected segmentation result corresponding to each historical blood vessel image is obtained, and a model accuracy when the training sample set is currently used to train the blood vessel segmentation model is determined according to an error between the expected segmentation result and the current blood vessel segmentation result; when the model accuracy meets a preset accuracy threshold, it is determined that the current blood vessel segmentation model can end the training; when the model accuracy does not meet the preset accuracy threshold, the prejudgment parameter of the blood vessel segmentation model is adjusted according to an error between the current blood vessel segmentation result and the expected segmentation result, so as to reduce the error between the current blood vessel segmentation result and the expected segmentation result, until the model accuracy meets the preset accuracy threshold.

[0094] The expected segmentation result can be a medical gold standard segmentation image corresponding to the historical blood vessel image. The preset accuracy threshold can be determined by a technician according to a need or an experience value.

[0095] The embodiment of the present application determines the corresponding historical blood vessel distribution map according to at least one historical blood vessel image before using the blood vessel segmentation model to segment the blood vessel, inputs the historical blood vessel image and the historical blood vessel distribution map into the blood vessel segmentation model to be trained to obtain the current blood vessel segmentation result corresponding to the historical blood vessel image, and adjusts the prejudgment parameter of the blood vessel segmentation model according to the error between the current blood vessel segmentation result and the expected segmentation result. By using the above technical solution, the training mechanism of the blood vessel segmentation model is improved, and by using the blood vessel segmentation model, the effective segmentation of the blood vessel tissue in the blood vessel image to be segmented is realized, and the segmentation accuracy of the blood vessel is improved.

[0096] Embodiment four

[0097] Figure 4A is a flowchart of a blood vessel segmentation method in the embodiment four of the present application. The embodiment of the present application provides a preferred implementation mode on the basis of the technical solutions of the above-mentioned embodiments, and the segmentation process of the coronary CT image is exemplarily described.

[0098] As shown in a blood vessel segmentation method in the embodiment four of the present application, the method comprises the following steps. Figure 4A

[0099] S410, acquiring a coronary angiogram.

[0100] Referring to the coronary angiogram shown in the embodiment four of the present application. Figure 4B

[0101] S420, filtering the coronary angiogram by using a plurality of Gaussian filters with different kernels to obtain a plurality of initial coronary images.

[0102] Specifically, the Gaussian kernel of 3x3, 5x5 and 7x7 is respectively used to perform Gaussian smoothing on the coronary angiogram to obtain the initial coronary image, so as to remove the sensitivity of the subsequent processing process to the Gaussian noise, and at the same time, the target object of different sizes is used by the way of multi-scale filtering. The target object can be understood as each component of the blood vessel tissue to be segmented.

[0103] S430, calculating the Hessian matrix of each pixel point in the initial coronary image, and obtaining the eigenvalue of each pixel point by the Jacobi method.

[0104] Specifically, the Hessian matrix of each pixel point of each initial coronary image is calculated, and the eigenvalue is calculated for each Hessian matrix. Since the initial coronary image is a three-dimensional image, each pixel point corresponds to three eigenvalues.

[0105] ​​S440, enhancing the eigenvalue of each pixel point by using a tubular filter to obtain a response value.

[0106] Since the blood vessels generally present a tubular shape, when a certain pixel point is located in a blood vessel, the response value is relatively large, and vice versa. By reasonably setting the bandwidth of the tubular filter, the non-target objects can be removed to a certain extent.

[0107] Specifically, a function F = |a2| * (|a2| - |a3|) / |a1| * sigma is used 2 The eigenvalue of each pixel point is enhanced, wherein a1, a2 and a3 are the eigenvalues of each pixel point obtained in different dimensions, F is a response value, and sigma is a filter bandwidth.

[0108] S450, comparing the response values of the same pixel points in different initial coronary artery maps, and selecting the maximum response value corresponding to each pixel point to obtain a blood vessel extraction map.

[0109] Referring to Figure 4C , a blood vessel extraction map is shown. Among them, the response value of the non-target object is small, so the brightness of the corresponding pixel point in the map is relatively weak; the response value of the target object is large, so the brightness of the corresponding pixel point in the map is relatively strong. However, due to the limitation of tubular filtering, there are still some pixel points corresponding to the brightness of the rib, trachea and other blood vessels in the map.

[0110] S460, inputting the coronary angiography map and the blood vessel extraction map as two input sources into the pre-trained V-Net neural network model to obtain a blood vessel segmentation result.

[0111] The use process of the model can be seen from Figure 4D , a model test process schematic diagram is shown. Since the input end of the V-Net neural network model includes two data sources of the coronary angiography map and the blood vessel segmentation image obtained based on the coronary angiography map, the model can learn more rich features in the model training stage, and the learning difficulty is reduced. At the same time, since a large amount of non-target objects are removed in the blood vessel extraction image, the model is introduced with relatively effective prior information, and the accuracy of the model is improved.

[0112] Specifically, the blood vessel segmentation result of the embodiment of the present application can be seen from Figure 4E , the result of blood vessel segmentation only using the coronary angiography map can be seen from Figure 4F , and the corresponding medical gold standard map can be seen from Figure 4G .

[0113] S470, extracting the maximum connected region in the blood vessel segmentation result as a target segmentation result.

[0114] Since the obtained blood vessel segmentation result has a disconnected phenomenon in the coronary artery branch, and the visual effect of the blood vessel segmentation result is poor due to the interference of noise. In order to make the obtained blood vessel segmentation result more beautiful, the noise and non-connected target objects in the blood vessel segmentation result are filtered out by the maximum connected domain method. For details, please refer to Figure 4H The target segmentation result shown in FIG. 8.

[0115] Embodiment five

[0116] Figure 5 FIG. 1 is a structural diagram of a blood vessel segmentation device in an embodiment of the present application. The embodiment of the present application is suitable for the case of separating blood vessel tissues from background regions in a blood vessel image. The device is realized by software and / or hardware, and is specifically configured in a medical imaging device. The medical imaging device can be a CT (Computed Tomography) device, an MRI (Magnetic Resonance Imaging) device, a DSA (Digital Subtraction Angiography) device, and the like.

[0117] As shown in FIG. 2, a blood vessel segmentation device includes an acquisition module 210 and a segmentation module 220. Figure 5 The acquisition module 210 is configured to acquire a blood vessel image to be segmented and a blood vessel distribution map corresponding to the blood vessel image to be segmented.

[0118] The segmentation module 220 is configured to input the blood vessel image to be segmented and the blood vessel distribution map into a pre-trained blood vessel segmentation model to obtain a blood vessel segmentation result.

[0119] The segmentation module 220 is configured to input the blood vessel image to be segmented and the blood vessel distribution map into a pre-trained blood vessel segmentation model to obtain a blood vessel segmentation result.

[0120] The embodiment of the present application acquires a blood vessel image to be segmented and a blood vessel distribution map corresponding to the blood vessel image to be segmented through the acquisition module. The blood vessel image to be segmented and the blood vessel distribution map are input into a pre-set blood vessel segmentation model through the segmentation module to obtain a blood vessel segmentation result. The above technical solution solves the problem that the blood vessel segmentation method in the prior art is not suitable for the segmentation of relatively complex blood vessel images such as coronary images, and improves the blood vessel segmentation accuracy.

[0121] Further, the segmentation module 520, when performing inputting of the blood vessel image to be segmented and the blood vessel distribution map into a pre-set blood vessel segmentation model, includes:

[0122] A fusion unit is configured to fuse the blood vessel image to be segmented and the blood vessel distribution map, and input the fused image into a pre-set blood vessel segmentation model.

[0123] Further, the fusion unit is specifically configured to:

[0124] parallelly input the to-be-segmented blood vessel image and the blood vessel distribution map to different channels of the preset blood vessel segmentation model input, and input the images of each channel into the preset blood vessel segmentation model respectively; or

[0125] multiply or add each pixel point of the to-be-segmented blood vessel image and the blood vessel distribution map to obtain a fused image, and input the fused image into the preset blood vessel segmentation model.

[0126] Further, the acquisition module 510 comprises:

[0127] The extraction unit is configured to process the to-be-segmented blood vessel image by using at least one filtering algorithm to obtain the blood vessel distribution map.

[0128] Further, the extraction unit is specifically configured to:

[0129] calculate eigenvalues of a Hessian matrix of each pixel point in the to-be-segmented blood vessel image;

[0130] process the eigenvalues by using a tubular filter to obtain response values;

[0131] combine the response values corresponding to each pixel point to obtain the blood vessel distribution map.

[0132] Further, the extraction unit is further configured to, before the calculation of the eigenvalues of the Hessian matrix of each pixel point in the to-be-segmented blood vessel image, perform smoothing processing on the to-be-segmented blood vessel image by using at least one target filter, and replace the to-be-segmented blood vessel image with the image after the smoothing processing.

[0133] Correspondingly, the extraction unit, when performing the combination of the target response values corresponding to each pixel point to form the blood vessel distribution map, is specifically configured to:

[0134] compare the blood vessel response values corresponding to the same pixel point based on different target filters to obtain a maximum response value as a target response value;

[0135] combine the target response values corresponding to each pixel point to form the blood vessel distribution map.

[0136] Further, the device further comprises a training module configured to:

[0137] train the blood vessel segmentation model before inputting the to-be-segmented blood vessel image and the blood vessel distribution map into the preset blood vessel segmentation model to obtain a blood vessel segmentation result.

[0138] Further, the training module comprises:

[0139] a historical blood vessel distribution map determination unit configured to determine a corresponding historical blood vessel distribution map according to at least one historical blood vessel image;

[0140] a current blood vessel segmentation result obtaining unit configured to input the historical blood vessel image and the historical blood vessel distribution map into the blood vessel segmentation model to be trained to obtain a current blood vessel segmentation result corresponding to each historical blood vessel image;

[0141] a model prediction parameter adjustment unit configured to adjust a prediction parameter of the blood vessel segmentation model according to an error between the current blood vessel segmentation result and an expected segmentation result.

[0142] Further, the current blood vessel segmentation result obtaining unit comprises:

[0143] an image fusion subunit configured to fuse the historical blood vessel image and the historical blood vessel distribution map and input the fused image into the blood vessel segmentation model to be trained.

[0144] Further, the image fusion subunit is specifically configured to:

[0145] parallel the historical blood vessel image and the historical blood vessel distribution map into different channels of input of the blood vessel segmentation model to be trained, and input the images of each channel into the blood vessel segmentation model to be trained respectively; or

[0146] multiply or add each pixel point of the historical blood vessel image and the historical blood vessel distribution map to obtain a fused image, and input the fused image into the blood vessel segmentation model to be trained.

[0147] Further, the historical blood vessel distribution map determination unit comprises:

[0148] a filtering processing subunit configured to process each historical blood vessel image using at least one filtering algorithm to obtain the historical blood vessel distribution map.

[0149] Further, the filtering processing subunit is specifically configured to:

[0150] calculate eigenvalues of a Hessian matrix of each pixel point in the historical blood vessel image;

[0151] enhance the eigenvalues using a tubular filter to obtain response values;

[0152] combine the response values corresponding to each pixel point to obtain the historical blood vessel distribution map.

[0153] Further, the filtering processing subunit is further configured to:

[0154] Before the eigenvalues of the Hessian matrix of each pixel point in the historical blood vessel image are calculated, at least one target filter is used to smooth the historical blood vessel image, and the smoothed image is used to replace the historical blood vessel image.

[0155] The target filter includes at least one of a Gaussian filter, a linear filter, and a Wiener filter.

[0156] Further, the filter processing subunit, when combining the response values corresponding to each pixel point to obtain the historical blood vessel distribution map, includes:

[0157] The maximum response value is obtained by comparing the blood vessel response values corresponding to the same pixel point based on different target filters, and the maximum response value is used as the target response value.

[0158] The target response values corresponding to each pixel point are combined to form the historical blood vessel distribution map.

[0159] Further, the blood vessel segmentation model is a U-Net convolutional neural network or a V-Net convolutional neural network.

[0160] Further, the device further includes an extraction module configured to:

[0161] After the blood vessel segmentation result is obtained, the largest blood vessel connected graph in the blood vessel segmentation result is extracted to obtain a target blood vessel segmentation map.

[0162] The above-mentioned blood vessel segmentation device can execute the blood vessel segmentation method provided by any embodiment of the present application, and has the corresponding function modules and beneficial effects of executing the blood vessel segmentation method.

[0163] Embodiment six

[0164] Figure 6 is a structural schematic diagram of a medical image device in embodiment six of the present application, which includes an input device 610, an output device 620, a processor 630, and a storage device 640.

[0165] The input device 610 is configured to acquire a blood vessel image to be segmented.

[0166] The output device 620 is configured to display the blood vessel image to be segmented, and also display the blood vessel segmentation result.

[0167] One or more processors 630;

[0168] The storage device 640 is configured to store one or more programs.

[0169] Figure 6The input device 610 in the medical imaging device can be connected with the output device 620, the processor 630 and the storage device 640 through a bus or other means, and the processor 630 and the storage device 640 are also connected through a bus or other means, Figure 6 For example, the medical imaging device is connected through a bus.

[0170] In the embodiment, the processor 630 in the medical imaging device can obtain the to-be-segmented blood vessel image and the blood vessel distribution graph corresponding to the to-be-segmented blood vessel image from the input device 610 or the storage device 640; and input the to-be-segmented blood vessel image and the blood vessel distribution graph into a blood vessel segmentation model pre-trained in the storage device 640 to obtain a blood vessel segmentation result.

[0171] The storage device 640 in the medical imaging device is a computer readable storage medium, which can be used to store one or more programs, such as software programs, computer executable programs and modules, for example, program instructions / modules (for example, the acquisition module 510 and the segmentation module 520 shown in the embodiment of the present application) corresponding to the blood vessel segmentation method in the embodiment of the present application. Figure 5 The processor 630 executes the software programs, instructions and modules stored in the storage device 640, thereby performing various functional applications and data processing of the medical imaging device, that is, implementing the blood vessel segmentation method in the above-mentioned method embodiment.

[0172] The storage device 640 can include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application program required by a function; and the data storage area can store data and the like (for example, the to-be-segmented blood vessel image, the blood vessel distribution graph, the blood vessel segmentation model and the blood vessel segmentation result in the above-mentioned embodiment). In addition, the storage device 640 can include a high-speed random access memory, and can also include a non-volatile memory, for example, at least one magnetic disk storage device, a flash memory device or other non-volatile solid-state storage device. In some examples, the storage device 640 can further include a memory remotely arranged relative to the processor 630, and these remote memories can be connected to a server through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0173] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the program is executed by a blood vessel segmentation device to implement the blood vessel segmentation method provided by the embodiment of the present application, and the method comprises the following steps: obtaining a to-be-segmented blood vessel image and a blood vessel distribution graph corresponding to the to-be-segmented blood vessel image; inputting the to-be-segmented blood vessel image and the blood vessel distribution graph into a blood vessel segmentation model pre-trained to obtain a blood vessel segmentation result.

[0174] Note that the above merely describes preferred embodiments of the present application and the principles of the technology applied. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, modifications and substitutions can be made without departing from the scope of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A method for segmenting blood vessels, characterized in that, include: Acquire the image of the blood vessel to be segmented; The blood vessel image to be segmented is processed by filters of multiple scales to obtain multiple candidate distribution maps; the filters include tubular filters. Calculate the eigenvalues ​​of the Hessian matrix for each pixel in each of the candidate distribution maps; The tubular filter is used to enhance the feature values ​​of each pixel to obtain the response values ​​of each pixel in each candidate distribution map; The blood vessel distribution map is determined based on the response value of each pixel in each candidate distribution map; the blood vessel distribution map is the blood vessel image obtained after extracting blood vessels from the blood vessel image to be segmented. The blood vessel image to be segmented and the blood vessel distribution map are input into a preset blood vessel segmentation model to obtain the blood vessel segmentation result; The determination of the blood vessel distribution map based on the response values ​​of each pixel in each of the candidate distribution maps includes any of the following methods: Compare the response values ​​corresponding to the same pixel in each of the candidate distribution maps; obtain the maximum blood vessel response value as the target response value; combine the target response values ​​corresponding to each pixel to form the blood vessel distribution map; or... Count the number of pixels with response values ​​greater than a set threshold in each candidate distribution map; determine the candidate distribution map with the most pixels as the blood vessel distribution map; or... The mean response value corresponding to the candidate distribution map is determined by the response value corresponding to each pixel. The candidate distribution map with the largest mean response is selected as the blood vessel distribution map.

2. The method according to claim 1, characterized in that, Before calculating the eigenvalues ​​of the Hessian matrix for each pixel in each of the candidate distribution maps, the method further includes: The candidate distribution map is smoothed using at least one target filter to obtain a smoothed candidate partial map.

3. The method according to claim 1, characterized in that, If the blood vessel image to be segmented is a three-dimensional image, then each pixel corresponds to a feature value in three dimensions. Accordingly, for any given pixel, the response value of the pixel is calculated based on the feature values ​​of the three dimensions corresponding to the pixel.

4. The method according to claim 1, characterized in that, The step of inputting the image of the blood vessel to be segmented and the blood vessel distribution map into a preset blood vessel segmentation model to obtain the blood vessel segmentation result includes: The image of the blood vessel to be segmented and the blood vessel distribution map are input in parallel to different channels of the preset blood vessel segmentation model, and the images of each channel are input to the preset blood vessel segmentation model respectively; or... The pixels of the blood vessel image to be segmented and the blood vessel distribution map are multiplied or added together to obtain a fused image, and the fused image is input into the preset blood vessel segmentation model.

5. The method according to any one of claims 1 to 4, characterized in that, The method further includes: Based on the vessel segmentation results, extract the maximum vessel connectivity graph from the vessel segmentation results; The largest vessel connectivity map is determined as the target vessel segmentation map.

6. The method according to any one of claims 1 to 4, characterized in that, Before inputting the image of the blood vessel to be segmented and the blood vessel distribution map into a preset blood vessel segmentation model to obtain the blood vessel segmentation result, the process also includes: The blood vessel segmentation model was trained.

7. The method according to claim 6, characterized in that, The training of the blood vessel segmentation model includes: Based on at least one historical vascular image, determine the corresponding historical vascular distribution map; The historical blood vessel images and the historical blood vessel distribution map are input into the blood vessel segmentation model to be trained to obtain the current blood vessel segmentation result corresponding to each of the historical blood vessel images; The prediction parameters of the blood vessel segmentation model are adjusted based on the error between the current blood vessel segmentation result and the expected segmentation result.

8. A blood vessel segmentation device, characterized in that, include: An acquisition module is used to acquire a blood vessel image to be segmented; process the blood vessel image to be segmented using filters of multiple scales to obtain multiple candidate distribution maps; calculate the eigenvalues ​​of the Hessian matrix of each pixel in each candidate distribution map, wherein the filters include tubular filters; enhance the eigenvalues ​​of each pixel using the tubular filters to obtain the response values ​​of each pixel in each candidate distribution map; determine the blood vessel distribution map based on the response values ​​of each pixel in each candidate distribution map; the blood vessel distribution map is a blood vessel image obtained after extracting blood vessels from the blood vessel image to be segmented. The segmentation module is used to input the blood vessel image to be segmented and the blood vessel distribution map into a preset blood vessel segmentation model to obtain the blood vessel segmentation result; The filter includes a tubular filter. The acquisition module is specifically used to calculate the eigenvalues ​​of the Hessian matrix of each pixel in each of the candidate distribution maps; to enhance the eigenvalues ​​of each pixel using the tubular filter to obtain the response values ​​of each pixel in each of the candidate distribution maps; and to determine the blood vessel distribution map based on the response values ​​of each pixel in each of the candidate distribution maps. The determination of the blood vessel distribution map based on the response values ​​of each pixel in each of the candidate distribution maps includes any of the following methods: Compare the response values ​​corresponding to the same pixel in each of the candidate distribution maps; obtain the maximum blood vessel response value as the target response value; combine the target response values ​​corresponding to each pixel to form the blood vessel distribution map; or... Count the number of pixels with response values ​​greater than a set threshold in each candidate distribution map; determine the candidate distribution map with the most pixels as the blood vessel distribution map; or... The mean response value corresponding to the candidate distribution map is determined by the response value corresponding to each pixel. The candidate distribution map with the largest mean response is selected as the blood vessel distribution map.

9. A medical imaging device, comprising an input device and an output device, characterized in that, Also includes: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a blood vessel segmentation method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements a blood vessel segmentation method as described in any one of claims 1-7.

Images