Liver portal vein system thrombus identification method and device

Abstract

The invention discloses a hepatic portal vein system thrombus identification method and device, and relates to the technical field of medical image processing, and the method comprises the steps: inputting a to-be-processed CTA image into a rough segmentation model, and outputting a blood vessel cavity; the vascular cavity comprises a blood vessel and thrombus in the blood vessel; the rough segmentation model is obtained by training an nnunet network; carrying out statistics on HU value distribution of the pixels in the blood vessel cavity, and determining a segmentation threshold value for segmenting thrombus in combination with a human liver tissue standard HU value; based on a density clustering algorithm, determining a clustering center coordinate of the thrombus in the blood vessel; taking the clustering center coordinate as a seed point, carrying out region growth according to the similarity between a pixel HU value in a blood vessel cavity and the segmentation threshold value until a blood vessel boundary is encountered or no to-be-grown pixel point exists, and obtaining a fine segmentation result; the fine segmentation result comprises each pixel where the thrombus is located. The thrombus recognition accuracy of the hepatic portal vein system can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method and device for identifying thrombus in the portal vein system. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention that are recited in the claims. No statement herein is admitted to be prior art by virtue of its inclusion in this section.

[0003] Portal vein system thrombosis (PVST) refers to thrombosis of the main portal vein and / or its left and right branches, with or without thrombosis of the mesenteric and splenic veins. Acute PVST can lead to serious adverse outcomes such as mesenteric ischemia and even intestinal necrosis. Chronic PVST can lead to portal vein occlusion or cavernous degeneration, with subsequent portal hypertension. Therefore, early and accurate identification of thrombosis is crucial.

[0004] In existing technologies, human thrombus identification is mostly based on image processing, combined with other data such as pressure measurement for judgment. However, the human abdominal organs are stacked on top of each other and there is no bony structure to stabilize the organs. The relative positions of abdominal organs in different people vary greatly. The vascular network of the portal vein system is intertwined with the hepatic artery network and the hepatic vein network. Thrombi in the portal vein system are small structures attached to the vascular walls of the portal vein system. Identifying such thrombi is very difficult. Existing thrombus identification methods cannot accurately identify thrombi in the portal vein system. Currently, thrombi still need to be identified with the naked eye, with an accuracy of only 40% to 50%. Therefore, the accuracy of existing portal vein system thrombus identification needs to be improved. Summary of the Invention

[0005] An embodiment of the present invention provides a method for identifying thrombosis in the portal vein system to improve the accuracy of identifying thrombosis in the portal vein system. The method includes:

[0006] Inputting a CTA (Computed Tomography Angiography) image to be processed into a pre-trained coarse segmentation model to output a vascular lumen in the CTA image to be processed; the vascular lumen includes the blood vessels and thrombi within the blood vessels; the coarse segmentation model is trained on a nnunet (Self-adapting Framework for U-Net-Based Medical Image Segmentation) network using historical images of the human portal vein system with annotated blood vessels and thrombi;

[0007] Counting the HU value distribution of the pixels in the blood vessel lumen, and determining a segmentation threshold for segmenting the thrombus based on a standard HU value of human liver tissue;

[0008] Determining cluster center coordinates of the thrombus in the blood vessel based on a density clustering algorithm;

[0009] Taking the cluster center coordinates as the seed point, region growing is performed according to the similarity between the pixel HU value in the vascular cavity and the segmentation threshold until the vascular boundary is encountered or there are no pixels to be grown, and a fine segmentation result is obtained; the fine segmentation result includes each pixel where the thrombus is located.

[0010] An embodiment of the present invention further provides a device for identifying thrombus in the portal vein system, for improving the accuracy of identifying thrombus in the portal vein system. The device comprises:

[0011] A coarse segmentation processing module is configured to input the CTA image to be processed into a pre-trained coarse segmentation model and output the vascular lumen in the CTA image to be processed; the vascular lumen includes the blood vessels and the thrombus in the blood vessels; the coarse segmentation model is trained on the nnunet network using historical images of the human portal vein system with annotated blood vessels and thrombus;

[0012] A grayscale processing module, configured to calculate the HU value distribution of pixels within the blood vessel lumen and determine a segmentation threshold for thrombus segmentation based on the standard HU value of human liver tissue;

[0013] A fine segmentation processing module is used to determine the cluster center coordinates of the thrombus in the blood vessel based on a density clustering algorithm; using the cluster center coordinates as a seed point, regional growth is performed according to the similarity between the pixel HU value in the blood vessel cavity and the segmentation threshold until the blood vessel boundary is encountered or there are no pixels to be grown, thereby obtaining a fine segmentation result; the fine segmentation result includes each pixel where the thrombus is located.

[0014] An embodiment of the present invention further provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for identifying portal vein system thrombosis when executing the computer program.

[0015] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for identifying thrombosis in the portal vein system is implemented.

[0016] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the computer program implements the above-mentioned method for identifying thrombosis in the portal vein system.

[0017] In an embodiment of the present invention, a CTA image to be processed is input into a pre-trained coarse segmentation model, and a vascular cavity in the CTA image to be processed is output; the vascular cavity includes blood vessels and thrombi in the blood vessels; the coarse segmentation model is obtained by training the nnunet network using historical images of the human portal vein system with blood vessels and thrombi annotated; the HU value distribution of pixels in the vascular cavity is statistically analyzed, and the segmentation threshold for segmenting the thrombus is determined in combination with the standard HU value of human liver tissue; the cluster center coordinates of the thrombus in the blood vessel are determined based on a density clustering algorithm; the cluster center coordinates are used as seed points, and region growing is performed according to the similarity between the HU values ​​of the pixels in the vascular cavity and the segmentation threshold until a blood vessel boundary is encountered or there are no pixels to be grown, thereby obtaining a fine segmentation result; the fine segmentation result includes each pixel where the thrombus is located. In the embodiment of the present invention, the nnunet network is first used to automatically and roughly segment the blood vessels and thrombi to obtain a complete vascular cavity. The segmentation threshold is determined based on the HU value, and the segmentation threshold is used to construct a region growing criterion for secondary precise segmentation and identification of the thrombus. Compared with the existing methods, the thrombus characteristics of the portal vein system are targeted and the thrombus identification of the portal vein system is performed, which greatly improves the accuracy of the identification of the portal vein system thrombus. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0019] Figure 1 Schematic diagram of the process of the method for identifying thrombosis in the portal vein system according to an embodiment of the present invention;

[0020] Figure 2 Schematic diagram of vascular thrombus segmentation in an embodiment of the present invention;

[0021] Figure 3 is a histogram of HU value distribution of pixels in the blood vessel cavity in an embodiment of the present invention;

[0022] Figure 4 This is a schematic diagram of the thrombus fine segmentation result in an embodiment of the present invention;

[0023] Figure 5 Schematic diagram of a device for identifying thrombus in the portal vein system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0025] First, the technical terms involved in the embodiments of the present invention are explained.

[0026] The Dice metric, also known as the Dice coefficient or Sørensen-Dice coefficient, is a statistic used to measure the similarity between two sample sets in the fields of machine learning and image processing, especially in binary segmentation tasks. It is often used to evaluate the similarity between model predictions and true labels.

[0027] HU value: The Hounsfield Unit is a standardized unit used to quantify tissue density in CT images. It is named after the British scientist Godfrey Hounsfield (co-inventor of CT). Essentially, it converts the X-ray attenuation coefficients of different tissues into a unified numerical scale. Higher HU values ​​indicate brighter (whiter) CT images; lower HU values ​​indicate darker (blacker) images.

[0028] nnunet: nnU-Net is a fully automatic medical image segmentation framework based on the classic U-Net architecture, which can automatically adjust hyperparameters according to dataset properties.

[0029] The existing technology has poor accuracy in automatically identifying thrombi in the portal vein system. Typically, thin-layer abdominal venous phase CTA images are used to visually identify thrombi, with an accuracy of only 40-50%. Therefore, an embodiment of the present invention proposes a method for identifying thrombi in the portal vein system to improve the accuracy of automatic identification of thrombi in the portal vein system.

[0030] Figure 1 FIG. 1 is a flow chart of a method for identifying thrombosis in the portal vein system according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0031] Step 101: Input the CTA image to be processed into a pre-trained coarse segmentation model, and output the vascular lumen in the CTA image to be processed; the vascular lumen includes the blood vessels and the thrombus in the blood vessels; the coarse segmentation model is trained on the nnunet network using historical images of the human portal vein system with annotated blood vessels and thrombus;

[0032] Step 102: Counting the HU value distribution of the pixels in the blood vessel cavity, and determining a segmentation threshold for thrombus segmentation based on the standard HU value of human liver tissue;

[0033] Step 103: determining the cluster center coordinates of the thrombus in the blood vessel based on a density clustering algorithm;

[0034] Step 104: Using the cluster center coordinates as seed points, perform region growing based on the similarity between the HU values ​​of the pixels in the vascular cavity and the segmentation threshold until the vascular boundary is encountered or there are no more pixels to be grown, thereby obtaining a fine segmentation result; the fine segmentation result includes each pixel where the thrombus is located.

[0035] The following is a detailed explanation of the method for identifying thrombosis in the portal vein system according to an embodiment of the present invention.

[0036] Part 1: Blood vessel and thrombus segmentation.

[0037] During specific implementation, a rough segmentation model is pre-trained, and the CTA image to be processed is processed using the rough segmentation model to obtain the vascular cavity in the CTA image to be processed.

[0038] In the embodiment, the rough segmentation model is obtained by training a convolutional network (unet) for biomedical image segmentation using historical images of the human portal vein system with annotated blood vessels and thrombi.

[0039] In one embodiment, the coarse segmentation model can be trained as follows:

[0040] The training set is constructed using historical images of the human portal vein system with annotated blood vessels and thrombi.

[0041] Using the training set, 5-fold cross validation and adam (Adaptive Moment Estimation) gradient optimization method are used for training to obtain a trained rough segmentation model.

[0042] For example, a large number of abdominal venous phase CTA images were collected, first automatically labeled by a machine, then reviewed and corrected by professionals, and finally sample data was obtained through morphological opening and closing operations. The sample data included positive data and negative data, and the training set and validation set were split according to an 8:2 ratio. The training was performed on a V100 GPU with a 500 epoch, a deep supervision mode, a learning rate of 0.001, and Adam gradient optimization.

[0043] Specifically, there were 84 positive cases and 7 negative cases. The test set was divided into a 22-case test set for our hospital and a 25-case test set for site A and site B, respectively. The dice index for our hospital data was 0.71, the dice index for the site A test set was 0.682, and the dice index for the site B test set was 0.643.

[0044] In the prior art, it is difficult to collect imaging data on thrombosis in the portal vein system, and the abdominal background is complex. Accurately labeling thrombi layer by layer on thin-layer data requires very high professional skills from the labelers. To solve this problem, in one embodiment of the present invention, a rough segmentation model is obtained through federated learning training. During implementation, a federated learning architecture is constructed, including a server and various clients. The server is used to collect model update parameters from each client, and each client trains the model using the local data it holds. Through federated learning training, the accuracy and generalization ability of the model can be improved.

[0045] Step 101 inputs the CTA image to be processed into a pre-trained coarse segmentation model to output the vascular lumen in the CTA image to be processed. This may include: inputting the CTA image to be processed into the pre-trained coarse segmentation model to output the blood vessels and thrombi therein; and merging the blood vessels and thrombi therein using a morphological closing method to obtain the vascular lumen in the CTA image to be processed. Filling gaps and smoothing boundaries using the dilation and erosion operations in the morphological closing method helps improve the quality of subsequent thrombus segmentation and reduce noise.

[0046] Figure 2 FIG. 1 is a schematic diagram of vascular thrombus segmentation in an embodiment of the present invention, as shown in FIG. Figure 2 As shown in the figure, the left image shows the CTA image to be processed, and the right image shows the identified blood vessel lumen. Green indicates the blood vessel, and red indicates the thrombus within the vessel. It should be noted that during specific processing, the identified thrombus within a vessel may include multiple small thrombi in multiple locations within the vessel, which are shown as multiple small thrombus areas in the figure.

[0047] The second part is the pixel distribution processing inside the lumen.

[0048] In step 102, the HU value distribution of the pixels in the blood vessel cavity is counted, and the segmentation threshold for segmenting the thrombus is determined in combination with the standard HU value of human liver tissue.

[0049] In the embodiment of the present invention, each pixel of the CTA image to be processed has a HU value. Considering the standard HU value of human liver tissue, a segmentation threshold for segmenting thrombus is designed to perform fine segmentation on the vascular lumen processed by the rough segmentation model.

[0050] In one embodiment, counting the HU value distribution of pixels within the blood vessel lumen and combining it with the standard HU value of human liver tissue to determine the segmentation threshold for thrombus segmentation may include:

[0051] A histogram is constructed using the HU values ​​of the vascular lumen pixels; the horizontal axis of the histogram is the HU value, and the vertical axis is the number of pixels at each HU value;

[0052] If there are two peaks in the histogram, the HU value with the smaller horizontal axis value in the two peaks is recorded as the segmentation threshold;

[0053] If there is no double peak in the histogram, it is determined that there is no thrombus in the CTA image to be processed.

[0054] In the embodiment of the present invention, the data for segmenting the portal vein system blood vessels are venous phase CTA images, that is, the images are collected during the period when contrast agent is injected and the venous blood vessels are visualized. The characteristics of blood vessels are high-density shadows, and thrombi are emboli attached to the blood vessel walls, which are low-density shadows. If the blood vessel segmentation is not accurate, such as if the blood vessel boundary is expanded several times, other non-vascular tissues will also be included. The mis-segmented non-vascular part is the liver tissue part. When counting the pixels in the lumen, the blood vessel label is three-dimensionally corroded. Since the thrombus is located inside the blood vessel lumen, this operation will not have a significant effect on the thrombus. The grayscale value in the lumen is counted. Figure 3 is the HU value distribution histogram of the pixels in the blood vessel cavity in the embodiment of the present invention, as shown in FIG. Figure 3 As shown in the figure, the horizontal axis represents the HU value, and the vertical axis represents the number of pixels. If the grayscale distribution shows a clear bimodal pattern, with one peak corresponding to a grayscale value (unit: HU) higher than the liver tissue and the other lower than the liver tissue, and the liver tissue HU value on CTA images is 70-80 HU, then a thrombus is present in the lumen; otherwise, it is absent. The HU values ​​corresponding to the two peaks are obtained, and the lower HU value is used as the thrombus segmentation threshold.

[0055] Part 3: Fine segmentation of thrombus.

[0056] In step 103 , the cluster center coordinates of the thrombus in the blood vessel are determined based on a density clustering algorithm, wherein the density clustering algorithm includes a DBSCAN (Density-Based Spatial Clustering with Noise) algorithm.

[0057] During implementation, the identified cluster center coordinates may include cluster center coordinates of multiple thrombi.

[0058] In step 104, the cluster center coordinates are used as seed points, and region growing is performed based on the similarity between the HU values ​​of the pixels in the vascular cavity and the segmentation threshold until the vascular boundary is encountered or there are no more pixels to be grown, thereby obtaining a fine segmentation result; the fine segmentation result includes each pixel where the thrombus is located.

[0059] For example, the cluster center coordinates are used as seed points, and the difference between the HU value of the pixel in the blood vessel cavity and the segmentation threshold is less than 10 HU values ​​as the growth criterion. Pixel points are searched in 26 neighboring directions until the blood vessel boundary is encountered or there are no pixels to be grown, thereby obtaining a fine segmentation result.

[0060] The three elements of the region growing algorithm are: seed point, growth criterion, and stopping condition. In the embodiment, the DBSCAN algorithm is used to obtain the cluster center coordinates of the thrombus. The thrombus can be single or multiple, and the cluster center corresponds to a single or multiple ones, which are set as seed points. The corresponding seed points are also single or multiple. Growth criterion: similarity of pixel grayscale values. The difference between the grayscale value of the pixel in the lumen and the segmentation threshold is <10HU as the growth criterion, and similar pixels are searched in 26 neighborhood directions. Stopping condition: encountering the blood vessel boundary or there are no seed points to be grown.

[0061] Figure 4 Schematic diagram of thrombus fine segmentation results in an embodiment of the present invention, refer to Figure 4 There are 4 cluster centers, and 4 thrombi are identified with the 4 cluster centers as the centers. The fine segmentation result can be, for example, marking pixels as 1 or 0 to show whether it is a thrombus, 1 represents a thrombus area, and 0 represents a non-thrombus area.

[0062] In one embodiment, after obtaining the fine segmentation result, the method may further include performing a closing operation on the fine segmentation result to obtain a final thrombus segmentation result.

[0063] For example, after preliminary segmentation using nnunet, two steps of region growing based on lumen grayscale statistics are used to obtain the precise segmentation result of the thrombus. Combined with post-processing of a 3D morphological one-scale closing operation, the pixel gaps caused by the interference of white noise on region growing are compensated to obtain the final thrombus segmentation result.

[0064] Afterwards, based on the final thrombus segmentation results, thrombus quantitative parameter results such as vascular volume, vascular surface area, vascular curvature, thrombus volume, thrombus surface area, vascular thrombus volume ratio, maximum thrombus cross-section ratio, and average thrombus cross-section ratio can be further calculated.

[0065] In summary, the embodiments of the present invention, based on the open-source framework nnunet, automatically segment blood vessels and thrombi. By leveraging the physiological properties of thrombi attached to blood vessels, a morphological closing operation is employed to merge the thrombi and blood vessels, forming a complete vascular lumen. Within the vascular lumen, the grayscale value of the thrombus is determined using histogram statistics. This grayscale value serves as a growth criterion, and the DBSCAN clustering results of the initially segmented thrombi are used as seed points. The thrombi are then refined using a region growing method, resulting in accurate thrombus segmentation. This method achieves precise identification of thrombi in the portal vein system.

[0066] The present invention also provides a device for identifying thrombi in the portal vein system, as described in the following embodiments. Because the principles of this device are similar to those of the method for identifying thrombi in the portal vein system, the implementation of this device can be referenced to the implementation of the method for identifying thrombi in the portal vein system, and any repetitions will not be repeated.

[0067] Figure 5 FIG. 1 is a schematic diagram of a device for identifying thrombus in the portal vein system according to an embodiment of the present invention. Figure 5 As shown, the apparatus 500 includes:

[0068] The coarse segmentation processing module 501 is configured to input the CTA image to be processed into a pre-trained coarse segmentation model and output the vascular lumen in the CTA image to be processed; the vascular lumen includes the blood vessels and the thrombus in the blood vessels; the coarse segmentation model is trained on the nnunet network using historical images of the human portal vein system with annotated blood vessels and thrombus;

[0069] Grayscale processing module 502, for counting the HU value distribution of pixels in the blood vessel lumen and determining a segmentation threshold for thrombus segmentation based on the standard HU value of human liver tissue;

[0070] The fine segmentation processing module 503 is used to determine the cluster center coordinates of the thrombus in the blood vessel based on the density clustering algorithm; using the cluster center coordinates as the seed point, regional growth is performed according to the similarity between the pixel HU value in the blood vessel cavity and the segmentation threshold until the blood vessel boundary is encountered or there are no pixels to be grown, thereby obtaining a fine segmentation result; the fine segmentation result includes each pixel where the thrombus is located.

[0071] In one embodiment, the coarse segmentation model is trained as follows:

[0072] The training set is constructed using historical images of the human portal vein system with annotated blood vessels and thrombi.

[0073] Using the training set, 5-fold cross validation and Adam gradient optimization method are used for training to obtain a trained rough segmentation model.

[0074] In one embodiment, the rough segmentation processing module 501 is specifically configured to:

[0075] The CTA image to be processed is input into a pre-trained rough segmentation model to output blood vessels and thrombi in the blood vessels;

[0076] The morphological closing operation method is used to merge the blood vessels and the thrombus in the blood vessels to obtain the vascular lumen in the CTA image to be processed.

[0077] In one embodiment, the grayscale processing module 502 is specifically configured to:

[0078] A histogram is constructed using the HU values ​​of the pixels in the blood vessel cavity; the horizontal axis of the histogram is the HU value, and the vertical axis is the number of pixels with each HU value;

[0079] If there are two peaks in the histogram, the HU value with the smaller horizontal axis value in the two peaks is recorded as the segmentation threshold;

[0080] If there is no double peak in the histogram, it is determined that there is no thrombus in the CTA image to be processed.

[0081] In one embodiment, the fine segmentation processing module 503 is specifically configured to:

[0082] The cluster center coordinates are used as seed points, and the difference between the HU value of the pixel in the vascular cavity and the segmentation threshold is less than 10 HU values ​​as the growth criterion. Pixel points are searched in 26 neighboring directions until the vascular boundary is encountered or there are no pixels to be grown, and a fine segmentation result is obtained.

[0083] In one embodiment, the apparatus 500 further includes:

[0084] The final result determination module is used to perform a closing operation on the fine segmentation result after the fine segmentation processing module 503 obtains the fine segmentation result, so as to obtain the final thrombus segmentation result.

[0085] An embodiment of the present invention further provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for identifying portal vein system thrombosis when executing the computer program.

[0086] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for identifying thrombosis in the portal vein system is implemented.

[0087] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the computer program implements the above-mentioned method for identifying thrombosis in the portal vein system.

[0088] In an embodiment of the present invention, a CTA image to be processed is input into a pre-trained coarse segmentation model, and a vascular cavity in the CTA image to be processed is output; the vascular cavity includes blood vessels and thrombi in the blood vessels; the coarse segmentation model is obtained by training the nnunet network using historical images of the human portal vein system with blood vessels and thrombi annotated; the HU value distribution of pixels in the vascular cavity is statistically analyzed, and the segmentation threshold for segmenting the thrombus is determined in combination with the standard HU value of human liver tissue; the cluster center coordinates of the thrombus in the blood vessel are determined based on a density clustering algorithm; the cluster center coordinates are used as seed points, and region growing is performed according to the similarity between the HU values ​​of the pixels in the vascular cavity and the segmentation threshold until a blood vessel boundary is encountered or there are no pixels to be grown, thereby obtaining a fine segmentation result; the fine segmentation result includes each pixel where the thrombus is located. In the embodiment of the present invention, the nnunet network is first used to automatically and roughly segment the blood vessels and thrombi to obtain a complete vascular cavity. The segmentation threshold is determined based on the HU value, and the segmentation threshold is used to construct a region growing criterion for secondary precise segmentation and identification of the thrombus. Compared with the existing methods, the thrombus characteristics of the portal vein system are targeted and the thrombus identification of the portal vein system is performed, which greatly improves the accuracy of the identification of the portal vein system thrombus.

[0089] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0091] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0093] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying thrombosis in the portal vein system, characterized in that: include: Inputting a computed tomography angiography (CTA) image to be processed into a pre-trained coarse segmentation model to output a vascular lumen in the CTA image to be processed; the vascular lumen includes the blood vessel and the thrombus in the blood vessel; the coarse segmentation model is trained on a nnunet network, an adaptive medical image segmentation framework based on a U-shaped network structure, using historical images of the human portal vein system with annotated blood vessels and thrombus; Counting the HU value distribution of the pixels in the blood vessel lumen, and determining a segmentation threshold for segmenting the thrombus based on a standard HU value of human liver tissue; Determining cluster center coordinates of the thrombus in the blood vessel based on a density clustering algorithm; Taking the cluster center coordinates as the seed point, region growing is performed according to the similarity between the pixel HU value in the vascular cavity and the segmentation threshold until the vascular boundary is encountered or there are no pixels to be grown, and a fine segmentation result is obtained; the fine segmentation result includes each pixel where the thrombus is located.

2. The method according to claim 1, wherein The coarse segmentation model is trained as follows: The training set is constructed using historical images of the human portal vein system with annotated blood vessels and thrombi. Using the training set, 5-fold cross validation and adaptive moment estimation adam gradient optimization method are used for training to obtain a trained rough segmentation model.

3. The method according to claim 2, wherein The CTA image to be processed is input into the pre-trained coarse segmentation model, and the vascular lumen in the CTA image to be processed is output, including: The CTA image to be processed is input into a pre-trained rough segmentation model to output blood vessels and thrombi in the blood vessels; The morphological closing operation method is used to merge the blood vessels and the thrombus in the blood vessels to obtain the vascular lumen in the CTA image to be processed.

4. The method according to claim 1, wherein Counting the HU value distribution of the pixels in the blood vessel lumen and combining it with the standard HU value of human liver tissue to determine a segmentation threshold for thrombus segmentation, including: A histogram is constructed using the HU values ​​of the vascular lumen pixels; the horizontal axis of the histogram is the HU value, and the vertical axis is the number of pixels at each HU value; If there are two peaks in the histogram, the HU value with the smaller horizontal axis value in the two peaks is recorded as the segmentation threshold; If there is no double peak in the histogram, it is determined that there is no thrombus in the CTA image to be processed.

5. The method according to claim 1, wherein Using the cluster center coordinates as the seed point, region growing is performed based on the similarity between the HU values ​​of the pixels in the vascular cavity and the segmentation threshold until the vascular boundary is encountered or there are no more pixels to be grown, and a fine segmentation result is obtained, including: The cluster center coordinates are used as seed points, and the difference between the HU value of the pixel in the vascular cavity and the segmentation threshold is less than 10 HU values ​​as the growth criterion. Pixel points are searched in 26 neighboring directions until the vascular boundary is encountered or there are no pixels to be grown, and a fine segmentation result is obtained.

6. The method according to claim 5, wherein After obtaining the fine segmentation results, it also includes: The fine segmentation results are closed and processed to obtain the final thrombus segmentation results.

7. A device for identifying thrombosis in the portal vein system, characterized in that: include: A coarse segmentation processing module is configured to input the CTA image to be processed into a pre-trained coarse segmentation model and output the vascular lumen in the CTA image to be processed; the vascular lumen includes the blood vessels and the thrombus in the blood vessels; the coarse segmentation model is trained on the nnunet network using historical images of the human portal vein system with annotated blood vessels and thrombus; A grayscale processing module, configured to calculate the HU value distribution of pixels within the blood vessel lumen and determine a segmentation threshold for thrombus segmentation based on the standard HU value of human liver tissue; A fine segmentation processing module is used to determine the cluster center coordinates of the thrombus in the blood vessel based on a density clustering algorithm; using the cluster center coordinates as a seed point, regional growth is performed according to the similarity between the pixel HU value in the blood vessel cavity and the segmentation threshold until the blood vessel boundary is encountered or there are no pixels to be grown, thereby obtaining a fine segmentation result; the fine segmentation result includes each pixel where the thrombus is located.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images