A CT medical image segmentation method, device and storage medium

Abstract

The application discloses a CT medical image segmentation method and device and a storage medium, and the method comprises the following steps: constructing a human standard model, virtually slicing organs and / or tissues in the human standard model to obtain a CT-like semantic segmentation graph, and selecting a corresponding target CT image in a target individual; constructing a difference quantization system, adjusting the CT-like semantic segmentation graph by using a finite element method to obtain a new CT-like semantic segmentation graph; determining whether the CT-like semantic segmentation graph is adjusted by using the difference quantization system; if yes, updating the human standard model by using the new CT-like semantic segmentation graph, and determining whether the corresponding organs and / or tissues are adjusted; if yes, obtaining a digital twin of the organs and / or tissues corresponding to the target CT image; and traversing the human standard model to complete evolution of the human standard model to the target individual, virtually slicing at a required position, and obtaining a segmentation result of the CT medical image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building structures, and in particular to a CT medical image segmentation method, device and storage medium. BACKGROUND

[0002] Medical image data is an important part of medical data, and the diagnosis of medical images currently mainly relies on subjective analysis by artificial. In recent years, with the outbreak of the third artificial intelligence revolution, using deep learning method to process medical images has become the mainstream research.

[0003] The current mainstream CT medical image segmentation method attempts to input the original image and the image with fine annotation to let the neural network learn the segmentation strategy of organs and tissues independently, and ignores whether the doctor has enough cognition of the three-dimensional shape and spatial relationship of human organs and tissues. In this case, even the best segmentation method performs much worse than experienced radiologists on medical images with blurred boundaries.

[0004] Therefore, it is urgent to propose an effective means to integrate the knowledge of the three-dimensional shape and spatial relationship of human organs and tissues into the CT medical image segmentation algorithm to realize more accurate organ and tissue segmentation. SUMMARY

[0005] The purpose of the present application is to provide a CT medical image segmentation method, device and storage medium to solve one or more technical problems existing in the prior art and at least provide a beneficial choice or create conditions.

[0006] The solution to the technical problem of the present application is to provide a CT medical image segmentation method, device and storage medium.

[0007] According to the first aspect of the present application, a CT medical image segmentation method is provided, comprising:

[0008] According to the obtained original CT image and fine annotated CT image, a human standard model and a difference quantization system are constructed, the organs and / or tissues in the human standard model are virtually sliced to obtain a CT-like semantic segmentation map, and a target CT image corresponding to the target individual is selected;

[0009] According to the CT-like semantic segmentation map and the target CT image, a difference quantization system is constructed, and the CT-like semantic segmentation map is adjusted by using the finite element method to obtain a new CT-like semantic segmentation map;

[0010] According to the CT-like semantic segmentation map and the new CT-like semantic segmentation map, whether the CT-like semantic segmentation map is adjusted is judged by using the difference quantization system;

[0011] If yes, the human standard model is updated by using the new CT-like semantic segmentation map, and it is determined whether the corresponding organs and / or tissues are adjusted or not;

[0012] If yes, the digital twin of the organs and / or tissues corresponding to the target CT image is obtained;

[0013] The human standard model is traversed, the evolution of the human standard model to the target individual is completed, and a virtual slice is performed at a required position to obtain a segmentation result of the CT medical image.

[0014] Further, the obtaining process of the CT-like semantic segmentation map specifically comprises:

[0015] In the human standard model, the direction of the foot pointing to the head is the z-axis, the intersection of the organs and / or tissues in the human standard model and the plane perpendicular to the z-axis is calculated to obtain the corresponding slice contour;

[0016] The inside and outside of the slice contour are distinguished to obtain the CT-like semantic segmentation map.

[0017] Further, the obtaining process of the new CT-like semantic segmentation map specifically comprises:

[0018] In the CT-like semantic segmentation map, one contour point in the organ contour or the tissue contour is selected as a moving contour point, a tangent line of the moving contour point is constructed, and a tangent direction is determined;

[0019] The first contour point is fixed, the moving contour point is moved along the tangent direction, and the second contour point moves with the moving contour point under the constraint of the finite element method to obtain a moving position, the CT-like semantic segmentation map is updated according to the moving position to obtain a new CT-like semantic segmentation map;

[0020] The distance from the first contour point to the moving contour point is greater than the distance from the second contour point to the moving contour point.

[0021] Further, the construction process of the difference quantification system specifically comprises:

[0022] An image semantic generation network is constructed, and the image semantic generation network is trained according to the original CT image and the fine-labeled CT image, wherein the image semantic generation network is composed of a generation sub-network and a discrimination sub-network;

[0023] The discrimination sub-network after training is used to quantify the difference value between the target CT image and the new CT-like semantic segmentation map or the CT-like semantic segmentation map.

[0024] Further, the judgment process of whether to adjust the CT-like semantic segmentation map specifically comprises:

[0025] Input the new CT-like semantic segmentation map and the target CT image into the difference quantization system to obtain a first difference value; input the CT-like semantic segmentation map and the target CT image into the difference quantization system to obtain a second difference value.

[0026] Compare the first and second difference values, and determine whether to adjust the CT-like semantic segmentation map based on the comparison results.

[0027] Furthermore, the determination of whether the corresponding organ and / or tissue has completed the adjustment specifically includes:

[0028] Determine whether the first difference value is less than the set threshold;

[0029] If so, then based on the target CT image, determine whether all adjustments to the corresponding organ and / or tissue have been completed;

[0030] If so, the corresponding organ and / or tissue is considered to have completed the adjustment.

[0031] Furthermore, when not all adjustments to the corresponding organs and / or tissues have been completed, this specifically includes:

[0032] Fix all contour points in the organ and / or tissue contours in the new CT-like semantic segmentation image, and adjust the remaining contour points of the corresponding organs and / or tissues using the three-dimensional spatial finite element method.

[0033] Select a new target CT image and adjust it again.

[0034] Furthermore, determining whether the first difference value is less than the set threshold also includes:

[0035] When the first difference value is greater than the set threshold, the organs and / or tissues in the human standard model are virtually sliced ​​again and adjusted again.

[0036] According to an embodiment of a second aspect of the present invention, an electronic device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement a CT medical image segmentation method as described in an embodiment of the first aspect of the present invention.

[0037] According to an embodiment of a third aspect of the present invention, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements a CT medical image segmentation method as described in an embodiment of the first aspect of the present invention.

[0038] The beneficial effects of this invention are as follows: This invention utilizes digital twin technology to bring knowledge of the three-dimensional morphology and spatial location of human organs and tissues to computer-aided medical image segmentation methods, making its information content as comparable as possible to that of medical experts. By employing finite element analysis and semantic segmentation, the segmentation problem of CT images is transformed into a problem of gradually approximating the CT image's description of the human body using a digital spatial human body model. This cleverly combines the aforementioned knowledge with CT image information to achieve CT image segmentation. This differs significantly from existing technologies that use finely annotated CT medical images to train neural networks for medical image segmentation, solving the problem of unclear boundaries between organs and tissues in CT images caused by insufficient scanning resolution or similar physical properties. Attached Figure Description

[0039] Figure 1 This is a schematic flowchart of a CT medical image segmentation method provided by the present invention;

[0040] Figure 2 This is a schematic diagram illustrating the method for obtaining CT-like semantic segmentation maps from a standard human body model, as provided by the present invention for CT medical image segmentation.

[0041] Figure 3 This is a schematic diagram illustrating a new CT-like semantic segmentation map obtained through a CT medical image segmentation method provided by the present invention; wherein:

[0042] Figure (a) is a schematic diagram of the selection of moving contour points and tangent calculation;

[0043] Figure (b) is a schematic diagram of the movement of the contour point along the tangent direction;

[0044] Figure (c) is a schematic diagram of the new CT-like semantic segmentation map. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and should not be construed as limiting the scope of the invention.

[0046] It should be noted that although functional modules are divided in the system diagram, in some cases, the steps shown or described may be executed in a different order than the module division or flowchart shown in the system. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0047] In the description of this invention, it should be noted that, unless otherwise explicitly defined, terms such as "setting," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0048] Reference Figures 1 to 3 According to an embodiment of the first aspect of the present invention, a CT medical image segmentation method includes the following steps:

[0049] S100: Based on the acquired original CT images and finely annotated CT images, construct a human standard model and a differential quantification system, perform virtual slicing of organs and / or tissues in the human standard model to obtain a CT-like semantic segmentation map, and select the corresponding target CT image in the target individual.

[0050] In this embodiment, three-dimensional models of male and female human organs and tissues can be obtained from publicly available data, and a large number of finely annotated CT images can be obtained from publicly available datasets from hospitals. The publicly available datasets include: original CT images and finely annotated CT images.

[0051] Guided by meticulously annotated CT images, the three-dimensional models of the aforementioned human organs and tissues were refined and adjusted one by one using 3D modeling software to construct a standard human body model.

[0052] We obtain raw CT images and finely annotated CT images from public datasets, and construct a differential quantification system using these raw and finely annotated CT images.

[0053] In a standard human body model, the slice outlines of organs and tissues are obtained by calculating the intersection points of the organs and tissues with a plane perpendicular to the z-axis. By distinguishing between the inside and outside of the slice outlines, a CT-like semantic segmentation map is obtained.

[0054] That is, to perform virtual slicing of organs and / or tissues in a standard human body model. Since some tissues and organs in the human body are on the same plane, while others are not, virtual slices can be organs, tissues, or both.

[0055] Select the actual CT image of the target human individual at the corresponding layer, that is, select the corresponding target CT image. In other words, the organs and / or tissues in the target CT image are the organs and / or tissues displayed in the CT semantic segmentation map.

[0056] S200: Based on the CT-like semantic segmentation map and the target CT image, a difference quantization system is constructed. Using the finite element method, the CT-like semantic segmentation map is adjusted to obtain a new CT-like semantic segmentation map.

[0057] In this embodiment, a difference quantification system is constructed between virtual slices of a standard human body model and the target CY image. Specifically, a difference quantification system is established using a CT-like semantic segmentation map and the target CT image.

[0058] By using the discriminant subnetwork of the image semantic generation network, the goal is to quantify the difference between the semantically segmented CT image and the target CT image. The semantically segmented CT image is used as the input to the image semantic generation network, and the target CT image is used as the output. After training the network, the discriminant subnetwork gains the ability to determine whether the target CT image and the semantically segmented CT image match, and its output is the numerical difference between the two.

[0059] By using the finite element method, the contours of organs or tissues in a CT-like semantic segmentation image are adjusted to obtain a new CT-like semantic segmentation image. Compared with existing technologies that input CT images into a trained neural network and the neural network outputs a semantic segmentation image, this application constructs a standard human body model in digital space and continuously adjusts the structural features in the standard human body model using the finite element method to approximate the target CT image and overcome the problem of missing information.

[0060] S300 uses a difference quantization system to determine whether to adjust the CT-like semantic segmentation map based on the CT-like semantic segmentation map and the new CT-like semantic segmentation map.

[0061] In this embodiment, a difference quantization system is used to calculate the difference between the target CT image and the CT-like semantic segmentation map, as well as the difference between the target CT image and the new CT-like semantic segmentation map. By calculating the differences before and after adjustment, it is determined whether the CT-like semantic segmentation map needs to be adjusted.

[0062] S400, if so, then update the human standard model using the new CT-like semantic segmentation map and determine whether the corresponding organs and / or tissues have been adjusted.

[0063] In this embodiment, when it is determined that adjustments to the CT-like semantic segmentation map are needed, the human body standard model is updated using the new CT-like semantic segmentation map.

[0064] Based on the updated human standard model and the target CT image, it can be determined whether the corresponding organs and / or tissues have been adjusted.

[0065] S500, if so, then a digital twin of the organ and / or tissue corresponding to the target CT image is obtained;

[0066] In this embodiment, when it is determined that the corresponding organ and / or tissue has been adjusted, the corresponding organ and / or tissue is a digital twin of the organ and / or tissue in the target CT image.

[0067] S600 traverses the standard human body model, completes the evolution from the standard human body model to the target individual, and performs virtual slicing at the required locations to obtain the segmentation results of CT medical images.

[0068] In this embodiment, all organs and tissues in the standard human body model are traversed to complete the evolution of the adjusted standard human body model into the target individual. The required positions of the human body model that has completed the evolution into the target individual are then virtually sliced ​​again to obtain the segmentation results of the CT medical image.

[0069] Addressing the issue of unclear boundaries between organs and tissues in CT images due to insufficient scanning resolution or similar physical properties is a major challenge in CT image segmentation. Current CT image segmentation methods rely solely on training neural networks with finely annotated CT images before and after image processing, attempting to enable the network to autonomously learn the three-dimensional morphology and spatial relationships of human organs and tissues. However, this method is incomplete in terms of information. Training the neural network with sufficiently broad and large-scale training data can partially overcome the information gap problem. However, acquiring a large amount of precisely annotated data is itself a challenge in the field of medical imaging. Therefore, it is necessary not only to overcome the information gap problem but also to acquire a large amount of precisely annotated data to obtain accurate and complete CT images.

[0070] This invention does not follow the previous approach to solving medical image segmentation problems. Instead, it directly uses digital twins to provide knowledge of the three-dimensional morphology and spatial relationships of human organs and tissues in digital space. This allows for the processing of medical image segmentation problems based on more complete information, achieving more accurate segmentation results compared to existing technologies from the perspective of information acquisition.

[0071] This invention utilizes digital twins to bring knowledge of the three-dimensional morphology and spatial location of human organs and tissues into computer-aided medical image segmentation methods, enabling them to match the information content of medical experts as closely as possible. By employing finite element analysis and semantic segmentation, the segmentation problem of CT images is transformed into a problem of gradually approximating the CT image's description of the human body using a digital spatial human body model. This cleverly combines the aforementioned knowledge with CT image information to achieve CT image segmentation. This differs significantly from existing technologies that use finely annotated CT medical images to train neural networks for medical image segmentation, thus solving the problem of unclear boundaries between organs and tissues in CT images caused by insufficient scanning resolution or similar physical properties.

[0072] Reference Figures 1 to 3 In some embodiments of the present invention, the process of obtaining the CT-like semantic segmentation map in S100 specifically includes the following steps:

[0073] S110, set the z-axis in the direction from the feet of the human standard model to the head of the human standard model, determine the intersection point of the organ and / or tissue with a certain plane, which is perpendicular to the z-axis, and thus obtain the corresponding slice outline.

[0074] In this embodiment, the z-axis is defined as the direction from the feet of the human standard model to the head of the human standard model. By calculating the intersection point of the organ or tissue with a certain plane that is perpendicular to the z-axis, the slice outline of the organ or tissue is obtained based on the intersection point and the human standard model.

[0075] In other words, in the standard human body model, the z-axis is defined as the direction from the feet to the head. Since a plane perpendicular to the z-axis in the standard human body model may intersect with both tissues and organs, or only with tissues and only with organs, the intersection points of organs and / or tissues with the plane perpendicular to the z-axis are calculated. Based on these intersection points and the standard human body model, the slice contours corresponding to the organs and / or tissues are obtained.

[0076] S120, by distinguishing between the inside and outside of the slice outline, a CT-like semantic segmentation map is obtained.

[0077] In this embodiment, since there are many organs and tissues in the standard human body model, the plane perpendicular to the z-axis may intersect with both tissues and organs, or with only tissues and only organs. To facilitate precise adjustment of the contours of organs and tissues, the standard human body model is virtually sliced. This allows for subsequent adjustments to each slice to achieve the evolution of a particular organ or tissue into a target organ or tissue.

[0078] Reference Figures 1 to 3 In some embodiments of the present invention, the construction of the difference quantification system in S100 specifically includes the following steps:

[0079] S130: Construct an image semantic generation network by training the network using original CT images and finely annotated CT images.

[0080] In this embodiment, raw CT images and finely annotated CT images are obtained from a public dataset, and the image semantic generation network is trained using the raw CT images and finely annotated CT images.

[0081] The image semantic generation network consists of two networks: a generation subnetwork and a discriminator subnetwork.

[0082] For example, finely annotated CT images are input into a generative subnetwork, which then generates and outputs fake CT images. A discrimination subnetwork takes both the finely annotated and fake CT images as input and outputs a quantitative indicator showing the similarity between them. Through this training process, accurate quantitative indicators can be obtained.

[0083] S140, the trained discriminant subnetwork is used to quantify the difference between the semantic segmentation map and the target CT image.

[0084] In this embodiment, the trained discriminant subnetwork can output the difference value between the semantic segmentation map and the target CT image, which can be used in subsequent steps to determine whether the CT-like semantic segmentation map needs to be adjusted.

[0085] The discriminant subnetwork of the trained image semantic generation network is used to quantify the difference between the target CT image and other semantic segmentation maps, where other semantic segmentation maps can be CT-like semantic segmentation maps or new CT-like semantic segmentation maps.

[0086] That is, the discriminant subnetwork of the trained image semantic generation network can quantify the difference between the CT-like semantic segmentation map and the target CT image, and can also quantify the difference between the new CT-like semantic segmentation map and the target CT image.

[0087] Reference Figures 1 to 3 In some embodiments of the present invention, obtaining a new CT-like semantic segmentation map in step S200 specifically includes the following steps:

[0088] S210, Select a point in the contour of an organ or tissue in the CT-like semantic segmentation image as the moving contour point, calculate its tangent, and determine the tangent direction of the tangent.

[0089] In this embodiment, the CT-like semantic segmentation map may contain slices of organs, slices of tissues, and slices of both organs and tissues. Any contour point within the contour of an organ is selected as a moving contour point, or any contour point within the contour of a tissue is selected as a moving contour point. A tangent line to this moving contour point is constructed, and the tangent direction of this tangent line is determined.

[0090] S220: Fix the first contour point that is far away from the moving contour point, move the moving contour point in the tangential direction, and obtain the position of the second contour point that is close to the moving contour point as the moving contour point moves under the constraints of the finite element method. Determine the moving position and obtain a new CT-like semantic segmentation map.

[0091] In this embodiment, a first contour point that is relatively far from the moving contour point is fixed, and the selected moving contour point is moved back and forth in the tangential direction. Under the constraints of the finite element method, the second contour point adjacent to the moving contour point will move accordingly under energy constraints. That is, the distance from the second contour point to the moving contour point is less than the distance from the first contour point to the moving contour point; the second contour point is closer to the moving contour point, and the first contour point is farther from the moving contour point.

[0092] Obtain the position of the second contour point after it has been moved, i.e., the moved position. Determine the moved position and update the CT-like semantic segmentation map to obtain a new CT-like semantic segmentation map.

[0093] Reference Figure 3 In this embodiment, compared with the prior art of directly inputting CT images into a trained neural network and then having the neural network output semantically segmented CT images, the present invention first constructs a standard human body model in digital space, and then uses the finite element method to continuously adjust the structural features of organs and / or tissues in the CT-like semantic segmentation map to make it approximate the target CT image. Figure (a) is a schematic diagram of the selection of moving contour points and tangent calculation; Figure (b) is a schematic diagram of the movement of moving contour points along the tangent direction; and Figure (c) is a schematic diagram of the new CT-like semantic segmentation map.

[0094] Reference Figures 1 to 3 In some embodiments of the present invention, in step S300, determining whether to adjust the CT-like semantic segmentation map specifically includes the following steps:

[0095] S310, using the difference quantization system, input a new CT-like semantic segmentation map and a target CT image to obtain the first difference value; then input the CT-like semantic segmentation map and the target CT image again to obtain the second difference value.

[0096] In this embodiment, a difference quantization system is used to calculate the difference between the CT-like semantic segmentation map before adjustment and the target CT image, i.e., the second difference value, and to calculate the difference between the new CT-like semantic segmentation map after adjustment and the target CT image, i.e., the first difference value.

[0097] S320, compare the first difference value and the second difference value to obtain the comparison result, and based on the comparison result, determine whether to adjust the CT-like semantic segmentation map;

[0098] In this embodiment, the difference values ​​before and after adjustment are calculated and compared. Based on the comparison results, it is determined whether to adopt the new CT-like semantic segmentation map. In other words, it is determined whether to adjust the CT-like semantic segmentation map and adopt the new CT-like semantic segmentation map for subsequent steps.

[0099] When the calculation shows that the adjusted virtual slice is closer to the target CT image, the moving contour points are moved to make adjustments; otherwise, the moving contour points are not moved and no adjustments are made.

[0100] In other words, the first and second difference values ​​are compared to obtain the comparison results. When the first difference value is larger, it is confirmed that the CT-like semantic segmentation map needs to be adjusted, and the new CT-like semantic segmentation map is used to update the human standard model; when the second difference value is larger, it is confirmed that the CT-like semantic segmentation map does not need to be adjusted, and the CT-like semantic segmentation map is used to update the human standard model.

[0101] In this embodiment, a differential quantization system is used to select a semantic segmentation map that is closer to the target CT image based on the virtual slices before and after adjustment. This allows the semantic segmentation map to gradually approach the target CT image and describe the human body problems of the target individual.

[0102] Reference Figures 1 to 3 In some embodiments of the present invention, the determination process for whether the corresponding organ and / or tissue has completed adjustment in S400 specifically includes the following steps:

[0103] S410, confirm whether the first difference value is less than the set threshold.

[0104] In this embodiment, when the first difference value is smaller, it is determined that the CT-like semantic segmentation map needs to be adjusted, and the human standard model is updated with a new CT-like semantic segmentation map. Then, it is determined whether the set threshold is greater than the first difference value; that is, whether the new CT-like semantic segmentation map is close to the target CT image, and whether the difference between it and the target CT image is small.

[0105] It should be noted that, in this embodiment, S410 further includes:

[0106] S411, when the first difference value is greater than the set threshold, return to S100, use the updated human body standard model to perform virtual slicing again, and make adjustments again.

[0107] In this embodiment, when the first difference value is greater than the set threshold, the process returns to S100, and virtual slicing is performed again using the updated human body standard model.

[0108] S420, if so, use the target CT image to confirm whether all adjustments to the corresponding organ and / or tissue have been completed.

[0109] In this embodiment, when the first difference value is less than a set threshold, it is confirmed whether the organs and tissues corresponding to the new CT-like semantic segmentation map have all been adjusted. According to S100, the organs and / or tissues in the target CT image are the organs and / or tissues displayed in the CT-like semantic segmentation map.

[0110] In other words, it is necessary to determine whether the remaining virtual slices of the organ and tissue corresponding to the new CT-like semantic segmentation map have been adjusted. That is, it is necessary to determine whether the outline of the corresponding organ and / or tissue in the new CT-like semantic segmentation map has approached the outline of the corresponding organ and / or tissue in the target CT image.

[0111] S430, if yes, then the corresponding organ and / or tissue is considered to have completed the adjustment.

[0112] In this embodiment, once the above two judgments are satisfied, it is considered that the corresponding organ and / or tissue has been adjusted, and a digital twin of the organ and / or tissue corresponding to the target CT image is obtained.

[0113] Reference Figures 1 to 3 In some embodiments of the present invention, S420 specifically includes the following steps when the corresponding organ and / or tissue has not completed all adjustments:

[0114] S421, fix all contour points in the new CT-like semantic segmentation map, and apply the three-dimensional spatial finite element method to adjust the remaining contour points of organs and / or tissues.

[0115] In this embodiment, when the remaining virtual slices of the organ or tissue corresponding to the new CT-like semantic segmentation map have not been fully adjusted, the contour of the corresponding organ and / or tissue in the new CT-like semantic segmentation map is inconsistent with the contour of the corresponding organ and / or tissue in the target CT image.

[0116] Fix all contour points on the current slice, that is, fix all contour points in the new CT-like semantic segmentation map, and apply the three-dimensional spatial finite element method to adjust the remaining contour points of the corresponding organ and / or tissue, so as to achieve the convergence of the organ and / or tissue contours in the remaining virtual slices to the target organ and / or tissue contours.

[0117] S422, select a new target CT image again and make adjustments.

[0118] In this embodiment, based on the vertical distance of the CT image slices, the actual target CT image of the organ or tissue that has not been adjusted is selected, and the process returns to S100 for adjustment.

[0119] According to an embodiment of a second aspect of the present invention, an electronic device includes:

[0120] A memory is used for computer programs; a processor is used for executing the computer programs stored in the memory. When the processor executes the programs stored in the memory, the processor is used to execute a CT medical image segmentation method as described in the first aspect embodiment of the present invention.

[0121] The processor and memory can be connected via a bus or other means.

[0122] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs, such as the CT medical image segmentation method described in the embodiments of the present invention. The processor implements the CT medical image segmentation method of the first aspect of the present invention by running the non-transitory software program and instructions stored in the memory.

[0123] The memory may include a program storage area and a parameter storage area. The program storage area may store the operating system and an application program required for at least one function. The parameter storage area may store a CT medical image segmentation method described above. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0124] According to an embodiment of a third aspect of the present invention, a storage medium is characterized in that it comprises: storing a computer program for being executed by a processor as a CT medical image segmentation method according to the first aspect of the present invention.

[0125] The non-transitory software program and instructions required to implement the above-described terminal selection method are stored in memory and, when executed by one or more processors, are executed in the first CT medical image segmentation method of the present invention.

[0126] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, parameter structures, program modules, or other parameters). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, parameter structures, program modules, or other parameters in modulation parameter signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0127] The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A CT medical image segmentation method, characterized in that, include: Based on the acquired original CT images and finely annotated CT images, a standard human body model and a differential quantification system are constructed. The organs and / or tissues in the standard human body model are virtually sliced ​​to obtain a CT-like semantic segmentation map, and the corresponding target CT image in the target individual is selected. Using the finite element method, the CT-like semantic segmentation map is adjusted to obtain a new CT-like semantic segmentation map; Based on the CT-like semantic segmentation map and the new CT-like semantic segmentation map, a difference quantization system is used to determine whether to adjust the CT-like semantic segmentation map. If so, the new CT-like semantic segmentation map is used to update the human standard model, and it is determined whether the corresponding organs and / or tissues have been adjusted. If so, a digital twin of the organ and / or tissue corresponding to the target CT image is obtained; The process iterates through the standard human model, transforming it into the target individual, and performs virtual slicing at the required locations to obtain the segmentation results of the CT medical image.

2. The CT medical image segmentation method according to claim 1, characterized in that, The process of obtaining the CT-like semantic segmentation map specifically includes: In the standard human body model, the direction from the foot to the head is set as the z-axis. The intersection points of organs and / or tissues in the standard human body model with the plane perpendicular to the z-axis are calculated to obtain the corresponding slice contours. By distinguishing the inside and outside of the slice contour, the CT-like semantic segmentation map is obtained.

3. The CT medical image segmentation method according to claim 2, characterized in that, The process of obtaining the new CT-like semantic segmentation map specifically includes: In the CT-like semantic segmentation map, a contour point in the organ contour or tissue contour is selected as a moving contour point, the tangent of the moving contour point is constructed, and the direction of the tangent is determined. The first contour point is fixed, the moving contour point is moved along the tangent direction, and under the constraint of the finite element method, the second contour point moves with the moving contour point to obtain the moving position. Based on the moving position, the CT-like semantic segmentation map is updated to obtain a new CT-like semantic segmentation map. Wherein, the distance from the first contour point to the moving contour point is greater than the distance from the second contour point to the moving contour point.

4. The CT medical image segmentation method according to claim 1, characterized in that, The construction process of the difference quantification system specifically includes: An image semantic generation network is constructed and trained based on the original CT images and finely annotated CT images. The image semantic generation network consists of a generation subnetwork and a discriminator subnetwork. The trained discriminant subnetwork is used to quantify the difference between the target CT image and the new CT-like semantic segmentation map or CT-like semantic segmentation map.

5. The CT medical image segmentation method according to claim 1, characterized in that, The process of determining whether to adjust the CT-like semantic segmentation map specifically includes: Input the new CT-like semantic segmentation map and the target CT image into the difference quantization system to obtain a first difference value; input the CT-like semantic segmentation map and the target CT image into the difference quantization system to obtain a second difference value. Compare the first and second difference values, and determine whether to adjust the CT-like semantic segmentation map based on the comparison results.

6. A CT medical image segmentation method according to claim 5, characterized in that, The determination of whether the corresponding organ and / or tissue has completed the adjustment specifically includes: Determine whether the first difference value is less than the set threshold; If so, then based on the target CT image, determine whether all adjustments to the corresponding organ and / or tissue have been completed; If so, the corresponding organ and / or tissue is considered to have completed the adjustment.

7. A CT medical image segmentation method according to claim 6, characterized in that, Specifically, when not all adjustments to the corresponding organ and / or tissue have been completed, this includes: Fix all contour points in the organ and / or tissue contours in the new CT-like semantic segmentation image, and adjust the remaining contour points of the corresponding organs and / or tissues using the three-dimensional spatial finite element method. Select a new target CT image and adjust it again.

8. A CT medical image segmentation method according to claim 5, characterized in that, The step of determining whether the first difference value is less than the set threshold also includes: When the first difference value is greater than the set threshold, the organs and / or tissues in the human standard model are virtually sliced ​​again and adjusted again.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, The processor executes the computer program to implement a CT medical image segmentation method as described in any one of claims 1 to 8.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a CT medical image segmentation method as described in any one of claims 1 to 8.

Images