Image segmentation method, device, computer equipment and storage medium

Abstract

The embodiments of the present application disclose an image segmentation method, apparatus, computer equipment, and storage medium, belonging to the field of computer technology. The method includes: acquiring a first pathological image of a target object, wherein the position points in the first pathological image have absorption parameters; extracting a second pathological image and a third pathological image from the first pathological image, wherein the absorption parameters of the position points in the second pathological image belong to a first parameter range, and the absorption parameters of the position points in the third pathological image belong to a second parameter range; performing feature extraction on the target image to obtain target image features, wherein the target image is synthesized based on the second pathological image and the third pathological image; performing image segmentation on the target image based on the target image features to obtain a segmented image. The target image features extracted by this method can more accurately represent the lesion area, thereby improving the image segmentation effect when performing image segmentation based on the target image features.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular to an image segmentation method, apparatus, computer equipment, and storage medium. Background Art

[0002] With the development of computer technology and image processing technology, image segmentation technology is increasingly being used in the medical field. In the medical field, image segmentation technology is often used to segment edema and hematoma areas in CT (Computed Tomography) images, or edema and tumor areas in MRI (Magnetic Resource Imaging) images.

[0003] Taking the segmentation of edema and hematoma areas on the head as an example, related techniques directly extract features from acquired head CT images and then segment the edema and hematoma areas in the head CT images based on the extracted image features. However, because CT images contain not only edema and hematoma areas but also other regions, the extracted image features are difficult to accurately represent the edema and hematoma areas, resulting in poor image segmentation results. Summary of the Invention

[0004] The embodiments of the present application provide an image segmentation method, apparatus, computer device, and storage medium, which improve the image segmentation effect. The technical solution is as follows:

[0005] In one aspect, a method for image segmentation is provided, the method comprising:

[0006] Acquiring a first pathological image of a target object, where the first pathological image is obtained by scanning the target object using radiation, wherein a position point in the first pathological image has an absorption parameter, and the absorption parameter indicates a degree of absorption of the radiation by the position point in the target object;

[0007] Extracting a second pathological image and a third pathological image from the first pathological image, respectively, wherein absorption parameters of the position points in the second pathological image belong to a first parameter range, which refers to a range within which absorption parameters of a lesion belong; and the absorption parameters of the position points in the third pathological image belong to a second parameter range, which refers to a range within which absorption parameters of a lesion of the first lesion type belong;

[0008] performing feature extraction on a target image to obtain target image features, wherein the target image is synthesized based on the second pathological image and the third pathological image;

[0009] Based on the target image features, the target image is segmented to obtain a segmented image, in which the lesion area of the first lesion type and the lesion area of the second lesion type are respectively marked, and the second lesion type refers to other lesion types except the first lesion type.

[0010] In another aspect, an image segmentation apparatus is provided, the apparatus comprising:

[0011] an image acquisition module, configured to acquire a first pathological image of a target object, wherein the first pathological image is obtained by scanning the target object using radiation, wherein a position point in the first pathological image has an absorption parameter, wherein the absorption parameter indicates a degree of absorption of the radiation by the position point in the target object;

[0012] an image extraction module, configured to extract a second pathological image and a third pathological image from the first pathological image, respectively, wherein the absorption parameters of the position points in the second pathological image belong to a first parameter range, which refers to a range within which the absorption parameters of the lesion site belong; and the absorption parameters of the position points in the third pathological image belong to a second parameter range, which refers to a range within which the absorption parameters of the lesion site of the first lesion type belong;

[0013] a feature extraction module, configured to extract features from a target image to obtain features of the target image, wherein the target image is synthesized based on the second pathological image and the third pathological image;

[0014] An image segmentation module is used to perform image segmentation on the target image based on the target image features to obtain a segmented image, in which the lesion area of the first lesion type and the lesion area of the second lesion type are respectively marked, and the second lesion type refers to other lesion types except the first lesion type.

[0015] In one possible implementation, the minimum value of the first parameter range is the first absorption parameter and the maximum value is the second absorption parameter. The image extraction module is used to adjust the absorption parameters that are smaller than the first absorption parameter among the multiple absorption parameters corresponding to the first pathological image to the first absorption parameter, and adjust the absorption parameters that are greater than the second absorption parameter among the multiple absorption parameters to the second absorption parameter, to obtain the second pathological image.

[0016] In another possible implementation, the minimum value of the second parameter range is the third absorption parameter and the maximum value is the fourth absorption parameter. The image extraction module is used to adjust the absorption parameters that are smaller than the third absorption parameter among the multiple absorption parameters corresponding to the first pathological image to the third absorption parameter, and adjust the absorption parameters that are greater than the fourth absorption parameter among the multiple absorption parameters to the fourth absorption parameter, so as to obtain the third pathological image.

[0017] In another possible implementation, the target object is a head, and the apparatus further includes:

[0018] The image extraction module is further configured to extract a fourth pathological image from the first pathological image, wherein the absorption parameter of the position point in the fourth pathological image belongs to a third parameter range, and the third parameter range refers to a range to which the absorption parameter of the subdural area belongs;

[0019] An image synthesis module is configured to synthesize the target image based on the first pathological image, the second pathological image, and the third pathological image.

[0020] In another possible implementation, the image segmentation module includes:

[0021] a probability determination unit, configured to determine, based on the target image feature, a first probability and a second probability corresponding to a position point in the target image, wherein the first probability indicates a possibility that the position point belongs to the first lesion type, and the second probability indicates a possibility that the position point belongs to the second lesion type;

[0022] a type determination unit, configured to determine a maximum probability between a first probability and a second probability corresponding to a position point in the target image, and determine the type represented by the maximum probability as the type to which the position point belongs;

[0023] The image segmentation unit is used to mark the position points in the target image based on the types of the position points in the target image to obtain the segmented image.

[0024] In another possible implementation, the image segmentation module includes:

[0025] a probability determination unit, configured to determine, based on the target image feature, a first probability, a second probability, and a third probability corresponding to a position point in the target image, wherein the first probability indicates a possibility that the position point belongs to the first lesion type, the second probability indicates a possibility that the position point belongs to the second lesion type, and the third probability indicates a possibility that the position point belongs to a non-lesion type;

[0026] a type determination unit, configured to determine a maximum probability among a first probability, a second probability, and a third probability corresponding to a position point in the target image, and determine the type represented by the maximum probability as the type to which the position point belongs;

[0027] The image segmentation unit is used to mark the position points in the target image based on the types of the position points in the target image to obtain the segmented image.

[0028] In another possible implementation, the target object is a head, the skull-removed model includes a first downsampling sub-model, a first upsampling sub-model, and a skull-removed sub-model, and the apparatus further includes:

[0029] a skull removal module, configured to call the first downsampling sub-model to downsample the first pathological image to obtain a first image feature;

[0030] The skull removal module is further configured to call the first upsampling sub-model to upsample the first image feature to obtain a second image feature, where the second image feature represents a fourth probability corresponding to a position point in the first pathological image, and the fourth probability represents a possibility that the position point belongs to a skull type;

[0031] The skull removal module is also used to call the skull removal sub-model. When the fourth probability corresponding to the position point in the first pathological image is greater than the reference probability, the position point is determined as a skull position point, the skull position point in the first pathological image is removed, and the first pathological image after skull removal is obtained.

[0032] In another possible implementation, the image segmentation model includes a second downsampling sub-model, a second upsampling sub-model, and an image segmentation sub-model, and the feature extraction module is configured to:

[0033] calling the second downsampling sub-model to downsample the target image to obtain a third image feature;

[0034] calling the second upsampling sub-model to upsample the third image feature to obtain the target image feature;

[0035] The image segmentation module is used to call the image segmentation sub-model and perform image segmentation on the target image based on the target image features to obtain the segmented image.

[0036] In another possible implementation, the apparatus further includes:

[0037] a model training module, configured to obtain a sample pathology image and a sample segmentation image corresponding to the sample pathology image, wherein the sample segmentation image is respectively marked with a sample lesion region of the first lesion type and a sample region of the second lesion type;

[0038] The model training module is further configured to call the image segmentation model to process the sample pathological image to obtain a predicted segmented image, wherein the predicted segmented image is respectively marked with a predicted area of the first lesion type and a predicted area of the second lesion type;

[0039] The model training module is further used to process the predicted segmentation image and the sample segmentation image based on a loss function to obtain a loss value, and train the image segmentation model based on the loss value.

[0040] In another possible implementation, the loss function includes at least one of a first loss function, a second loss function, and a third loss function;

[0041] The first loss function represents a relationship between an actual type of a position point in the sample segmented image and a predicted type of a position point in the predicted segmented image, and a first loss value, wherein the actual type refers to a type of a sample region containing the position point, and the predicted type refers to a type of a predicted region containing the position point.

[0042] The second loss function represents a relationship between a sample region of the first lesion type in the sample segmented image and a predicted region of the first lesion type in the predicted segmented image, and a second loss value;

[0043] The third loss function represents the relationship between the sample area of the second lesion type in the sample segmentation image and the predicted area of the second lesion type in the predicted segmentation image, and a third loss value.

[0044] In another possible implementation, the apparatus further includes:

[0045] A normalization module is used to normalize the absorption parameters of the position points in the second pathological image and the position points in the third pathological image respectively.

[0046] On the other hand, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the image segmentation method described in the above aspects.

[0047] On the other hand, a computer-readable storage medium is provided, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to implement the operations performed by the image segmentation method described in the above aspects.

[0048] On the other hand, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the operations performed by the image segmentation method described in the above aspects are implemented.

[0049] The technical solution provided in the embodiment of the present application utilizes the characteristic that different parts of the target object correspond to different absorption parameters, and can extract a second pathological image and a third pathological image from the first pathological image according to a first parameter range corresponding to the lesion part in the target object and a second parameter range corresponding to the lesion part of the first lesion type. By extracting the pathological image, the area in the first pathological image that is not related to the lesion can be removed, and then feature extraction is performed based on the extracted pathological image, so that the feature extraction will not be affected by the area that is not related to the lesion, so that the extracted target image features can more accurately represent the lesion area, so that when image segmentation is performed based on the target image features, the lesion areas of different lesion types can be more accurately determined, thereby improving the image segmentation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0051] Figure 1 This is a schematic diagram of an implementation environment provided by an embodiment of the present application;

[0052] Figure 2 This is a flowchart of an image segmentation method provided by an embodiment of the present application;

[0053] Figure 3 is a flowchart of another image segmentation method provided in an embodiment of the present application;

[0054] Figure 4 is a schematic diagram of an image segmentation model provided in an embodiment of the present application;

[0055] Figure 5 This is a flowchart of another image segmentation method provided in an embodiment of the present application;

[0056] Figure 6 is a schematic diagram of an image segmentation process provided in an embodiment of the present application;

[0057] Figure 7 This is a flowchart of an image segmentation model training method provided in an embodiment of the present application;

[0058] Figure 8 Schematic diagram of a skull removal model provided in an embodiment of the present application;

[0059] Figure 9 This is a flow chart of a method for image skull removal provided in an embodiment of the present application;

[0060] Figure 10 is a schematic diagram of a skull removal process provided in an embodiment of the present application;

[0061] Figure 11 This is a flowchart of another image segmentation method provided in an embodiment of the present application;

[0062] Figure 12 Schematic diagram of a brain window image, an edema window image, and a subdural window image provided in an embodiment of the present application;

[0063] Figure 13 is a schematic diagram of an image segmentation result provided in an embodiment of the present application;

[0064] Figure 14 is a schematic diagram of another image segmentation result provided in an embodiment of the present application;

[0065] Figure 15 This is a structural diagram of an image segmentation device provided in an embodiment of the present application;

[0066] Figure 16 is a structural diagram of another image segmentation device provided in an embodiment of the present application;

[0067] Figure 17 This is a schematic diagram of the structure of a terminal provided in an embodiment of the present application;

[0068] Figure 18 This is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0069] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0070] It is understood that the terms "first," "second," and the like used herein may be used to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are used solely to distinguish one concept from another. For example, a first arrangement sequence may be referred to as a second arrangement sequence, and a second arrangement sequence may be referred to as a first arrangement sequence, without departing from the scope of this application.

[0071] As used herein, the terms "at least one," "plurality," "each," and "any" include one, two, or more, "plurality" includes two or more, "each" refers to each of the corresponding plurality, and "any" refers to any one of the plurality. For example, if the plurality of locations includes three locations, then "each" refers to each of the three locations, and "any" refers to any one of the three locations, which can be the first, second, or third.

[0072] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0073] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, as well as machine learning / deep learning, autonomous driving, and smart transportation.

[0074] Computer vision (CV) is the science of making machines "see." Specifically, it refers to machine vision, where cameras and computers replace the human eye in identifying and measuring objects, performing further image processing to create images more suitable for human observation or transmission to instrumentation. As a scientific discipline, computer vision studies related theories and technologies, aiming to build artificial intelligence systems capable of extracting information from images or multidimensional data. Computer vision technologies typically include image processing, image recognition, image semantic understanding, image retrieval, optical character recognition (OCR), video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, autonomous driving, and smart transportation. It also includes common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0075] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning through demonstration.

[0076] With the research and progress of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, smart customer service, Internet of Vehicles, automatic driving, smart transportation, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0077] The solution provided in the embodiment of the present application is based on artificial intelligence machine learning technology to train an image segmentation model. The trained image segmentation model can be used to segment the lesion area of the first lesion type and the lesion area of the second lesion type in the image.

[0078] The image segmentation method provided in the embodiment of the present application is performed by a computer device. Optionally, the computer device is a terminal or a server. Optionally, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal is a smart phone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, smart voice interaction device, smart home appliance and car terminal, etc., but is not limited to this.

[0079] Figure 1 This is a schematic diagram of an implementation environment provided by an embodiment of the present application. Figure 1 , the implementation environment includes a terminal 101 and a server 102. The terminal 101 and the server 102 are connected via a wireless or wired network, which is not limited in this application.

[0080] The terminal 101 is used to collect a first pathological image of the target object and send the first pathological image to the server 102. The server 102 is used to extract a second pathological image and a third pathological image from the first pathological image, perform feature extraction on a target image synthesized based on the first pathological image and the second pathological image to obtain target image features, and segment the target image based on the target image features to obtain a segmented image, thereby realizing image segmentation.

[0081] In one possible implementation, a target application provided by server 102 is installed on terminal 101, and terminal 101 can implement functions such as data analysis and image display through the target application. Optionally, the target application is a target application in the operating system of terminal 101, or a target application provided by a third party. For example, the target application is an image archiving and communication application, which has an image segmentation function. Of course, the image archiving and communication application can also have other functions, such as image acquisition function, image management function, query function, etc. Optionally, server 102 is the background server of the target application or a cloud server that provides cloud computing and cloud storage services.

[0082] Figure 2 This is a flow chart of an image segmentation method provided by an embodiment of the present application. The execution subject of the embodiment of the present application is a computer device. Figure 2 , the method comprises the following steps:

[0083] 201. A computer device acquires a first pathological image of a target object. The first pathological image is obtained by scanning the target object using rays. Position points in the first pathological image have absorption parameters.

[0084] The target object is any part of the body that can be scanned using radiation, such as the head, chest, etc. The absorption parameter represents the degree of radiation absorption at a point in the target object.

[0085] 202. The computer device extracts a second pathological image and a third pathological image from the first pathological image, wherein the absorption parameters of the position points in the second pathological image belong to the first parameter range, and the absorption parameters of the position points in the third pathological image belong to the second parameter range.

[0086] Among them, the first parameter range refers to the range of the absorption parameter of the lesion site, and the second parameter range refers to the range of the absorption parameter of the lesion site of the first lesion type. The first parameter range includes the second parameter range, that is, the lesion site includes the lesion site of the first lesion type.

[0087] In an embodiment of the present application, based on the range of the absorption parameters of the lesion site, a second pathological image containing the lesion site can be extracted from the first pathological image. Similarly, a third pathological image can be extracted. In the process of extracting the second pathological image and the third pathological image, some areas in the first pathological image that are not related to the lesion site are removed, which facilitates subsequent feature extraction of the lesion site.

[0088] 203. The computer device extracts features from the target image to obtain target image features. The target image is synthesized based on the second pathological image and the third pathological image.

[0089] By synthesizing the second pathological image and the third pathological image into a target image, the target image includes the entire lesion area and can highlight the lesion area of the first lesion type. Therefore, by performing feature extraction on the target image, the extracted target image features can more accurately represent the lesion area.

[0090] 204. The computer device performs image segmentation on the target image based on the target image features to obtain a segmented image.

[0091] The segmented image is marked with lesion areas of a first lesion type and lesion areas of a second lesion type, respectively. The second lesion type refers to other lesion types except the first lesion type.

[0092] The method provided in the embodiment of the present application utilizes the characteristic that different parts of the target object correspond to different absorption parameters, and can extract a second pathological image and a third pathological image from a first pathological image based on a first parameter range corresponding to the lesion part in the target object and a second parameter range corresponding to the lesion part of the first lesion type. By extracting the pathological image, the area in the first pathological image that is not related to the lesion can be removed, and then feature extraction is performed based on the extracted pathological image, so that the feature extraction will not be affected by the area that is not related to the lesion, so that the extracted target image features can more accurately represent the lesion area, so that when image segmentation is performed based on the target image features, the lesion areas of different lesion types can be more accurately determined, thereby improving the image segmentation effect.

[0093] Figure 3 This is a flow chart of another image segmentation method provided by an embodiment of the present application. The execution subject of the embodiment of the present application is a computer device. Figure 3 , the method comprises the following steps:

[0094] 301. A computer device obtains a first pathological image of a target object.

[0095] The target object is any part of the body that can be scanned using radiation, such as the head, chest, and other parts of the body. The first pathological image is obtained by scanning the target object using radiation, such as a CT image or other image obtained by scanning the target object using radiation. The first pathological image includes multiple position points, each position point having a corresponding absorption parameter, which represents the degree of absorption of radiation by the position point in the target object. For example, when the first pathological image is a CT image, the absorption parameter is the CT value, and the range of the absorption parameter is -1000HU-1000HU, where HU (Hounsfield Units) is the unit of the absorption parameter.

[0096] Optionally, the first pathological image is collected by the computer device, or is sent to the computer device by other devices, which is not limited in this embodiment of the present application.

[0097] Optionally, the first pathological image is an image of any size, for example, the first pathological image is an image of 176*176*176, or the first pathological image is an image of 160*160*160. The embodiment of the present application does not limit the size of the first pathological image.

[0098] Optionally, when the first pathological image is a CT image, since different devices may have different orientations when acquiring CT images, to ensure uniformity in subsequent processing, the orientation of the CT image is adjusted to a target orientation, for example, the RA (right abdominal) orientation, before subsequent processing. Optionally, a reorientation model is used to correct the orientation of the CT image.

[0099] 302. The computer device extracts a second pathological image and a third pathological image from the first pathological image based on the first parameter range and the second parameter range, respectively.

[0100] The first parameter range is the range of absorption parameters of the location points in the second pathological image, and the first parameter range refers to the range of absorption parameters of the lesion site. The second parameter range is the range of absorption parameters of the location points in the third pathological image, and the second parameter range refers to the range of absorption parameters of the lesion site of the first lesion type. The lesion site is the site where a lesion occurs in the target object, and the lesion site includes the lesion site of the first lesion type, that is, the first parameter range includes the second parameter range.

[0101] The lesion type includes a first lesion type and a second lesion type, wherein the second lesion type refers to a lesion type other than the first lesion type. For example, the first lesion type is edema, and the second lesion type is hematoma; or, the first lesion type is edema, and the second lesion type is a tumor. For example, the target object is the head, and hematoma and edema occur in the brain. The lesion site is the brain, the lesion site of the first lesion type is the site of edema in the brain, and the lesion site of the second lesion type is the site of hematoma in the brain. When the first pathological image is a CT image, the first parameter range is [0, 80], and the second parameter range is [5, 35].

[0102] In one possible implementation, because different locations in a first pathological image correspond to different absorption parameters, a desired pathological image can be extracted from the first pathological image by changing the absorption parameters at those locations. When the minimum value of a first parameter range is the first absorption parameter and the maximum value is the second absorption parameter, the computer device extracts the second pathological image from the first pathological image based on the first parameter range, including adjusting absorption parameters less than the first absorption parameter among multiple absorption parameters corresponding to the first pathological image to the first absorption parameter, and adjusting absorption parameters greater than the second absorption parameter among the multiple absorption parameters to the second absorption parameter, thereby obtaining the second pathological image. For example, the first absorption parameter is 0 and the second absorption parameter is 80.

[0103] In one possible implementation, when the minimum value of the second parameter range is the third absorption parameter and the maximum value is the fourth absorption parameter, extracting the third pathological image from the first pathological image by the computer device based on the second parameter range includes: adjusting, by the computer device, absorption parameters less than the third absorption parameter among the plurality of absorption parameters corresponding to the first pathological image to the third absorption parameter, and adjusting, by the computer device, absorption parameters greater than the fourth absorption parameter to the fourth absorption parameter, thereby obtaining the third pathological image. For example, the third absorption parameter is 5 and the second absorption parameter is 35.

[0104] 303. The computer device normalizes the absorption parameters of the position points in the second pathological image and the position points in the third pathological image respectively.

[0105] In an embodiment of the present application, since the ranges of absorption parameters corresponding to the second pathological image and the third pathological image are different, the effect of directly synthesizing the second pathological image and the third pathological image is poor. By normalizing the second pathological image and the third pathological image, the absorption parameters of the position points in the second pathological image and the third pathological image after normalization are adjusted to the range of 0 to 1, thereby improving the synthesis effect when synthesizing the normalized second pathological image and the third pathological image.

[0106] For example, the following formula can be used to extract the second pathological image from the first pathological image, and normalize the absorption parameters of the position points in the second pathological image:

[0107]

[0108] Among them, I out represents the normalized second pathological image, I in represents the first pathological image, w min represents the first absorption parameter, w max represents the second absorption parameter, and clip(·) represents the I in Smaller than w min The absorption parameter is adjusted to w min , greater than w max The absorption parameter is adjusted to w max .

[0109] Similarly, replace w in the above formula min is replaced by the third absorption parameter, w max By replacing it with the fourth absorption parameter, a normalized third pathological image can be obtained.

[0110] 304. The computer device synthesizes a target image based on the normalized second pathological image and the normalized third pathological image.

[0111] The synthesized target image refers to taking the normalized second pathological image and the normalized third pathological image as an image channel.

[0112] It should be noted that the embodiments of the present application are only described using the extraction of the second and third pathological images as an example. In another embodiment, when the target object is the head, the first lesion type is edema, and the second lesion type is hematoma, the computer device extracts a fourth pathological image from the first pathological image, and the absorption parameters of the position points in the fourth pathological image fall within a third parameter range. The third parameter range refers to the range of absorption parameters of the subdural area, where the subdural area is a part of the head and is related to hematoma. Based on the second pathological image, the third pathological image, and the fourth pathological image, a target image is synthesized.

[0113] 305. The computer device extracts features from the target image to obtain target image features.

[0114] The target image feature is used to indicate the type of the position point in the target image.

[0115] In one possible implementation, the target image is downsampled to obtain downsampled features, and then the downsampled features are upsampled to obtain target image features. The embodiment of the present application does not limit the implementation method for extracting target image features.

[0116] 306. The computer device performs image segmentation on the target image based on the target image features to obtain a segmented image.

[0117] The target image feature indicates the type of the position point in the target image.

[0118] In one possible implementation, based on target image features, a first probability and a second probability corresponding to a location point in the target image are determined, where the first probability indicates the likelihood that the location point belongs to a first lesion type, and the second probability indicates the likelihood that the location point belongs to a second lesion type. A greater first probability for a location point indicates a greater likelihood that the location point belongs to the first lesion type, while a smaller first probability for a location point indicates a smaller likelihood that the location point belongs to the first lesion type. A greater second probability for a location point indicates a greater likelihood that the location point belongs to the second lesion type, while a smaller second probability for a location point indicates a smaller likelihood that the location point belongs to the second lesion type.

[0119] The maximum probability between the first and second probabilities corresponding to the location point in the target image is then determined, and the type represented by the maximum probability is determined as the type to which the location point belongs. That is, if the first probability corresponding to the location point is greater than the second probability, the type to which the location point belongs is determined to be the first lesion type; if the second probability corresponding to the location point is greater than the first probability, the type to which the location point belongs is determined to be the second lesion type. The type to which the location point belongs is consistent with the type of the location to which the location point corresponds in the target object. For example, if a location in the target object is edematous, then the location point corresponding to that location in the target image is also edematous.

[0120] Based on the type of the location point in the target image, the location point in the target image is marked to obtain a segmented image, in which the lesion area of the first lesion type and the lesion area of the second lesion type are respectively marked. Optionally, in the segmented image, different colors are used to mark the lesion areas of different lesion types, for example, red is used to mark a hematoma area, and green is used to mark an edema area. Alternatively, other methods can be used to mark the lesion areas of different lesion types.

[0121] In another possible implementation, in addition to marking lesion areas of the first lesion type and lesion areas of the second lesion type in the segmented image, the computer device also marks background areas that do not belong to any lesion type. Specifically, the computer device determines a first probability, a second probability, and a third probability corresponding to a location point in the target image based on the target image features. The third probability represents the likelihood that the location point belongs to a non-lesion type. A greater third probability for a location point indicates a greater likelihood that the location point belongs to a non-lesion type, while a smaller third probability for a location point indicates a lower likelihood that the location point belongs to a non-lesion type.

[0122] Then, the maximum probability among the first, second, and third probabilities corresponding to the location point in the target image is determined, and the type represented by the maximum probability is determined as the type to which the location point belongs. Specifically, if the first probability corresponding to the location point is greater than the second and third probabilities, the type to which the location point belongs is determined to be the first lesion type; if the second probability corresponding to the location point is greater than the first and third probabilities, the type to which the location point belongs is determined to be the second lesion type; and if the third probability corresponding to the location point is greater than the first and second probabilities, the type to which the location point belongs is determined to be the non-lesion type.

[0123] Then, based on the type of the position point in the target image, the position point in the target image is marked to obtain a segmented image, in which the lesion area of the first lesion type, the lesion area of the second lesion type and the background area of the non-lesion type are marked respectively.

[0124] It should be noted that, for multiple position points in the target image, the above method can be used to determine the type of each position point.

[0125] The method provided in the embodiment of the present application utilizes the characteristic that different parts of the target object correspond to different absorption parameters, and can extract a second pathological image and a third pathological image from a first pathological image based on a first parameter range corresponding to the lesion part in the target object and a second parameter range corresponding to the lesion part of the first lesion type. By extracting the pathological image, the area in the first pathological image that is not related to the lesion can be removed, and then feature extraction is performed based on the extracted pathological image, so that the feature extraction will not be affected by the area that is not related to the lesion, so that the extracted target image features can more accurately represent the lesion area, so that when image segmentation is performed based on the target image features, the lesion areas of different lesion types can be more accurately determined, thereby improving the image segmentation effect.

[0126] Moreover, in an embodiment of the present application, the segmentation of the lesion area of the first lesion type and the lesion area of the second lesion type is achieved simultaneously, that is, when segmenting the lesion area of the first lesion type, the lesion area of the second lesion type will be considered, and when segmenting the lesion area of the second lesion type, the lesion area of the first lesion type will also be considered. In this way, during the image segmentation process, the lesion areas of the two lesion types constrain each other, further improving the image segmentation effect.

[0127] In a possible implementation, for the above steps 304 and 305, an image segmentation model can be called to extract target image features corresponding to the target image and perform image segmentation on the target image. Figure 4 The image segmentation model includes a second downsampling sub-model 401, a second upsampling sub-model 402 and an image segmentation sub-model 403, wherein the second downsampling sub-model 401 is used to downsample the target image, the second upsampling sub-model 402 is used to upsample the image features output by the second downsampling sub-model 401, and the image segmentation sub-model 403 is used to segment the target image based on the image features output by the second upsampling sub-model 402. Figure 5 The illustrated embodiment explains this in detail.

[0128] Figure 5 This is a flowchart of another image segmentation method provided by an embodiment of the present application. The execution subject of the embodiment of the present application is a computer device. Figure 5 , the method comprises the following steps:

[0129] 501. The computer device calls the second downsampling sub-model to downsample the target image to obtain a third image feature.

[0130] 502. The computer device calls the second upsampling sub-model to upsample the third image feature to obtain the target image feature.

[0131] In one possible implementation, the second downsampling submodel includes n1 second downsampling networks, and the second upsampling submodel includes n1 second upsampling networks. The computer device calls the first second downsampling network to downsample the target image to obtain the first third image feature; calls the k1th second downsampling network to downsample the k1-1th third image feature output by the k1-1th second downsampling network to obtain the k1th third image feature, and so on, until the n1th third image feature output by the n1th second downsampling network is obtained; calls the first second upsampling network to upsample the n1th third image feature to obtain the first fourth image feature; calls the k1th second upsampling network to upsample the k1-1th fourth image feature output by the k1-1th second upsampling network to obtain the k1th fourth image feature, and so on, until the n1th fourth image feature output by the n1th second upsampling network is obtained, and the n1th fourth image feature is the target image feature.

[0132] Optionally, the image segmentation model further includes a mapping submodel, which is called to perform identity mapping on n1 third image features to obtain n1 first mapping features, and the k1th first mapping feature is the same as the k1th third image feature. The computer device calls the second upsampling submodel to upsample the third image features to obtain target image features, including: calling the first second upsampling network to upsample the n1th third image feature and the n1th first mapping feature to obtain the first fourth image feature; calling the k1th second upsampling network to upsample the k1-1th fourth image feature output by the k1-1th second upsampling network and the k1-1th first mapping feature to obtain the k1th fourth image feature, and so on until the n1th fourth image feature output by the n1th second upsampling network is obtained.

[0133] For example, the second downsampling submodel includes 4 second downsampling networks, the second upsampling submodel includes 4 second upsampling networks, and each second downsampling network includes a convolution subnetwork and a 1*1 convolution layer with a step size of 2, and the convolution subnetwork includes a residual block and a dense block, and each second upsampling network includes an upsample layer.

[0134] It should be noted that the embodiment of the present application uses a U-shaped network to extract target image features as an example. In another embodiment, models with other structures can be used to extract target image features, and the embodiment of the present application does not limit this.

[0135] 503. The computer device calls the image segmentation sub-model and performs image segmentation on the target image based on the target image features to obtain a segmented image.

[0136] Optionally, the image segmentation sub-model includes an activation function and a labeling layer. For example, the activation function is a Softmax function. The activation function is called to convert the target image features into corresponding first, second, and third probabilities. Each location point is then labeled based on the labeling layer and the obtained probabilities.

[0137] Optionally, the image segmentation sub-model further includes a three-dimensional visualization layer, which is called to convert the segmented image into a three-dimensional visualization image.

[0138] For example, see Figure 6 The schematic diagram of the image segmentation process shown is as follows: a target image 601 is input into an image segmentation model 602 , and the image segmentation model 602 is called to process the target image to obtain a three-dimensional visualization image 603 .

[0139] The method provided in the embodiment of the present application calls an image segmentation model and performs feature extraction based on the extracted pathological image, so that the feature extraction will not be affected by areas unrelated to the lesion, and the extracted target image features can more accurately represent the lesion area. Therefore, when image segmentation is performed based on the target image features, the lesion areas of different lesion types can be more accurately determined, thereby improving the image segmentation effect, and the end-to-end image segmentation model also improves the image segmentation efficiency.

[0140] For the above Figure 4 The image segmentation model shown in the figure needs to be trained before the image segmentation model is used to segment the target image. Figure 7 The illustrated embodiment illustrates the training process of the image segmentation model.

[0141] Figure 7 This is a flowchart of an image segmentation model training method provided by an embodiment of the present application. The execution subject of the embodiment of the present application is a computer device. Figure 7 , the method comprises the following steps:

[0142] 701. A computer device obtains a sample pathology image and a sample segmentation image corresponding to the sample pathology image, wherein a sample lesion region of a first lesion type and a sample region of a second lesion type are respectively marked in the sample segmentation image.

[0143] The sample pathological image is a pathological image corresponding to the target part. The method for acquiring the sample pathological image is the same as the method for acquiring the first pathological image in step 301 above, and will not be repeated here.

[0144] 702. The computer device calls the image segmentation model to process the sample pathological image to obtain a predicted segmented image, wherein the predicted segmented image is marked with a predicted area of the first lesion type and a predicted area of the second lesion type.

[0145] The implementation of step 702 is similar to the implementation of steps 501 to 503 above, and will not be repeated here.

[0146] 703. The computer device processes the predicted segmentation image and the sample segmentation image based on the loss function to obtain a loss value, and trains the image segmentation model based on the loss value.

[0147] Among them, the smaller the loss value, the more accurate the image segmentation model.

[0148] In one possible implementation, the loss function includes at least one of the following first loss function, second loss function, and third loss function:

[0149] (1) The first loss function.

[0150] The first loss function represents the relationship between the actual type of the position point in the sample segmentation image and the predicted type of the position point in the predicted segmentation image, and the first loss value, where the actual type refers to the type of the sample area containing the position point, and the predicted type refers to the type of the predicted area containing the position point.

[0151] For example, the first loss function WCE (Weighted Cross Entropy) loss function:

[0152]

[0153] Among them, l ce represents the first loss value, N represents the number of position points in the sample pathology image, i represents the i-th position point in the sample pathology image, C represents the number of types, c represents the c-th type, w c represents the reference parameter corresponding to the cth type, g represents the sample segmentation image, p represents the predicted segmentation image, Indicates that the i-th position point in the sample segmentation image belongs to the c-th type, It indicates the probability that the th position point in the predicted segmented image belongs to the cth type, and log(·) indicates the logarithm.

[0154] The types corresponding to C may include at least two lesion types, or may also include non-lesion types. For example, the c-th type may be an edema type, a hematoma type, and a non-lesion type.

[0155] (2) The second loss function.

[0156] The second loss function represents the relationship between the sample region of the first lesion type in the sample segmentation image and the predicted region of the first lesion type in the predicted segmentation image, and the second loss value.

[0157] For example, the second loss function is the DSC (Dice Similarity Coefficient) loss function:

[0158]

[0159] Wherein, l1 represents the second loss value, X1 represents the sample area of the first lesion type, Y1 represents the predicted area of the first lesion type, |X1| represents the area of the sample area of the first lesion type, |Y1| represents the area of the predicted area of the first lesion type, and |X1∩Y1| represents the area of the overlapping part between the sample area of the first lesion type and the predicted area of the first lesion type.

[0160] (3) The third loss function.

[0161] The third loss function represents the relationship between the sample region of the second lesion type in the sample segmentation image and the predicted region of the second lesion type in the predicted segmentation image, and the third loss value.

[0162] For example, the third loss function is the DSC function:

[0163]

[0164] Among them, l2 represents the third loss value, X2 represents the sample area of the second lesion type, Y2 represents the predicted area of the second lesion type, |X2| represents the area of the sample area of the second lesion type, |Y2| represents the area of the predicted area of the second lesion type, and |X2∩Y2| represents the area of the overlapping part between the sample area of the second lesion type and the predicted area of the second lesion type.

[0165] In another possible implementation, after obtaining corresponding loss values using at least two of the first, second, and third loss functions, a weighted sum of the at least two loss values is performed to obtain the loss value corresponding to the loss function. For example, referring to the following formula, the final loss value is obtained based on the three loss functions:

[0166] L total =λ1l ce +λ2l1+λ3l2

[0167] Among them, L total Represents the loss value, l ce, l1, l2 represent the first loss value, the second loss value and the third loss value respectively, λ1, λ2, λ3 represent the weights corresponding to the first loss value, the second loss value and the third loss value respectively.

[0168] It should be noted that the above steps 701 to 703 are described by taking one training process as an example. In another embodiment, the image segmentation model can be iteratively trained multiple times until a trained image segmentation model is obtained.

[0169] The method provided in the embodiment of the present application adopts three loss functions when training the image segmentation model, wherein the first loss function can accurately constrain the segmentation accuracy of each position point in the segmented image, the second loss function and the third loss function can respectively constrain the segmentation accuracy of the lesion area of the first lesion type and the lesion area of the second lesion type in the segmented image, and adjust the accuracy of the image segmentation model from different angles, thereby improving the accuracy of the image segmentation model training.

[0170] In the case where the target object is a head, the first pathological image of the target object also includes a skull region. The presence of the skull region will affect the image segmentation process shown in steps 302 to 305 above. Therefore, in order to achieve more accurate image segmentation, the first pathological image is first subjected to skull removal processing before image segmentation. In one possible implementation, a skull removal model is called to remove the skull from the first pathological image. Figure 8 The skull removal model includes a first downsampling sub-model 801, a first upsampling sub-model 802 and a skull removal sub-model 803, wherein the first downsampling sub-model 801 is used to downsample the first pathological image, the first upsampling sub-model 802 is used to upsample the image features output by the first downsampling sub-model 801, and the skull removal sub-model 803 is used to remove the skull of the first pathological image based on the image features output by the first upsampling sub-model 802. Figure 9 The illustrated embodiment illustrates the process of calling the skull removal model to remove the skull of the first pathological image.

[0171] Figure 9 This is a flowchart of a method for removing skull bones from an image provided by an embodiment of the present application. The execution subject of the embodiment of the present application is a computer device. Figure 9 , the method comprises the following steps:

[0172] 901. Call a first downsampling sub-model to downsample a first pathological image to obtain a first image feature.

[0173] 902. Call the first upsampling sub-model to upsample the first image feature to obtain a second image feature.

[0174] The second image feature represents a fourth probability corresponding to the location point in the first pathological image, and the fourth probability represents a likelihood that the location point belongs to a skull type. A larger fourth probability indicates a greater likelihood that the location point belongs to a skull type, and a smaller fourth probability indicates a smaller likelihood that the location point belongs to a skull type.

[0175] In one possible implementation, the first downsampling sub-model includes n2 first downsampling networks, and the first upsampling sub-model includes n2 first upsampling networks. The computer device calls the first first downsampling network to downsample the target image to obtain the first first image feature; calls the k2th first downsampling network to downsample the k2-1th first image feature output by the k2-1th first downsampling network to obtain the k2th first image feature, and so on until the n2th first image feature output by the n2th first downsampling network is obtained; calls the first first upsampling network to upsample the n2th first image feature to obtain the first second image feature; calls the k2th first upsampling network to upsample the k2-1th second image feature output by the k2-1th first upsampling network to obtain the k2th second image feature, and so on until the n2th second image feature output by the n2th first upsampling network is obtained.

[0176] Optionally, the image segmentation model further includes a mapping submodel, which is called to perform identity mapping on the n2 first image features to obtain n2 first mapping features, and the k2th first mapping feature is the same as the k2th first image feature. The computer device calls the first upsampling submodel to upsample the first image features to obtain target image features, including: calling the first first upsampling network to upsample the n2th first image feature and the n2th first mapping feature to obtain the first second image feature; calling the k2th first upsampling network to upsample the k2-1th second image feature output by the k2-1th first upsampling network and the k2-1th first mapping feature to obtain the k2th second image feature, until the n2th second image feature output by the n2th first upsampling network is obtained.

[0177] 903. Call the skull removal sub-model, and when the fourth probability corresponding to the position point in the first pathological image is greater than the reference probability, determine the position point as a skull position point, remove the skull position point in the first pathological image, and obtain the first pathological image after skull removal.

[0178] The reference probability is a preset probability, for example, 0.5, 0.6, or other values less than 1. When the fourth probability corresponding to the location point is greater than the reference probability, the location point can be determined to be a skull location point.

[0179] In one possible implementation, skull position points in a first pathological image are removed to obtain the first pathological image after skull removal, including: determining pixel values corresponding to skull position points as 0, determining pixel values corresponding to non-skull position points as 1, obtaining a binary mask image, and superimposing the binary mask image on the first pathological image to obtain the first pathological image after skull removal. In the binary mask image, for positions where the pixel points are 1, multiplication by the corresponding position points in the first pathological image keeps the absorption parameters of the positions in the first pathological image unchanged, while for positions where the pixel points are 0 in the binary mask image, multiplication by the corresponding position points in the first pathological image turns the positions in the first pathological image into a black color consistent with the background area, thereby removing the skull area.

[0180] In a possible implementation, after obtaining the first pathological image after skull removal, the first pathological image is cropped so that the first pathological image contains as little background area as possible.

[0181] For example, see Figure 10 The schematic diagram of the skull removal process shown is as follows: a first pathological image 1001 is input into a skull removal model 1002, the skull removal model 1002 is called, the skull removal is performed on the first pathological image, and a first pathological image 1003 after skull removal is obtained.

[0182] In one possible implementation, before using the skull removal model, the skull removal model is first trained, that is, a sample pathological image containing a skull area and a sample pathological image after skull removal are obtained, the skull removal model is called, and the sample pathological image containing a skull area is subjected to skull removal processing to obtain a predicted pathological image after skull removal, and the skull removal model is trained based on the difference between the sample pathological image after skull removal and the predicted pathological image after skull removal.

[0183] For example, the skull removal model is trained based on the following fourth loss function:

[0184]

[0185] Among them, l3 represents the fourth loss value, X3 represents the sample pathology image after skull removal, Y3 represents the predicted pathology image after skull removal, |X3| represents the area of the sample pathology image after skull removal, |Y3| represents the area of the predicted pathology image after skull removal, and |X3∩Y3| represents the area of the overlapping part between the sample pathology image after skull removal and the predicted pathology image after skull removal.

[0186] It should be noted that Figure 9This is only one way to remove the skull. In another embodiment, other ways can be used to remove the skull region in the pathological image. For example, a BET (Brain Extraction Tool) algorithm can be used to remove the skull.

[0187] The skull removal method provided in the embodiment of the present application can remove the skull area in the pathological image of the head, thereby avoiding the influence of the skull area on the subsequent processing process and improving the subsequent image processing effect.

[0188] The image segmentation method provided in the embodiments of the present application can be applied to a variety of scenarios, such as the segmentation of edema and hematoma, the segmentation of edema and tumor, etc. The image segmentation process is described below using the example of a target object being a head, a first lesion type being edema, a second lesion type being hematoma, and a first pathological image being a CT image.

[0189] Figure 11 This is a flowchart of another image segmentation method provided by an embodiment of the present application. The execution subject of the embodiment of the present application is a computer device. Figure 11 , the method comprises the following steps:

[0190] 1101. The computer device obtains a CT image of the head.

[0191] 1102. The computer device performs skull removal processing on the CT image to obtain a CT image after skull removal.

[0192] 1103. The computer device extracts a brain window image, an edema window image, and a subdural window image from the CT image after skull removal.

[0193] For example, see Figure 12 The computer device extracts a brain window image 1201 from the CT image based on the CT value range [0, 80], extracts an edema window image 1202 from the CT image based on the CT value range [5, 35], and extracts a subdural window image 1203 from the CT image based on the CT value range [-20, 180].

[0194] 1104. The computer device normalizes the brain window image, the edema window image, and the subdural window image respectively to obtain normalized brain window image, edema window image, and subdural window image.

[0195] 1105. The computer device synthesizes a target image based on the normalized brain window image, edema window image, and subdural window image.

[0196] 1106. The computer device extracts features from the target image to obtain target image features.

[0197] 1107. The computer device performs image segmentation on the target image based on the target image features to obtain a segmented image, wherein the segmented image is marked with an edema area and a hematoma area.

[0198] The implementation of steps 1101 to 1107 is the same as above Figure 3 The same is true for the embodiment shown, which will not be described again here.

[0199] For example, see Figure 13 As shown in the schematic diagram, the first pathological image 1301 is first de-skulled to obtain the first pathological image 1302 after the skull is removed, and then the first pathological image 1302 after the skull is removed is segmented to obtain the segmented image 1303, and then the segmented image 1303 is three-dimensionally visualized to obtain the three-dimensional visualized image 1304.

[0200] For example, see Figure 14 As shown in the schematic diagram, the first pathological image 1401 is first de-skulled to obtain the first pathological image 1402 after the skull is removed, and then the first pathological image 1402 after the skull is removed is segmented to obtain the segmented image 1403, and then the segmented image 1403 is three-dimensionally visualized to obtain the three-dimensional visualized image 1404.

[0201] In addition, in one possible implementation, the above-mentioned image segmentation method can be applied to PACS (Picture Archiving and Communication Systems) applications. Through the PACS application, the acquired CT images are segmented to obtain edema areas and hematoma areas, and then the volume of the hematoma area, the volume of the edema area, etc. can be further analyzed.

[0202] Figure 15 This is a schematic diagram of the structure of an image segmentation device provided in an embodiment of the present application. Figure 15 , the device comprises:

[0203] An image acquisition module 1501 is configured to acquire a first pathological image of a target object. The first pathological image is obtained by scanning the target object using radiation. A location in the first pathological image has an absorption parameter, which indicates a degree of absorption of the radiation by the location in the target object.

[0204] An image extraction module 1502 is configured to extract a second pathological image and a third pathological image from the first pathological image, wherein the absorption parameter of the position point in the second pathological image belongs to a first parameter range, wherein the first parameter range refers to a range of absorption parameters of the lesion site; and the absorption parameter of the position point in the third pathological image belongs to a second parameter range, wherein the second parameter range refers to a range of absorption parameters of the lesion site of the first lesion type;

[0205] A feature extraction module 1503 is configured to extract features from a target image to obtain features of the target image, where the target image is synthesized based on the second pathological image and the third pathological image;

[0206] The image segmentation module 1504 is used to perform image segmentation on the target image based on the target image features to obtain a segmented image, in which the lesion area of the first lesion type and the lesion area of the second lesion type are respectively marked, and the second lesion type refers to other lesion types except the first lesion type.

[0207] The device provided in the embodiment of the present application utilizes the characteristic that different parts of the target object correspond to different absorption parameters, and can extract a second pathological image and a third pathological image from a first pathological image based on a first parameter range corresponding to the lesion part in the target object and a second parameter range corresponding to the lesion part of the first lesion type. By extracting the pathological image, the area in the first pathological image that is not related to the lesion can be removed, and then feature extraction is performed based on the extracted pathological image, so that the feature extraction will not be affected by the area that is not related to the lesion, so that the extracted target image features can more accurately represent the lesion area, so that when image segmentation is performed based on the target image features, the lesion areas of different lesion types can be more accurately determined, thereby improving the image segmentation effect.

[0208] In one possible implementation, the minimum value of the first parameter range is the first absorption parameter and the maximum value is the second absorption parameter. The image extraction module 1502 is used to adjust the absorption parameters that are smaller than the first absorption parameter among the multiple absorption parameters corresponding to the first pathological image to the first absorption parameter, and adjust the absorption parameters that are greater than the second absorption parameter among the multiple absorption parameters to the second absorption parameter, so as to obtain the second pathological image.

[0209] In another possible implementation, the minimum value of the second parameter range is the third absorption parameter and the maximum value is the fourth absorption parameter. The image extraction module 1502 is used to adjust the absorption parameters that are smaller than the third absorption parameter among the multiple absorption parameters corresponding to the first pathological image to the third absorption parameter, and adjust the absorption parameters that are greater than the fourth absorption parameter among the multiple absorption parameters to the fourth absorption parameter, so as to obtain the third pathological image.

[0210] In another possible implementation, the target object is a head, see Figure 16 , the device further comprises:

[0211] The image extraction module 1502 is further configured to extract a fourth pathological image from the first pathological image, wherein the absorption parameter of the position point in the fourth pathological image belongs to a third parameter range, and the third parameter range refers to a range within which the absorption parameter of the subdural area belongs;

[0212] The image synthesis module 1505 is configured to synthesize the target image based on the first pathological image, the second pathological image, and the third pathological image.

[0213] In another possible implementation, see Figure 16 The image segmentation module 1504 includes:

[0214] a probability determination unit 1514 configured to determine, based on the target image feature, a first probability and a second probability corresponding to a position point in the target image, wherein the first probability indicates a likelihood that the position point belongs to the first lesion type, and the second probability indicates a likelihood that the position point belongs to the second lesion type;

[0215] A type determination unit 1524 is configured to determine the maximum probability between the first probability and the second probability corresponding to the position point in the target image, and determine the type represented by the maximum probability as the type to which the position point belongs;

[0216] The image segmentation unit 1534 is configured to mark the position points in the target image based on the types to which the position points in the target image belong, to obtain the segmented image.

[0217] In another possible implementation, see Figure 16 The image segmentation module 1504 includes:

[0218] a probability determination unit 1514 configured to determine, based on the target image feature, a first probability, a second probability, and a third probability corresponding to a position point in the target image, wherein the first probability indicates a probability that the position point belongs to the first lesion type, the second probability indicates a probability that the position point belongs to the second lesion type, and the third probability indicates a probability that the position point belongs to a non-lesion type;

[0219] A type determination unit 1524 is configured to determine the maximum probability among the first probability, the second probability, and the third probability corresponding to the position point in the target image, and determine the type represented by the maximum probability as the type to which the position point belongs;

[0220] The image segmentation unit 1534 is configured to mark the position points in the target image based on the types to which the position points in the target image belong, to obtain the segmented image.

[0221] In another possible implementation, the target object is a head, and the skull removal model includes a first downsampling sub-model, a first upsampling sub-model, and a skull removal sub-model, see Figure 16 , the device further comprises:

[0222] a skull removal module 1506 , configured to call the first downsampling sub-model to downsample the first pathological image to obtain a first image feature;

[0223] The skull removal module 1506 is further configured to call the first upsampling sub-model to upsample the first image feature to obtain a second image feature, where the second image feature represents a fourth probability corresponding to a location point in the first pathological image, where the fourth probability represents a likelihood that the location point belongs to a skull type.

[0224] The skull removal module 1506 is also used to call the skull removal sub-model. When the fourth probability corresponding to the position point in the first pathological image is greater than the reference probability, the position point is determined as a skull position point, and the skull position point in the first pathological image is removed to obtain the first pathological image after skull removal.

[0225] In another possible implementation, the image segmentation model includes a second downsampling sub-model, a second upsampling sub-model, and an image segmentation sub-model, and the feature extraction module 1503 is configured to:

[0226] Calling the second downsampling sub-model to downsample the target image to obtain a third image feature;

[0227] Calling the second upsampling sub-model to upsample the third image feature to obtain the target image feature;

[0228] The image segmentation module is used to call the image segmentation sub-model and perform image segmentation on the target image based on the target image features to obtain the segmented image.

[0229] In another possible implementation, see Figure 16 , the device further comprises:

[0230] A model training module 1507 is configured to obtain a sample pathology image and a sample segmentation image corresponding to the sample pathology image, wherein the sample segmentation image is respectively marked with a sample lesion region of the first lesion type and a sample region of the second lesion type;

[0231] The model training module 1507 is further configured to call the image segmentation model to process the sample pathological image to obtain a predicted segmented image, wherein the predicted region of the first lesion type and the predicted region of the second lesion type are respectively marked in the predicted segmented image;

[0232] The model training module 1507 is further configured to process the predicted segmented image and the sample segmented image based on a loss function to obtain a loss value, and train the image segmentation model based on the loss value.

[0233] In another possible implementation, the loss function includes at least one of a first loss function, a second loss function, and a third loss function;

[0234] The first loss function represents a relationship between an actual type of a position point in the sample segmented image and a predicted type of a position point in the predicted segmented image, and a first loss value, wherein the actual type refers to a type of a sample region containing the position point, and the predicted type refers to a type of a predicted region containing the position point.

[0235] The second loss function represents a relationship between a sample region of the first lesion type in the sample segmented image, a predicted region of the first lesion type in the predicted segmented image, and a second loss value;

[0236] The third loss function represents the relationship between the sample region of the second lesion type in the sample segmentation image, the predicted region of the second lesion type in the predicted segmentation image, and the third loss value.

[0237] In another possible implementation, see Figure 16 , the device further comprises:

[0238] The normalization module 1508 is configured to normalize the absorption parameters of the position points in the second pathological image and the position points in the third pathological image respectively.

[0239] All of the above optional technical solutions can be combined in any way to form optional embodiments of the present application, and will not be described in detail here.

[0240] It should be noted that the image segmentation device provided in the above embodiment is merely an example of the division of the above functional modules when segmenting an image. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the image segmentation device provided in the above embodiment and the image segmentation method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0241] An embodiment of the present application further provides a computer device comprising a processor and a memory, wherein the memory stores at least one computer program, which is loaded and executed by the processor to implement the operations performed by the image segmentation method of the above embodiment.

[0242] Optionally, the computer device is provided as a terminal. Figure 17 FIG1 is a schematic diagram of the structure of a terminal 1700 provided in an embodiment of the present application. The terminal 1700 includes: a processor 1701 and a memory 1702.

[0243] The processor 1701 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1701 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 1701 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 1701 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1701 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0244] Memory 1702 may include one or more computer-readable storage media, which may be non-transitory. Memory 1702 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in memory 1702 is used to store at least one computer program, which is executed by processor 1701 to implement the image segmentation method provided in the method embodiment of the present application.

[0245] In some embodiments, terminal 1700 may optionally include a peripheral device interface 1703 and at least one peripheral device. Processor 1701, memory 1702, and peripheral device interface 1703 may be connected via a bus or signal lines. Each peripheral device may be connected to peripheral device interface 1703 via a bus, signal lines, or circuit boards. Specifically, the peripheral device may include at least one of a radio frequency circuit 1704, a display screen 1705, a camera assembly 1706, an audio circuit 1707, and a power supply 1708.

[0246] The peripheral device interface 1703 can be used to connect at least one I / O (Input / Output)-related peripheral device to the processor 1701 and the memory 1702. In some embodiments, the processor 1701, the memory 1702, and the peripheral device interface 1703 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1701, the memory 1702, and the peripheral device interface 1703 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0247] RF circuit 1704 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. RF circuit 1704 communicates with communication networks and other communication devices via electromagnetic signals. RF circuit 1704 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. RF circuit 1704 may optionally include an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. RF circuit 1704 may communicate with other terminals via at least one wireless communication protocol. Such wireless communication protocols include, but are not limited to, the World Wide Web, metropolitan area networks, intranets, various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks, and / or WiFi (Wireless Fidelity) networks. In some embodiments, RF circuit 1704 may also include circuitry related to Near Field Communication (NFC), although this application does not limit this.

[0248] Display screen 1705 is used to display a UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When display screen 1705 is a touch screen display, display screen 1705 also has the ability to collect touch signals on or above the surface of display screen 1705. The touch signals can be input as control signals to processor 1701 for processing. In this case, display screen 1705 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards.

[0249] The camera assembly 1706 is used to capture images or videos. Optionally, the camera assembly 1706 includes a front camera and a rear camera. The front camera is set on the front panel of the terminal, and the rear camera is set on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions.

[0250] Audio circuit 1707 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and converts the sound waves into electrical signals that are input to processor 1701 for processing, or input to RF circuit 1704 for voice communication. The speaker is used to convert electrical signals from processor 1701 or RF circuit 1704 into sound waves. The speaker can be a traditional thin-film speaker or a piezoelectric ceramic speaker.

[0251] Power supply 1708 is used to power various components in terminal 1700. Power supply 1708 can be AC power, DC power, a disposable battery, or a rechargeable battery. When power supply 1708 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is charged via a wired line, while a wireless rechargeable battery is charged via a wireless coil. The rechargeable battery can also support fast charging technology.

[0252] Those skilled in the art will understand that Figure 17 The structure shown in the figure does not constitute a limitation on the terminal 1700, and the terminal 1700 may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.

[0253] Optionally, the computer device is provided as a server. Figure 18 1 is a schematic diagram of the structure of a server provided in an embodiment of the present application. The server 1800 may vary significantly due to different configurations or performances, and may include one or more processors (Central Processing Units, CPUs) 1801 and one or more memories 1802. The memories 1802 store at least one computer program, which is loaded and executed by the processor 1801 to implement the methods provided in the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and input / output interfaces for input and output. The server may also include other components for implementing device functions, which will not be described in detail here.

[0254] An embodiment of the present application further provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to implement the operations performed by the image segmentation method of the above embodiment.

[0255] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the operations performed by the image segmentation method of the above embodiment are implemented.

[0256] In some embodiments, the computer program involved in the embodiments of the present application may be deployed and executed on a computer device, or on multiple computer devices located at one location, or on multiple computer devices distributed at multiple locations and interconnected through a communication network. Multiple computer devices distributed at multiple locations and interconnected through a communication network may constitute a blockchain system.

[0257] It is understandable that in the specific implementation of this application, related data such as user information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0258] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0259] The above are only optional embodiments of the embodiments of the present application and are not intended to limit the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of this application.

Claims

1. An image segmentation method, characterized in that: The method comprises: Acquiring a first pathological image of a target object, where the first pathological image is obtained by scanning the target object using radiation, wherein a position point in the first pathological image has an absorption parameter, and the absorption parameter indicates a degree of absorption of the radiation by the position point in the target object; Extracting a second pathological image and a third pathological image from the first pathological image, respectively, wherein absorption parameters of the position points in the second pathological image belong to a first parameter range, which refers to a range within which absorption parameters of a lesion belong; and the absorption parameters of the position points in the third pathological image belong to a second parameter range, which refers to a range within which absorption parameters of a lesion of the first lesion type belong; performing feature extraction on a target image to obtain target image features, wherein the target image is synthesized based on the second pathological image and the third pathological image; performing image segmentation on the target image based on the target image feature to obtain a segmented image, wherein the segmented image is respectively marked with a lesion area of the first lesion type and a lesion area of a second lesion type, where the second lesion type refers to a lesion type other than the first lesion type; The minimum value of the first parameter range is the first absorption parameter, and the maximum value is the second absorption parameter. Extracting the second pathological image from the first pathological image includes: adjusting an absorption parameter smaller than the first absorption parameter among a plurality of absorption parameters corresponding to the first pathological image to the first absorption parameter, and adjusting an absorption parameter larger than the second absorption parameter among the plurality of absorption parameters to the second absorption parameter, to obtain the second pathological image; The minimum value of the second parameter range is the third absorption parameter, and the maximum value is the fourth absorption parameter. Extracting the third pathological image from the first pathological image includes: The absorption parameters smaller than the third absorption parameter among the multiple absorption parameters corresponding to the first pathological image are adjusted to the third absorption parameter, and the absorption parameters larger than the fourth absorption parameter among the multiple absorption parameters are adjusted to the fourth absorption parameter to obtain the third pathological image.

2. The method according to claim 1, characterized in that The target object is a head. After acquiring the first pathological image of the target object, the method further includes: extracting a fourth pathological image from the first pathological image, wherein the absorption parameter of a position point in the fourth pathological image belongs to a third parameter range, and the third parameter range refers to a range to which the absorption parameter of the subdural area belongs; Before extracting features from the target image to obtain target image features, the method further includes: The target image is synthesized based on the second pathological image, the third pathological image, and the fourth pathological image.

3. The method according to claim 1, characterized in that The step of performing image segmentation on the target image based on the target image features to obtain a segmented image includes: Determining, based on the target image feature, a first probability and a second probability corresponding to a position point in the target image, wherein the first probability indicates a likelihood that the position point belongs to the first lesion type, and the second probability indicates a likelihood that the position point belongs to the second lesion type; Determine a maximum probability between a first probability and a second probability corresponding to the position point in the target image, and determine the type represented by the maximum probability as the type to which the position point belongs; Based on the type of the position points in the target image, the position points in the target image are marked to obtain the segmented image.

4. The method according to claim 1, wherein The step of performing image segmentation on the target image based on the target image features to obtain a segmented image includes: Determining, based on the target image features, a first probability, a second probability, and a third probability corresponding to a position point in the target image, wherein the first probability indicates a likelihood that the position point belongs to the first lesion type, the second probability indicates a likelihood that the position point belongs to the second lesion type, and the third probability indicates a likelihood that the position point belongs to a non-lesion type; Determine a maximum probability among a first probability, a second probability, and a third probability corresponding to the position point in the target image, and determine the type represented by the maximum probability as the type to which the position point belongs; Based on the type of the position points in the target image, the position points in the target image are marked to obtain the segmented image.

5. The method according to claim 1, wherein The target object is a head, the skull removal model includes a first downsampling sub-model, a first upsampling sub-model, and a skull removal sub-model, and before extracting the second pathological image and the third pathological image from the first pathological image, the method further includes: calling the first downsampling sub-model to downsample the first pathological image to obtain a first image feature; calling the first upsampling sub-model to upsample the first image feature to obtain a second image feature, where the second image feature represents a fourth probability corresponding to a location point in the first pathological image, where the fourth probability represents a likelihood that the location point belongs to a skull type; The skull removal sub-model is called, and when the fourth probability corresponding to the position point in the first pathological image is greater than the reference probability, the position point is determined as a skull position point, the skull position point in the first pathological image is removed, and the first pathological image after skull removal is obtained.

6. The method according to any one of claims 1 to 5, characterized in that The image segmentation model includes a second downsampling sub-model, a second upsampling sub-model and an image segmentation sub-model. The feature extraction of the target image to obtain the target image features includes: calling the second downsampling sub-model to downsample the target image to obtain a third image feature; calling the second upsampling sub-model to upsample the third image feature to obtain the target image feature; The step of performing image segmentation on the target image based on the target image features to obtain a segmented image includes: The image segmentation sub-model is called to perform image segmentation on the target image based on the target image features to obtain the segmented image.

7. The method according to claim 6, characterized in that The training process of the image segmentation model includes the following steps: Acquire a sample pathological image and a sample segmentation image corresponding to the sample pathological image, wherein the sample segmentation image is respectively marked with a sample lesion region of the first lesion type and a sample region of the second lesion type; calling the image segmentation model to process the sample pathological image to obtain a predicted segmented image, wherein the predicted segmented image is respectively marked with a predicted region of the first lesion type and a predicted region of the second lesion type; Based on the loss function, the predicted segmentation image and the sample segmentation image are processed to obtain a loss value, and the image segmentation model is trained based on the loss value.

8. The method according to claim 7, characterized in that The loss function includes at least one of a first loss function, a second loss function, and a third loss function; The first loss function represents a relationship between an actual type of a position point in the sample segmented image and a predicted type of a position point in the predicted segmented image, and a first loss value, wherein the actual type refers to a type of a sample region containing the position point, and the predicted type refers to a type of a predicted region containing the position point. The second loss function represents a relationship between a sample region of the first lesion type in the sample segmented image and a predicted region of the first lesion type in the predicted segmented image, and a second loss value; The third loss function represents the relationship between the sample area of the second lesion type in the sample segmentation image and the predicted area of the second lesion type in the predicted segmentation image, and a third loss value.

9. The method according to any one of claims 1 to 5, characterized in that Before extracting features from the target image to obtain target image features, the method further includes: The absorption parameters of the position points in the second pathological image and the absorption parameters of the position points in the third pathological image are normalized respectively.

10. An image segmentation device, characterized in that: The device comprises: an image acquisition module, configured to acquire a first pathological image of a target object, wherein the first pathological image is obtained by scanning the target object using radiation, wherein a position point in the first pathological image has an absorption parameter, wherein the absorption parameter indicates a degree of absorption of the radiation by the position point in the target object; an image extraction module, configured to extract a second pathological image and a third pathological image from the first pathological image, respectively, wherein the absorption parameters of the position points in the second pathological image belong to a first parameter range, which refers to a range within which the absorption parameters of the lesion site belong; and the absorption parameters of the position points in the third pathological image belong to a second parameter range, which refers to a range within which the absorption parameters of the lesion site of the first lesion type belong; a feature extraction module, configured to extract features from a target image to obtain features of the target image, wherein the target image is synthesized based on the second pathological image and the third pathological image; an image segmentation module, configured to perform image segmentation on the target image based on the target image features to obtain a segmented image, wherein the segmented image is respectively marked with lesion areas of the first lesion type and lesion areas of the second lesion type, where the second lesion type refers to lesion types other than the first lesion type; The minimum value of the first parameter range is the first absorption parameter, and the maximum value is the second absorption parameter. The image extraction module is used to adjust the absorption parameters smaller than the first absorption parameter among the multiple absorption parameters corresponding to the first pathological image to the first absorption parameter, and adjust the absorption parameters larger than the second absorption parameter among the multiple absorption parameters to the second absorption parameter, to obtain the second pathological image; The minimum value of the second parameter range is the third absorption parameter, and the maximum value is the fourth absorption parameter. The image extraction module is used to adjust the absorption parameters smaller than the third absorption parameter among the multiple absorption parameters corresponding to the first pathological image to the third absorption parameter, and adjust the absorption parameters greater than the fourth absorption parameter among the multiple absorption parameters to the fourth absorption parameter, so as to obtain the third pathological image.

11. The device according to claim 10, characterized in that The target object is a head, and the device further includes: The image extraction module is further configured to extract a fourth pathological image from the first pathological image, wherein the absorption parameter of the position point in the fourth pathological image belongs to a third parameter range, and the third parameter range refers to a range to which the absorption parameter of the subdural area belongs; An image synthesis module is configured to synthesize the target image based on the first pathological image, the second pathological image, and the third pathological image.

12. The device according to claim 10, characterized in that The image segmentation module comprises: a probability determination unit, configured to determine, based on the target image feature, a first probability and a second probability corresponding to a position point in the target image, wherein the first probability indicates a possibility that the position point belongs to the first lesion type, and the second probability indicates a possibility that the position point belongs to the second lesion type; a type determination unit, configured to determine a maximum probability between a first probability and a second probability corresponding to a position point in the target image, and determine the type represented by the maximum probability as the type to which the position point belongs; The image segmentation unit is used to mark the position points in the target image based on the types of the position points in the target image to obtain the segmented image.

13. The device according to claim 10, characterized in that The image segmentation module comprises: a probability determination unit, configured to determine, based on the target image feature, a first probability, a second probability, and a third probability corresponding to a position point in the target image, wherein the first probability indicates a possibility that the position point belongs to the first lesion type, the second probability indicates a possibility that the position point belongs to the second lesion type, and the third probability indicates a possibility that the position point belongs to a non-lesion type; a type determination unit, configured to determine a maximum probability among a first probability, a second probability, and a third probability corresponding to a position point in the target image, and determine the type represented by the maximum probability as the type to which the position point belongs; The image segmentation unit is used to mark the position points in the target image based on the types of the position points in the target image to obtain the segmented image.

14. The device according to claim 10, characterized in that The target object is a head, the skull removal model includes a first downsampling sub-model, a first upsampling sub-model and a skull removal sub-model, and the device further includes: a skull removal module, configured to call the first downsampling sub-model to downsample the first pathological image to obtain a first image feature; The skull removal module is further configured to call the first upsampling sub-model to upsample the first image feature to obtain a second image feature, where the second image feature represents a fourth probability corresponding to a position point in the first pathological image, and the fourth probability represents a possibility that the position point belongs to a skull type; The skull removal module is also used to call the skull removal sub-model. When the fourth probability corresponding to the position point in the first pathological image is greater than the reference probability, the position point is determined as a skull position point, the skull position point in the first pathological image is removed, and the first pathological image after skull removal is obtained.

15. The device according to any one of claims 10 to 14, characterized in that The image segmentation model includes a second downsampling sub-model, a second upsampling sub-model and an image segmentation sub-model, and the feature extraction module is used to: calling the second downsampling sub-model to downsample the target image to obtain a third image feature; calling the second upsampling sub-model to upsample the third image feature to obtain the target image feature; The image segmentation module is used to call the image segmentation sub-model and perform image segmentation on the target image based on the target image features to obtain the segmented image.

16. The device according to claim 15, characterized in that The device further comprises: a model training module, configured to obtain a sample pathology image and a sample segmentation image corresponding to the sample pathology image, wherein the sample segmentation image is respectively marked with a sample lesion region of the first lesion type and a sample region of the second lesion type; The model training module is further configured to call the image segmentation model to process the sample pathological image to obtain a predicted segmented image, wherein the predicted segmented image is respectively marked with a predicted area of the first lesion type and a predicted area of the second lesion type; The model training module is further used to process the predicted segmentation image and the sample segmentation image based on a loss function to obtain a loss value, and train the image segmentation model based on the loss value.

17. The device according to claim 16, characterized in that The loss function includes at least one of a first loss function, a second loss function, and a third loss function; The first loss function represents a relationship between an actual type of a position point in the sample segmented image and a predicted type of a position point in the predicted segmented image, and a first loss value, wherein the actual type refers to a type of a sample region containing the position point, and the predicted type refers to a type of a predicted region containing the position point. The second loss function represents a relationship between a sample region of the first lesion type in the sample segmented image and a predicted region of the first lesion type in the predicted segmented image, and a second loss value; The third loss function represents the relationship between the sample area of the second lesion type in the sample segmentation image and the predicted area of the second lesion type in the predicted segmentation image, and a third loss value.

18. The device according to any one of claims 10 to 14, characterized in that The device further comprises: A normalization module is used to normalize the absorption parameters of the position points in the second pathological image and the position points in the third pathological image respectively.

19. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the image segmentation method according to any one of claims 1 to 9.

20. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the image segmentation method according to any one of claims 1 to 9.

21. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the operations performed by the image segmentation method according to any one of claims 1 to 9 are realized.

Images