A method for brain partitioning of cranial brain hemorrhage images and related products

Abstract

The application discloses a method for brain partition of a cranial hemorrhage image and related products. The method comprises the following steps: acquiring a cranial hemorrhage image and a healthy cranial image; performing image fusion on the cranial hemorrhage image and the healthy cranial image to obtain a fusion image; and inputting the fusion image and random noise into a trained generative model, and generating a brain partition result of the cranial hemorrhage image under the guidance of a hemorrhagic focus mask image. The scheme can improve the efficiency of brain partition and improve the accuracy of brain partition.

Description

Technical Field

[0001] This application generally relates to the field of brain partitioning technology. More specifically, this application relates to a method, apparatus, and computer-readable storage medium for brain partitioning of images of intracranial hemorrhage. Background Technology

[0002] Intracranial hemorrhage is often accompanied by brain tissue deformation, and the boundaries between different brain regions become blurred, making accurate segmentation and evaluation of these regions difficult. Brain region segmentation in intracranial hemorrhage imaging refers to the division and labeling of brain regions in images obtained through processing and analysis of imaging data from patients with intracranial hemorrhage. This helps doctors better understand the location, size, and morphological characteristics of the hemorrhage, providing precise quantitative information and visualization results, thus offering crucial information for clinical diagnosis and treatment decisions.

[0003] Currently, brain segmentation of intracranial hemorrhage images often relies on doctors' knowledge of the 3D anatomy of the brain, with doctors manually dividing the images into brain regions. However, this method is time-consuming and labor-intensive, and easily influenced by subjective factors. Due to the complex morphology and varying sizes of intracranial hemorrhages, as well as differences in image quality, different doctors may draw boundary lines based on personal experience and subjective judgment, leading to variations in the consistency and reliability of the results. Furthermore, manual segmentation is time-consuming, especially when segmenting large-scale image datasets, resulting in low efficiency. Additionally, manual segmentation requires specialized knowledge and skills, and some regions or medical institutions may lack sufficient personnel to perform this task.

[0004] Therefore, there is an urgent need to provide a scheme for brain partitioning of intracranial hemorrhage images in order to obtain brain partitioning results of intracranial hemorrhage images efficiently and accurately. Summary of the Invention

[0005] In order to at least address one or more of the technical problems mentioned above, this application proposes a scheme for brain partitioning in images of intracranial hemorrhage in several aspects.

[0006] In a first aspect, this application provides a method for brain partitioning of intracranial hemorrhage images, comprising: acquiring intracranial hemorrhage images and healthy brain images; performing image fusion on the intracranial hemorrhage images and the healthy brain images to obtain fused images; and inputting the fused images and random noise into a trained generative model, and generating brain partitioning results of the intracranial hemorrhage images under the guidance of hemorrhage foci mask images.

[0007] In some embodiments, the image fusion of the intracranial hemorrhage image and the healthy intracranial image to obtain a fused image includes: image registration of the healthy intracranial image and the intracranial hemorrhage image to obtain registration data; and image fusion of the registration data and the intracranial hemorrhage image to obtain the fused image.

[0008] In other embodiments, image fusion of the registration data and the intracranial hemorrhage images to obtain the fused image includes: performing an arithmetic average or a weighted average of the registration data and the intracranial hemorrhage images to obtain the fused image.

[0009] In some other embodiments, the process of inputting the fused image and random noise into a trained generative model and generating brain partitioning results of the intracranial hemorrhage image under the guidance of the hemorrhage foci mask image includes: generating an intermediate image using the trained generative model based on the fused image and the random noise; and performing a background masking operation on the intermediate image using the hemorrhage foci mask image to generate brain partitioning results of the intracranial hemorrhage image under the guidance of the hemorrhage foci mask image.

[0010] In some other embodiments, performing a background masking operation on the intermediate image using the hemorrhage mask image includes: multiplying the hemorrhage mask image with the intermediate image to perform a background masking operation on the intermediate image using the hemorrhage mask image.

[0011] In some other embodiments, the hemorrhage mask image includes a binary map or a probability map.

[0012] In some other embodiments, the trained generative model is obtained by: acquiring a healthy brain image and labeling the healthy brain image with brain regions; adding a noise distribution to the brain region labeling and inputting the added noise distribution and the healthy brain image into the generative model for feature extraction, so that the generative model has the ability to generate brain region results from the noise space based on the healthy brain image, thereby obtaining the trained generative model.

[0013] In a second aspect, this application provides an apparatus for brain partitioning of images of intracranial hemorrhage, comprising: a processor; and a memory storing program instructions for brain partitioning of images of intracranial hemorrhage, wherein when the program instructions are executed by the processor, the apparatus causes the apparatus to implement the various embodiments of the first aspect described above.

[0014] In a third aspect, this application provides a computer-readable storage medium storing computer-readable instructions for brain regions of images of intracranial hemorrhage, which, when executed by one or more processors, implement the various embodiments of the first aspect described above.

[0015] The present application provides a brain partitioning scheme for intracranial hemorrhage images through the above-described aspects and embodiments. This scheme involves inputting a fused image (e.g., a fusion of intracranial hemorrhage images and healthy intracranial images) and random noise into a trained generative model. Guided by a mask image of the hemorrhage foci, the model generates brain partitioning results for the intracranial hemorrhage images. By inputting the fused image into the generative model, it closely approximates the healthy intracranial images used in the model training process while preserving hemorrhage information, thereby improving the accuracy of brain partitioning. Furthermore, by using a mask image of the hemorrhage foci to guide the generation of brain partitioning results for the intracranial hemorrhage images, the extracted information near the hemorrhage becomes more complete, further improving the accuracy of brain partitioning. This method avoids manual segmentation and significantly improves the efficiency of brain partitioning for intracranial hemorrhage images. Attached Figure Description

[0016] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts, wherein:

[0017] Figure 1 This is an exemplary flowchart illustrating a method for brain partitioning of images of intracranial hemorrhage according to an embodiment of this application;

[0018] Figure 2 This is an exemplary flowchart illustrating the overall brain region for imaging of intracranial hemorrhage according to an embodiment of this application;

[0019] Figure 3 This is an exemplary flowchart illustrating the training and generation of a model according to an embodiment of this application;

[0020] Figure 4 This is a schematic structural block diagram illustrating a device for imaging brain regions of intracranial hemorrhage according to an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] It should be understood that the terms "comprising" and "including" used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0024] As used in this specification and claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0025] The specific embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0026] Figure 1 This is an exemplary flowchart illustrating a method 100 for brain partitioning of images of intracranial hemorrhage according to an embodiment of this application, which will be described below in conjunction with... Figure 1 The methods described in this application will be further explained. For example... Figure 1 As shown, in step S101, images of intracranial hemorrhage and healthy brain are acquired. It is important to understand that before delving into the content and scope of method 100, it is necessary to clarify its related basic concepts and applications.

[0027] Intracranial hemorrhage is an abnormal pooling of blood in the brain tissue. It can be caused by head trauma, ruptured arteries, or other damage to intracranial blood vessels. The presence of intracranial hemorrhage can pose a serious threat to a patient's life and health; therefore, accurate and timely detection and diagnosis of intracranial hemorrhage are crucial.

[0028] In the implementation scenario, the aforementioned images of intracranial hemorrhage and healthy brain are images acquired through medical imaging techniques, used to diagnose and assess lesions and diseases of the brain. Common brain images include those acquired through X-rays, computed tomography (CT), or magnetic resonance imaging (MRI). These images provide detailed information about brain structure and function, helping doctors observe and analyze brain structure, tissue, and function. Through these images, doctors can identify and assess important information such as the presence, location, size, and type of intracranial hemorrhage. Image data is typically represented in matrix form, where each element represents the grayscale value or signal intensity of a pixel. Therefore, segmentation of different regions in intracranial hemorrhage images is crucial for evaluating the extent of brain tissue damage. In some embodiments, the aforementioned healthy brain images can be obtained from a large number of healthy brain CT images. These brain CT image data can cover healthy individuals of different ages, sexes, and ethnicities to ensure the representativeness and comprehensiveness of the template.

[0029] Based on the obtained images of intracranial hemorrhage and healthy brain, in step S102, the images of intracranial hemorrhage and healthy brain are fused to obtain a fused image. This allows the fused image to closely resemble the healthy brain image obtained during model training while preserving hemorrhage information, thereby improving the accuracy of brain region mapping.

[0030] In one embodiment, healthy brain images and images of intracranial hemorrhage are image registered to obtain registration data. Then, the registration data and the intracranial hemorrhage images are fused to obtain a fused image. That is, this embodiment of the application obtains a fused image by registering healthy brain images and images of intracranial hemorrhage, and then fusing the registration result with the intracranial hemorrhage images. For registration, a transformation matrix between the healthy brain images and the intracranial hemorrhage images can first be calculated. Then, based on the transformation matrix, data transformed from the healthy brain images to the intracranial hemorrhage images, i.e., registration data, is calculated. Further, the registration data and the intracranial hemorrhage images are fused to obtain the fused image. In one implementation scenario, the registration data and the intracranial hemorrhage images can be averaged arithmetically or by weighted averaging to obtain the fused image.

[0031] As an example, suppose the transformation matrix is ​​X, and X(T) is the registration data for transforming a healthy brain image into a brain hemorrhage image. This is equivalent to the image structure being similar to X(T) if the data did not show hemorrhage. In this scenario, by denoting the brain hemorrhage image as I, and performing arithmetic mean (I+X(T)) / 2 or weighted mean on the brain hemorrhage image I and the registration data X(T), a fused image can be obtained. Based on this, by finding a transformation matrix X that allows the healthy brain image to be accurately mapped into the space of the brain hemorrhage image, the representation X(T) of the healthy brain image in the brain hemorrhage image space is obtained. In this process, if the image data does not show hemorrhage, then the brain hemorrhage image is similar to X(T). Furthermore, obtaining the fused image through arithmetic mean (I+X(T)) / 2 or weighted mean, for example, allows the fused image to contain both the features of the healthy brain data and the information of hemorrhage, thereby improving the accuracy of brain partitioning.

[0032] After obtaining the above fused image, in step S103, the fused image and random noise are input into the trained generative model, and the brain partitioning result of the intracranial hemorrhage image is generated under the guidance of the hemorrhage foci mask image.

[0033] In one embodiment, the aforementioned trained generative model can be performed by acquiring a healthy brain image, labeling the healthy brain image into brain regions, adding a noise distribution to the brain region labeling, and inputting the added noise distribution and the healthy brain image into the generative model for feature extraction. This enables the generative model to generate brain region results from the noise space based on the healthy brain image, thus obtaining a trained generative model. Specifically, any noise distribution (e.g., Gaussian noise) can be added to the brain region labeling to obtain a noise-added brain region labeling. Alternatively, multiple noise distributions can be added to the brain region labeling based on a predetermined signal-to-noise ratio to obtain multiple noise-added brain region labels. By inputting the added noise distribution and the healthy brain image together into the generative model for feature extraction, the training of the generative model maps the healthy brain image from image space to Gaussian space and then back from Gaussian space to image space to obtain brain region results, thereby completing the training of the generative model.

[0034] For example, when adding a noise distribution, a brain region with added noise (e.g., Xn) can be labeled to a Gaussian space. Regarding setting multiple noise distributions based on a predetermined signal-to-noise ratio, in an exemplary scenario, assume the image space has a signal-to-noise ratio of 1 (i.e., no noise, denoted by S0), the Gaussian space has a signal-to-noise ratio of 0 (i.e., pure noise, denoted by Sn), and the Gaussian spaces corresponding to sequentially designed signal-to-noise ratios s1, s2, ..., sn-1 can be represented as S1, S2, ..., Sn-1, where 1>s1>s2>...>sn-1>0. Specifically, the labeled brain region with added noise is used as input to the generative model, a healthy brain image is used as another input, and the brain region result is used as the expected output of the generative model to train the model. This allows the generative model to map features from one or more Gaussian spaces (e.g., S0, S1, S2, ..., Sn-1, Sn) back to the image space to obtain the brain region result, thus completing the training of the generative model.

[0035] In some embodiments, before training the generative model, preprocessing operations can be performed on the added noise distribution and healthy brain images to give the data a uniform format that facilitates processing by the generative model. These preprocessing operations include, but are not limited to, data format conversion, data size adjustment, and data augmentation. Furthermore, the generative model can be, for example, a Unet model, an FPN model, or a PSPNet model.

[0036] Based on a trained generative model, fused images and random noise can be input into the trained generative model, and brain partitioning results of intracranial hemorrhage images can be generated under the guidance of hemorrhage foci mask images. In one embodiment, an intermediate image can be obtained by using a trained generative model to generate an image based on fused images and random noise, and then the intermediate image can be subjected to background masking operation using the hemorrhage foci mask image to generate brain partitioning results of intracranial hemorrhage images under the guidance of the hemorrhage foci mask image. In some embodiments, the aforementioned random noise is Gaussian noise. The aforementioned hemorrhage foci mask image can be obtained by performing image segmentation operation on the hemorrhage foci mask in intracranial hemorrhage images, for example, using an image segmentation model. That is, in the process of mapping from noise space back to image space (or brain partitioning space), this application uses fused images as another input to the generative model to extract features from fused images to form intermediate images, and further uses hemorrhage foci mask images to perform background masking operation on the intermediate images to generate brain partitioning results of hemorrhage intracranial images.

[0037] In one implementation scenario, the hemorrhage foci mask image is multiplied with an intermediate image to perform a background masking operation on the intermediate image using the hemorrhage foci mask image. The hemorrhage foci mask image can be a binary map or a probabilistic map. It's important to understand that for a given feature space, an intermediate image x0 is generated by fusing the image and random noise. In this scenario, assuming the hemorrhage foci mask image is denoted as mask, the brain region result x0' of the hemorrhage brain image can be directly generated based on x0*mask.

[0038] For multiple feature spaces, the brain partitioning results of the hemorrhagic brain image can be generated step by step from Sn->Sn-1, Sn-1->Sn-2, ...,S1->S0, corresponding to multiple intermediate images such as xn-1, xn-2, ...,x0. For each intermediate image, a background masking operation is performed using a hemorrhage foci mask image to obtain the brain partitioning result x0 of the hemorrhagic brain image. For example, for the intermediate image xn-1, the final image xn-1' corresponding to Sn-1 is obtained by multiplying xn-1 by mask. Further, based on the final image xn-1' corresponding to Sn-1, the image xn-2' corresponding to Sn-2 is extracted, and the process of Sn-1->Sn-2, ...,S1->S0 is performed until x0' is generated, thus obtaining the brain partitioning result of the hemorrhagic brain image.

[0039] As described above, the embodiments of this application input a fused image obtained by fusing intracranial hemorrhage images with healthy intracranial images and random noise into a trained generative model, and generate brain partitioning results of intracranial hemorrhage images under the guidance of hemorrhage foci mask images. This allows the hemorrhage information to be preserved and complete information near the hemorrhage to be extracted, thereby improving the accuracy and efficiency of brain partitioning.

[0040] Figure 2 This is an exemplary flowchart illustrating the overall brain region for imaging of intracranial hemorrhage according to an embodiment of this application. Figure 2 As shown, in step S201, images of intracranial hemorrhage and healthy brain are acquired. As previously mentioned, these images can be acquired using techniques such as X-ray imaging, CT, or MRI. Next, in step S202, the healthy brain image and the intracranial hemorrhage image are image registered to obtain registration data.

[0041] It should be noted that image registration involves matching multiple (≥2) images taken in the same scene but at different times, from different perspectives, or with different acquisition devices. For example, assuming there are two images A and B, image registration aims to find a spatial transformation to transform A onto B, so that points corresponding to the same spatial location in the two images correspond one-to-one, thereby achieving information fusion. In this embodiment, by calculating the transformation matrix between healthy cranial images and images of intracranial hemorrhage, the data of healthy cranial images is transformed into images of intracranial hemorrhage based on the transformation matrix, thus achieving the fusion of healthy cranial images and images of intracranial hemorrhage.

[0042] Further, in step S203, the registration data and the intracranial hemorrhage images are fused to obtain a fused image. In one embodiment, the registration data and the intracranial hemorrhage images are averaged arithmetically or by weighted average to obtain the fused image. This ensures that the fused image closely resembles the healthy brain images obtained during model training while retaining hemorrhage information, thereby improving the accuracy of brain region mapping.

[0043] The figure further illustrates that in step S204, noise is randomly introduced, and in step S205, the fused image and random noise are input into the trained generative model. During the mapping from the noise space back to the image space (or brain partitioning space), the fused image is used as another input to the generative model to extract features from the fused image to form an intermediate image. As mentioned earlier, brain partitioning results of hemorrhagic brain images can be generated directly using a single feature space, or multiple feature spaces can be set, and brain partitioning results of hemorrhagic brain images can be generated step-by-step through multiple intermediate images.

[0044] Further, in step S206, a hemorrhage foci mask image is added for guidance to obtain brain region results in step S207. In one implementation scenario, the hemorrhage foci mask image is multiplied with an intermediate image to perform a background masking operation on the intermediate image using the hemorrhage foci mask image, thereby obtaining brain region results of the intracranial hemorrhage image under the guidance of the hemorrhage foci mask image. Based on this, richer information near the hemorrhage is extracted, improving the accuracy of brain region division.

[0045] Figure 3 This is an exemplary flowchart illustrating the training of a generative model according to an embodiment of this application. It is understood that the generative model needs to be trained before performing the process described above of using the trained generative model to partition brain images of intracranial hemorrhage. The input to the generative model is a noise distribution added to the brain partition annotations and a healthy brain image. The learning capabilities during training include: mapping healthy brain images to a noise feature space with certain characteristics and mapping from the noise feature space back to the brain partition image space.

[0046] In the embodiments of this application, the aforementioned generative model may include, but is not limited to, the Unet model, the FPN model, or the PSPNet model. The training process of this generative model includes collecting a large number of healthy cranial images, performing brain region annotation on the healthy cranial images, preprocessing, etc. The generative model learns the characteristics of healthy cranial images and acquires the ability to extract features from healthy cranial images and generate brain region results from Gaussian noise space.

[0047] like Figure 3 As shown, in step S301, healthy cranial images are acquired as a training set. Based on the acquired healthy cranial images, in step S302, brain region annotation is performed on the healthy cranial images, and in step S303, noise distribution is added to the brain region annotations. Next, in step S304, the healthy cranial images are used as another input to the generative model. Before being input into the generative model, in step S305, preprocessing operations are performed on the added noise and the healthy cranial images to give the data a uniform format for easy processing by the generative model. It can be understood that the preprocessing stage aims to remove noise, enhance image quality, and register the spatial and geometric relationships between different images. In some embodiments, the preprocessing operations of this application include, but are not limited to, filtering, enhancement, geometric correction, data format conversion, data size adjustment, and data augmentation.

[0048] Further, in step S306, the preprocessed noise and healthy brain images are input into the generative model for training. Specifically, a brain region label with added noise is mapped to a Gaussian space based on the healthy brain image, or multiple noise distributions are set according to a predetermined signal-to-noise ratio and mapped to multiple Gaussian spaces. Then, features from one or more Gaussian spaces (e.g., S0, S1, S2, ..., Sn-1, Sn) are mapped back to the image space to obtain the brain region result, thereby completing the training of the generative model.

[0049] It's important to understand that Gaussian noise is a type of random noise whose statistical properties follow a Gaussian distribution (also known as a normal distribution). In generative models, adding Gaussian noise as input can generate healthy brain images with a certain degree of variation and diversity. Different sampling methods and strategies can be employed when sampling healthy brain images. In one embodiment, random sampling can be used, for example. That is, a subset of pixels or regions are randomly selected from the healthy brain image as sampling samples. This sampling sample can be adjusted and controlled as needed to meet specific generation requirements. Furthermore, more complex sampling methods, such as uniform sampling and layered sampling, can be used to achieve more accurate sampling results.

[0050] Gaussian noise generation can be simulated using a Gaussian distribution function. The Gaussian distribution function is a continuous probability distribution function, defined by its mean and variance. By adjusting the values ​​of the mean and variance, the distribution characteristics of the generated Gaussian noise can be controlled. The generated Gaussian noise can be obtained using a random number generator, where each random number follows a Gaussian distribution. Based on this, the generative model can extract features from annotated healthy brain images with added noise distribution. During the training of the generative model, the healthy brain images are mapped from image space to Gaussian space, and then back from Gaussian space to image space to obtain brain partitioning results, thus completing the training of the generative model.

[0051] Figure 4 This is a schematic structural block diagram illustrating a device 400 for imaging intracranial hemorrhage according to an embodiment of this application. Figure 4 As shown, the device 400 may include a processor 401 and a memory 402, wherein the processor 401 and the memory 402 communicate via a bus. The memory 402 stores computer instructions for brain partitioning of intracranial hemorrhage images. When the computer instructions are executed by the processor 401, the device 400 performs the method steps described above in conjunction with the accompanying drawings: acquiring intracranial hemorrhage images and healthy brain images; performing image fusion of the intracranial hemorrhage images and the healthy brain images to obtain a fused image; and inputting the fused image and random noise into a trained generative model, and generating brain partitioning results of the intracranial hemorrhage images under the guidance of a hemorrhage foci mask image.

[0052] Based on the foregoing description in conjunction with the accompanying drawings, those skilled in the art will understand that the embodiments of this application can also be implemented by software programs. Therefore, this application also provides a computer-readable storage medium. This computer-readable storage medium stores computer-readable instructions thereon for brain regions of images of intracranial hemorrhage. When executed by one or more processors, these computer-readable instructions implement the embodiments of this application in conjunction with the accompanying drawings. Figure 1 The method described for brain partitioning in images of intracranial hemorrhage.

[0053] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0054] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0055] It should be understood that when the terms "first," "second," "third," and "fourth," etc., are used in the claims, specification, and drawings of this application, they are used only to distinguish different objects and not to describe a specific order. The terms "comprising" and "including" as used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0056] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0057] Although the embodiments of this application are described above, the content is merely an example adopted for the purpose of facilitating understanding of this application and is not intended to limit the scope and application scenarios of this application. Any person skilled in the art described in this application may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.

[0058] Furthermore, the collection and acquisition of various data in this application comply with relevant laws and regulations and are authorized by the data providers. Any organization or individual that needs to obtain external data shall obtain authorization in accordance with the law and ensure data security, and shall not illegally collect, use, process, or transmit unauthorized or unprotected data, nor shall it illegally buy, sell, provide, or disclose unauthorized or unprotected data.

Claims

1. A method for brain region division in images of intracranial hemorrhage, characterized in that, include: Obtain images of intracranial hemorrhage and healthy brain tissue; The images of the intracranial hemorrhage and the images of the healthy brain are fused to obtain a fused image; as well as The fused image and random noise are input into the trained generative model, and brain region results of the intracranial hemorrhage image are generated under the guidance of the hemorrhage foci mask image. The process of fusing the intracranial hemorrhage image with the healthy intracranial image to obtain a fused image includes: The healthy brain image and the brain hemorrhage image are image registered to obtain registration data; The registration data and the intracranial hemorrhage images are averaged or weighted to obtain the fused image. The trained generative model is obtained through the following operations: Acquire healthy cranial images and perform brain region annotation on the healthy cranial images; and A noise distribution is added to the labeled brain regions, and the added noise distribution and the healthy brain image are input into the generative model for feature extraction. This enables the generative model to generate brain region results from the noisy space based on the healthy brain image, thus obtaining the trained generative model. The fused image and random noise are input into the trained generative model, and the brain region results of the intracranial hemorrhage image generated under the guidance of the hemorrhage foci mask image include: Based on the fused image and the random noise, the trained generative model is used to generate an intermediate image. The hemorrhage lesion mask image is multiplied with the intermediate image to perform a background masking operation on the intermediate image using the hemorrhage lesion mask image.

2. The method according to claim 1, characterized in that, The hemorrhage mask image mentioned above includes a binary map or a probability map.

3. A device for brain region mapping in intracranial hemorrhage imaging, characterized in that, include: processor; as well as A memory storing program instructions for brain regions of images of intracranial hemorrhage, which, when executed by the processor, cause the device to implement the method according to any one of claims 1-2.

4. A computer-readable storage medium, characterized in that, It stores computer-readable instructions for brain regions used in images of intracranial hemorrhage, which, when executed by one or more processors, implement the method as described in any one of claims 1-2.

Images