Brain tomography image processing method and device and electronic equipment

Abstract

The invention provides a brain tomography image processing method and device and electronic equipment. The method comprises the steps of performing position adjustment on a first brain tomography image based on a brain reference direction to obtain a second brain tomography image after position adjustment; predicting registration parameters based on the second brain tomography image and a standard template; performing registration processing on the second brain tomography image based on the registration parameter to obtain a third brain tomography image; the third brain tomography image is subjected to tracer recognition, a tracer uptake value corresponding to the brain area is obtained, and the tracer uptake value is used for determining the brain amyloid protein deposition amount. According to the application, the efficiency and accuracy of detecting cerebral amyloid protein deposition can be improved.

Description

Technical Field

[0001] The present application relates to image processing and artificial intelligence technology, and in particular to a method, device and electronic device for processing brain tomography images. Background Art

[0002] Quantitative assessment of brain amyloid β (Aβ) deposition is of great significance in the research and diagnosis of neurodegenerative diseases such as Alzheimer's disease. In related technologies, the detection of brain amyloid deposition relies on magnetic resonance imaging (MRI) images and complex spatial normalization processes. This process involves processing of image rotation, translation, and flipping patterns, and usually requires manual intervention for coarse registration. The calculation process is complex and time-consuming, which limits the application of brain amyloid deposition detection in large-scale data analysis. Summary of the invention

[0003] The embodiments of the present application provide a method, device, electronic device, computer-readable storage medium and computer program product for processing brain tomography images, which can improve the efficiency and accuracy of detecting brain amyloid protein deposition.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] The present application provides a method for processing a brain tomography image, the method comprising:

[0006] Adjusting the position of the first brain tomography image based on the brain reference direction to obtain a second brain tomography image after the position adjustment;

[0007] predicting registration parameters based on the second brain tomography image and a standard template;

[0008] performing registration processing on the second brain tomography image based on the registration parameters to obtain a third brain tomography image;

[0009] The third brain tomography image is subjected to tracer identification to obtain a tracer uptake value corresponding to the brain region, wherein the tracer uptake value is used to determine the amount of amyloid protein deposition in the brain.

[0010] The present application provides a brain tomography image processing device, comprising:

[0011] A position adjustment module, used for adjusting the position of the first brain tomography image based on the brain reference direction to obtain a second brain tomography image after position adjustment;

[0012] A registration module, used for predicting registration parameters based on the second brain tomography image and a standard template;

[0013] The registration module is further used to perform registration processing on the second brain tomography image based on the registration parameters to obtain a third brain tomography image;

[0014] The identification module is used to perform tracer identification on the third brain tomography image to obtain a tracer uptake value corresponding to the brain region, wherein the tracer uptake value is used to determine the amount of amyloid protein deposition in the brain.

[0015] An embodiment of the present application provides an electronic device, the electronic device comprising:

[0016] A memory for storing computer executable instructions or computer programs;

[0017] The processor is used to implement the brain tomography image processing method provided in the embodiment of the present application when executing the computer executable instructions or computer program stored in the memory.

[0018] An embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions or a computer program, which is used to implement the brain tomography image processing method provided in the embodiment of the present application when executed by a processor.

[0019] An embodiment of the present application provides a computer program product, including a computer program or computer executable instructions. When the computer program or computer executable instructions are executed by a processor, the method for processing brain tomography images provided in the embodiment of the present application is implemented.

[0020] The embodiments of the present application have the following beneficial effects:

[0021] By adjusting the position of the brain tomography image, the brain tomography image is matched with the reference direction, the error in the subsequent registration process is reduced. The registration parameters are predicted based on the brain tomography image and the standard template. Compared with the manual registration scheme in the related technology, the accuracy and efficiency of the registration can be improved, and then the tracer uptake value is detected based on the registered brain tomography image, thereby improving the accuracy and efficiency of determining the amount of amyloid protein deposition in the brain. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a schematic diagram of an application mode of a method for processing a brain tomography image provided in an embodiment of the present application;

[0023] Figure 2 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0024] Figure 3A is a first flow chart of a method for processing a brain tomography image provided in an embodiment of the present application;

[0025] Figure 3B is a second flow chart of the method for processing brain tomography images provided in an embodiment of the present application;

[0026] Figure 4 is a third flow chart of the method for processing brain tomography images provided in an embodiment of the present application;

[0027] Figure 5 It is a schematic diagram of the principle of image processing provided by an embodiment of the present application;

[0028] Fig. 6A is a prediction result diagram of the anterior commissure position and reference direction provided in an embodiment of the present application;

[0029] Figure 6B It is a schematic diagram of affine transformation and elastic transformation provided in an embodiment of the present application;

[0030] Fig. 7A is a schematic diagram of the linear correlation provided in the embodiment of the present application;

[0031] Figure 7B is a first schematic diagram for comparing experimental results provided in the embodiments of the present application;

[0032] Figure 7C is a second schematic diagram for comparing experimental results provided in the embodiments of the present application;

[0033] Fig.7D is a third schematic diagram for comparing experimental results provided in the embodiments of the present application;

[0034] Figure 8 It is a schematic diagram of the interface of the application provided in the embodiment of the present application.

[0035] It should be pointed out that the above-mentioned "first" and "second" are only used to distinguish different solutions, and do not represent the degree of superiority or inferiority of the solutions or the priority in the implementation process. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0037] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0038] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0039] It should be pointed out that the collection and processing of relevant data in this application (for example, the user's positron tomography images and magnetic resonance imaging images) should be strictly in accordance with the requirements of relevant laws and regulations when applied in practice, and the informed consent or separate consent of the personal information subject should be obtained, and subsequent data use and processing should be carried out within the scope of authorization of laws and regulations and the personal information subject.

[0040] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0042] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0043] 1) Centiloid: is a unit used to measure the amount of amyloid beta deposits in the brain of patients with Alzheimer's disease (AD).

[0044] 2) Centiloid Scale: This is a method used to assess the accumulation of amyloid plaques (Aβ plaques) in the brains of patients with Alzheimer's disease. Aβ plaques are protein aggregates associated with Alzheimer's disease. The Centiloid Scale can quantify the deposition of Aβ plaques in the brain using positron emission tomography (PET) imaging technology.

[0045] 3) AβPET image: It is a positron emission tomography (PET) image specifically used for the diagnosis and evaluation of Alzheimer's disease (AD). AβPET images are performed by using specific radioactive tracers to label beta-amyloid protein (Aβ) in the brain. These tracers are able to bind to Aβ and then show the distribution of Aβ in PET scans. This technology can help doctors detect Aβ deposition in the brain, which is a major pathological feature of AD. The detection of Aβ deposition is crucial for the early diagnosis of AD, as early diagnosis is of great significance for the intervention and treatment of the disease.

[0046] 4) Spatial normalization: The process of matching the subject's brain structure image to a standard template by linear or nonlinear methods in the post-processing of brain functional magnetic resonance imaging. First, the high-resolution structural image is normalized, and then the functional image is normalized based on this parameter.

[0047] 5) Montreal Neurological Institute space: This is a standardized brain image space, hereinafter referred to as MNI space, which is a widely used reference framework in the field of neuroimaging. MNI space was developed by the Montreal Neurological Institute to provide a unified coordinate system for brain images to facilitate comparison and analysis between different studies.

[0048] 6) 3DSlicer is an open source, multi-platform medical imaging software. It provides a wide range of tools for processing, analyzing and visualizing medical imaging data such as MRI, CT, PET, etc. 3DSlicer was developed by a group of interdisciplinary researchers to provide a powerful and flexible platform for medical imaging researchers, doctors and clinicians.

[0049] 7) Anterior Commissure: It is a structure in the brain located near the midline between the cerebral hemispheres and is the intersection of some nerve fibers connecting the left and right cerebral hemispheres. It is an important landmark in neuroimaging and neuroanatomy and is used in the brain registration and standardization process.

[0050] 8) Anterior-Posterior: An anatomical term, where the anterior refers to the front or ventral direction of the body or structure, and the posterior refers to the back or dorsal direction of the body or structure.

[0051] 9) Superior-Inferior direction: An anatomical term, where Superior refers to the top or head of a body or structure, and Inferior refers to the bottom or feet of a body or structure.

[0052] 10) Voxel: A volume element that represents a tiny cubic unit in three-dimensional space. In PET imaging, voxel is the basic unit of data acquisition and processing, used to describe the concentration of radioactive tracers at each location in the image. A voxel is a point in three-dimensional space that has three-dimensional coordinates (x, y, z) and represents the tracer concentration or radioactivity count at that point. In PET images, the size of a voxel is usually fixed and is determined by the resolution of the image.

[0053] Quantitative assessment of brain amyloid β (Aβ) deposition is of great significance in the research and diagnosis of neurodegenerative diseases such as Alzheimer's disease. In related technologies, the detection and processing of brain amyloid deposition relies on magnetic resonance imaging images and complex spatial normalization processes. This process involves the processing of image change patterns such as rotation, translation, and flipping. Due to the diverse image change patterns, manual rough alignment is usually required, such as the need to use affine transformation to rotate and translate the image in advance. In related technologies, there is currently no better way to improve computational efficiency and the stability of quantitative results.

[0054] The embodiments of the present application provide a method for processing brain tomography images, a device for processing brain tomography images, an electronic device, a computer-readable storage medium, and a computer program product, which can improve the efficiency and accuracy of detecting brain amyloid protein deposition.

[0055] The following describes an exemplary application of the electronic device provided in the embodiments of the present application. The electronic device provided in the embodiments of the present application can be implemented as a terminal device, such as a laptop computer, a tablet computer, a desktop computer, a set-top box, a smart TV, a vehicle terminal, a virtual reality (VR) device, an augmented reality (AR) device, and other types of terminals, and can also be implemented as a server. The following describes an exemplary application when the electronic device is implemented as a terminal device or a server.

[0056] refer to Figure 1 , Figure 1 is a schematic diagram of an application mode of a method for processing a brain tomography image provided in an embodiment of the present application; for example, Figure 1 The invention involves a server 200, a network 300, a terminal device 400, and a database 500. The terminal device 400 is connected to the server 200 via the network 300, and the network 300 can be a wide area network or a local area network, or a combination of the two.

[0057] For example, the server 200 may be a server of a medical platform, and a large amount of medical image-related data is stored in the database 500, including pre-configured masks provided by the Centiloid scale method required for brain amyloid deposition detection. The user may be a doctor or a researcher. The terminal device 400 may be a computer or a mobile phone of the user.

[0058] For example, the user sends the positron tomography image of the brain to be detected to the server 200 through the terminal device 400 and the network 300. The server 200 calls the brain tomography image processing method provided in the embodiment of the present application to process the positron tomography image of the brain to be detected to obtain the registered positron tomography image. Based on the preconfigured mask provided by the Centiloid scale method required for brain amyloid protein deposition detection, the image of the specific area is extracted from the registered positron tomography image, and the standardized uptake value ratio (SUVr) is determined based on the image of the specific area. The standardized uptake value ratio is converted to obtain the Centiloid value (brain β-amyloid protein deposition amount).

[0059] The brain tomography image processing method provided in the embodiment of the present application can be applied in the following application scenarios: 1. Drug research, based on the brain tomography image processing method provided in the embodiment of the present application, determine the binding of brain proteins and drugs, thereby accelerating drug research; 2. For patients who have been diagnosed with Alzheimer's disease, based on the brain tomography image processing method provided in the embodiment of the present application, the effects of different treatment methods, such as drug treatment and non-drug treatment, can be evaluated based on the measured Centiloid value. In the embodiment of the present application, the amount of amyloid protein deposition in the brain is not used as a reference for disease diagnosis.

[0060] In some embodiments, the server 200 may be 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, content delivery networks (CDN), and big data and artificial intelligence platforms. The terminal and the server may be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiments of the present application.

[0061] The embodiments of the present application can be implemented through database technology. In short, a database can be regarded as an electronic file cabinet where electronic files are stored. Users can add, query, update, delete, etc. data in the files. The so-called "database" is a collection of data that is stored together in a certain way, can be shared with multiple users, has as little redundancy as possible, and is independent of the application program.

[0062] A database management system (DBMS) is a computer software system designed for managing databases. It generally has basic functions such as storage, retrieval, security, and backup. Database management systems can be classified according to the database model they support, such as relational, XML (Extensible Markup Language); or according to the type of computer they support, such as server clusters, mobile phones; or according to the query language used, such as Structured Query Language (SQL), XQuery; or according to performance focus, such as maximum scale, maximum operating speed; or other classification methods. Regardless of the classification method used, some DBMS can cross categories, for example, supporting multiple query languages ​​at the same time.

[0063] See also Figure 2 , Figure 2 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may be Figure 1 Server 200, Figure 2 The server 200 shown includes: at least one processor 410, a memory 450, and at least one network interface 420. The various components in the server 200 are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 440 is not described in detail. Figure 2 Various buses are labeled as bus system 440 .

[0064] Processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0065] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.

[0066] The memory 450 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0067] In some embodiments, memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0068] Operating system 451, including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0069] A network communication module 452, used to reach other electronic devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 include: Bluetooth, Wireless Compatibility Certification (WiFi), and Universal Serial Bus (USB), etc.;

[0070] In some embodiments, the device provided in the embodiments of the present application can be implemented in software. Figure 2 A processing device 455 for brain tomography images stored in a memory 450 is shown, which may be software in the form of programs and plug-ins, including the following software modules: a position adjustment module 4551, a registration module 4552, and a recognition module 4553. These modules are logical, and therefore may be arbitrarily combined or further split according to the functions implemented. The functions of each module will be described below.

[0071] The method for processing brain tomography images provided in the embodiment of the present application will be explained in combination with the exemplary application and implementation of the terminal device or server provided in the embodiment of the present application.

[0072] The following describes the brain tomography image processing method provided by the embodiment of the present application. As mentioned above, the electronic device that implements the brain tomography image processing method of the embodiment of the present application can be a terminal device or a server, or a combination of the two. Therefore, the execution subject of each step will not be repeatedly described below.

[0073] It should be noted that the following image processing examples are described using the detection of amyloid protein deposition in the brain as an example. Those skilled in the art, based on their understanding of the following, can apply the brain tomography image processing method provided in the embodiments of the present application to the processing of positron tomography images of other types of proteins or other parts of the human body.

[0074] Referring to FIG. 3, FIG. 3 is a flow chart of a method for processing a brain tomography image provided in an embodiment of the present application, which will be described in conjunction with the steps shown in FIG. 3. The execution subject of the steps in FIG. 3 is Figure 1 Server 200 in.

[0075] In step 301, the position of a first brain tomography image is adjusted based on a brain reference direction to obtain a second brain tomography image after the position adjustment.

[0076] For example, the brain tomography image is a positron emission tomography (PET) image of the brain, which is obtained by using specific radioactive tracers to label beta-amyloid protein (Aβ) in the brain, hereinafter referred to as Aβ, which can bind to Aβ and then show the distribution of Aβ in the PET scan. The brain reference direction is a direction in the brain image used as a reference to distinguish different areas in the brain image. The brain reference position can be the anterior commissure position in anatomy, and the brain reference direction can be the anterior-posterior direction or the superior-inferior direction.

[0077] In some embodiments, before step 301, the following processing is performed: performing image enhancement processing on the initial brain tomography image to obtain a data enhanced image, wherein the image enhancement processing includes at least one of the following processing: intensity normalization, affine transformation, flipping, pixel erasing, and adding Gaussian noise. The data enhanced image is scaled to a first preconfigured size to obtain a first brain tomography image.

[0078] For example, the first preconfigured size can be set according to the application scenario, such as 64×64×64 voxels. By performing image enhancement and size standardization on the initial brain tomography image, the accuracy of subsequent processing can be improved and errors can be reduced. Flipping, pixel erasing and adding Gaussian noise can be performed randomly, and the types of affine transformations include: scaling, translation and rotation. Image intensity normalization is the process of adjusting the pixel intensity values ​​in the image to a standard range. In an image, the pixel intensity value usually represents the brightness or color information of the image.

[0079] In some embodiments, step 301 can be implemented in the following manner: calling a convolutional neural network model to perform the following processing on the first brain tomography image: performing feature extraction processing on the first brain tomography image to obtain a first image feature. Predicting a brain reference position in the first brain tomography image based on the first image feature, wherein the brain reference position includes at least an anterior commissure position. Determining a brain reference direction corresponding to the first brain tomography image based on the brain reference position. Adjusting the placement direction of the first brain tomography image according to the brain reference direction to obtain a second brain tomography image.

[0080] For example, the convolutional neural network model can be a backbone convolutional neural network including multiple residual networks, and the convolutional neural network extracts features from the first brain tomography image through a feature extraction layer to obtain a first image feature of 1024-dimensional features, and the prediction head of the convolutional neural network determines the brain reference position based on the first image feature. The anterior commissure is a structure in the brain, located near the midline between the cerebral hemispheres. It is an important landmark in neuroimaging and neuroanatomy, so the anterior commissure position is used as the brain reference position. Furthermore, the brain reference direction can be determined based on the anterior commissure position. For example: the anterior-posterior direction or the superior-inferior direction, wherein the anterior-posterior direction is the direction from the abdomen of the body to the back, and the superior-inferior direction is the direction from the head to the feet. According to the brain reference direction, the reference direction in the first brain tomography image is adjusted to be parallel to the direction in the space to be registered, and a second brain tomography image is obtained.

[0081] In some embodiments, before step 301, the convolutional neural network model is trained in the following manner: a first sample set is obtained, wherein the first sample set includes first sample images and an actual reference position and an actual reference direction corresponding to each first sample image. Based on each first sample image, the initialized convolutional neural network model is called to perform prediction processing to obtain a predicted reference position, and a predicted reference direction is determined based on the predicted reference position. Based on the difference between the predicted reference position and the actual reference position, a first loss function is determined. Based on the difference between the predicted reference direction and the actual reference direction, a second loss function is determined. The parameters of the initialized convolutional neural network model are updated based on the first loss function and the second loss function to obtain a trained convolutional neural network model.

[0082] For example, the first sample image is a brain tomography image that has been processed by image enhancement. The source of the sample image can be a data set that has been annotated by researchers, or a real brain tomography image collected with the knowledge of the user in an actual application scenario. The first loss function can be an absolute value loss function (Absolute Value Loss Function). The actual reference position and the predicted reference position can be represented as points in space. The calculation method of the absolute value loss function is the absolute value of the difference between the predicted value and the actual value. The distance between the two positions in space can be used as the value of the absolute value loss function; the second loss function can be a cosine loss function (Cosine Loss Function). The predicted reference direction and the actual reference direction can be represented as vectors. The cosine loss function is a loss function used to measure the cosine similarity between two vectors. The sum of the first loss function and the second loss function is used as the total loss function, and the parameters of the convolutional neural network are updated based on the total loss function. The parameter update method can be a back propagation process based on gradient descent.

[0083] In step 302, registration parameters are predicted based on the second brain tomography image and a standard template.

[0084] For example, a standard template refers to a template used in the spatial standardization process. The standard template is set in a standardized brain image space, such as the Montreal Neurological Institute space, referred to as the MNI space. The MNI space is intended to provide a unified coordinate system for brain images to facilitate comparison and analysis between different studies. The registration of brain tomography images is the process of mapping brain tomography images to a standardized space. During the process, it is necessary to determine the registration parameters corresponding to the brain tomography images. The registration parameters are used for the direction and distance of voxel movement in the brain tomography images.

[0085] In some embodiments, reference Figure 3B , Figure 3B is a second flow chart of the method for processing brain tomography images provided in the embodiment of the present application; step 302 can be performed by Figure 3B Steps 3021 to 3025 in are implemented as described in detail below.

[0086] In step 3021, the affine transformation model is called to perform the following processing on the second brain tomography image: feature extraction processing is performed on the second brain tomography image to obtain second image features.

[0087] For example, the affine transformation model includes an affine (Affine) model and a voxel transformation model (Voxel Morph), the affine model is used to execute steps 3021 to 3023, and the voxel transformation model (Voxel Morph) is used to execute steps 3024 and 3025.

[0088] For example, the affine model uses a five-layer convolutional neural regression network to learn from the input image and predict the predicted affine matrix used in the affine transformation operation. The feature extraction process can refer to the convolutional neural network model mentioned above.

[0089] In step 3022, prediction processing is performed based on the second image feature to obtain a prediction affine matrix.

[0090] For example, the number of channels of the affine model is 16, 32, 64, 128, 256, and the step size is set to 2, which is used to predict all 12 parameters of the affine matrix. Affine transformation is a method of transforming an image while keeping the image lines parallel and collinear. It includes transformations such as rotation, scaling, translation, flipping, and shearing. The affine model predicts the probability of occurrence of each of the above transformations and the corresponding specific transformation parameters when they occur based on the second image features, and combines each predicted affine transformation parameter into a predicted affine matrix.

[0091] In step 3023, an affine transformation is performed on the second brain tomography image based on the predicted affine matrix to obtain a fourth brain tomography image.

[0092] As an example, continuing with the above example, corresponding transformations are sequentially performed on the second brain tomography image based on the 12 parameters in the predicted affine matrix to obtain a transformed fourth brain tomography image.

[0093] In step 3024, a mapping relationship between each first voxel in the fourth brain tomography image and each second voxel in the standard template is determined.

[0094] For example, the standard template is a brain image in a standard space, and a mapping relationship between a fourth brain tomography image and the standard template is determined, that is, the position of voxels in the fourth brain tomography image in the standard space is determined based on the standard template as a reference. For example: based on the anterior commissure position in the standard template, the position in the standard space corresponding to the anterior commissure position in the fourth brain tomography image is determined.

[0095] In step 3025, prediction processing is performed based on the mapping relationship to obtain the registration vector corresponding to each first voxel, and the registration vector corresponding to each first voxel is combined into a displacement field, and the displacement field is used as a registration parameter.

[0096] For example, the registration vector represents the displacement required to map the first voxel to the standard space corresponding to the second voxel. The registration vector is used to indicate the distance of movement and the direction of movement. By combining the registration vectors corresponding to each first voxel, a registration displacement vector field can be obtained, that is, the registration parameters of the second brain tomography image.

[0097] In the embodiment of the present application, the registration parameters are predicted by the model, which can improve the registration efficiency compared with the software and manual registration in the related technology. Only brain tomography images are used without magnetic resonance imaging, which reduces the calculation complexity and saves computing resources.

[0098] In some embodiments, before step 302, the affine transformation model can be trained in the following manner: obtain a second sample set, wherein the second sample set includes second sample images and an actual affine matrix, actual registration parameters, and actual registration images corresponding to each second sample image. Based on each second sample image, call the initialized affine transformation model for prediction processing to obtain a predicted affine matrix and predicted registration parameters. Perform registration processing on the second sample image based on the predicted affine matrix and the predicted registration parameters to obtain a predicted registered image. Determine a third loss function based on the difference between the predicted affine matrix and the actual affine matrix, and the difference between the predicted registration parameters and the actual registration parameters. Determine a fourth loss function based on the deformation difference between the predicted registered image and the actual registered image. Update the parameters of the initialized affine transformation model based on the third loss function and the fourth loss function to obtain a trained affine transformation model.

[0099] For example, the second sample image is a brain tomography image that has been image enhanced. The source of the sample image can be a data set that has been annotated by researchers, or a real brain tomography image collected with the knowledge of the user in an actual application scenario. Based on the difference between the predicted affine matrix and the actual affine matrix, and the difference between the predicted registration parameters and the actual registration parameters, it can be reflected by the mean square error, which is the square of the difference between the predicted value and the true value of each sample, and then the average of these square differences is taken. The third loss function can be the mean square error loss function (MeanSquared Error Loss Function). The fourth loss function can be a bending loss, which is used to measure the difference between the predicted deformation and the actual deformation. The weighted sum of the third loss function and the fourth loss function is used as the total loss function. The specific weight value can be set according to the application scenario, and the parameters of the training affine transformation model are updated based on the total loss function. The parameter update method can be a back propagation process based on gradient descent.

[0100] Continue to refer Figure 3AIn step 303, the second brain tomography image is registered based on the registration parameters to obtain a third brain tomography image.

[0101] For example, the registration process is to move the second brain tomography image so as to map the second brain tomography image to the third brain tomography image in the standard space.

[0102] In some embodiments, the registration parameters include a plurality of displacement vectors, each of which corresponds to a first voxel in the second brain tomography image. Step 303 may be implemented in the following manner: according to each displacement vector, a displacement process is performed on the corresponding first voxel in the second brain tomography image to obtain a third voxel after displacement. Each third voxel is combined into a third brain tomography image.

[0103] For example, for the displacement of each voxel, the displacement field formed by the displacement vector and the image features of the second brain tomography image may be multiplied to obtain an image that is spatially aligned with the standard template.

[0104] In the embodiment of the present application, the registration task is automatically performed through a neural network model, thereby reducing human errors in the manual alignment process. Compared with related technologies, it does not rely on magnetic resonance imaging technology and can save the computing resources required to obtain the amount of amyloid protein deposition in the brain.

[0105] In step 304, the third brain tomography image is subjected to tracer identification to obtain a tracer uptake value corresponding to the brain region.

[0106] Here, tracer uptake values ​​are used to determine the amount of amyloid deposits in the brain.

[0107] For example, tracers for proteins are a special class of chemical substances or biological molecules used to track and detect the distribution, movement and function of proteins in organisms. Tracers can characterize the content of proteins in a region.

[0108] In some embodiments, step 304 can be implemented by: extracting a target brain region image in a third brain tomography image based on a preconfigured mask, performing tracer identification on the target brain region image, and obtaining an actual tracer uptake average value in the target brain region image, wherein the tracer is used to trace brain beta amyloid protein, and determining a tracer uptake value corresponding to the brain region based on the actual uptake average value and the reference uptake average value.

[0109] For example, the preconfigured mask is a preconfigured mask provided by the Centiloid scale method. The tracer uptake value is a standardized uptake value ratio (SUVr), which is represented by meanVOI / meanRef. Here, meanVOI represents the average uptake value of the tracer in the volume of interest (VOI) extracted from the third brain tomography image by the mask, and meanRef usually refers to the average uptake value of the radioactive tracer in a PET image in a reference region (Ref). The ratio between the actual uptake average and the reference uptake average.

[0110] In the embodiment of the present application, the position of the brain tomography image is adjusted so that the brain tomography image matches the reference direction, thereby reducing errors in the subsequent registration process. The registration parameters are predicted based on the brain tomography image and the standard template. Compared with the manual registration scheme in the related art, the accuracy and efficiency of the registration can be improved, and then the tracer uptake value is detected based on the registered brain tomography image, thereby improving the accuracy and efficiency of determining the amount of amyloid protein deposition in the brain.

[0111] Next, an exemplary application of the method for processing brain tomography images according to an embodiment of the present application in a practical application scenario will be described.

[0112] Quantitative assessment of brain amyloid β (Aβ) deposition is of great significance in the research and diagnosis of neurodegenerative diseases such as Alzheimer's disease. Centiloid scale, as a standardized quantitative index, has been widely used in the assessment of Aβ deposition. In the related art, the standard Centiloid calculation method uses the Statistical Parametric Mapping (SPM) software package, relying on magnetic resonance imaging images and complex spatial normalization processes. This process involves the processing of image change patterns such as rotation, translation and flipping, and usually requires manual intervention for coarse registration. The calculation process is complex and time-consuming, which limits its application in large-scale data analysis. In the related art, the image normalization method based on neural network has begun to be applied to the processing of AβPET images. However, due to the diverse image change patterns, manual coarse registration steps are still required, such as the need to use affine transformation to rotate and translate the image in advance. Therefore, there is an urgent need for a fully automatic, fast and accurate AβPET image Centiloid calculation method to improve the computational efficiency and the stability of quantitative results, and reduce the dependence on magnetic resonance imaging images.

[0113] The related technologies have the following problems: 1. The Centiloid calculation method relies on MRI images and a complex spatial standardization process, and requires manual rough registration. The manual rough registration process is time-consuming and easily affected by human factors, making the entire process complex and inefficient. 2. Each new registration process needs to be calculated from scratch, which takes a long time.

[0114] In order to solve the above-mentioned problems existing in the related art, the embodiments of the present application can handle changes such as rotation, translation and flipping in the image, so that the method for processing brain tomography images provided in the embodiments of the present application can be applied in the rapid, automated and large-scale cross-center assessment of amyloid protein deposition.

[0115] For ease of understanding, the following explanation is provided in conjunction with the accompanying drawings. Figure 4 , Figure 4 is a third flow chart of the method for processing brain tomography images provided in the embodiment of the present application. Figure 1 The server 200 is the execution entity.

[0116] In step 401, data enhancement is performed on a brain tomography image to be detected to obtain a first brain tomography image.

[0117] For example, data enhancement may include at least one of the following processes: intensity normalization, random affine transformation (including scaling, translation, and rotation), random flipping, random erasing, and random Gaussian noise.

[0118] Image intensity normalization is the process of adjusting the pixel intensity values ​​in an image to a standard range. In an image, the pixel intensity values ​​usually represent the brightness or color information of the image, and these values ​​may be distributed in different ranges due to various reasons (such as differences in equipment, changes in shooting conditions, etc.). Affine transformation refers to the linear transformation of an image in a two-dimensional space.

[0119] For example, by augmenting the data, the robustness and generalization ability of the model can be increased. After data augmentation, the input image is scaled to an input size of 64×64×64 voxels to facilitate subsequent processing.

[0120] In step 402, the position of the first brain tomography image is adjusted to obtain a second brain tomography image.

[0121] For example, step 402 can be implemented by a convolutional neural network (CNN) model of a rigid transform, and the convolutional neural network can be a backbone convolutional neural network composed of 5 layers of residual connection blocks. The number of channels is 16, 32, 64, 128, and 256, respectively, and the step size of each layer is set to 2 (each layer is downsampled to 1 / 2 of the original resolution). Each residual connection block includes a convolutional layer, a batch normalization layer, a Gaussian Error Linear Unit (Gaussian Error Linear Unit, GeLU) activation function, and a jump connection. The Gaussian Error Linear Unit activation function is a function used to introduce nonlinear characteristics in a neural network, which combines a linear unit (Linear Unit) and a Gaussian error function. The advantage of the GeLU activation function is that it can provide a smooth and nonlinear activation mode while maintaining model performance, which is particularly important when dealing with tasks with high complexity and nonlinearity.

[0122] For example, a convolutional neural network is used to estimate the position of the anterior commissure, the anterior-posterior (Anterior-Posterior), and the superior-inferior (Superior-Inferior) directions. The anterior commissure is a structure in the brain, located between the frontal lobe and the parietal lobe of the brain, connecting the front of the two cerebral hemispheres. The position of the anterior commissure is used as a reference point to ensure accurate alignment and correction of the image. The first brain tomography image is mapped into 1024-dimensional features through the convolutional neural network backbone, and connected to the fully connected layers respectively. The anterior commissure position and image direction are predicted in the voxel space. The prediction head of the convolutional neural network backbone uses the GeLU activation function as the activation function. With the anterior commissure position and image direction as reference, the position of the first brain tomography image is adjusted to obtain the second brain tomography image. Reference Fig. 6A , Fig. 6A The prediction result diagram of the anterior commissure position and reference direction provided by the embodiment of the present application; the embodiment of the present application performs preliminary prediction of the anterior commissure position and the brain direction prediction result of the input PET image in voxel space for subsequent processing. Among them, PiB (Pittsburgh Compound B), Florbetapir, Florbetaben, and Flutemetamol are different types of tracers.

[0123] In some embodiments, the convolutional neural network can be trained by the absolute value loss function and the cosine loss function. The absolute value loss function is also called the L1 loss function, which is mainly used to measure the difference between the predicted value and the actual value in regression problems. The absolute value loss function is calculated as the absolute value of the difference between the predicted value and the actual value. The cosine similarity loss function is a loss function used to measure the similarity between two vectors, and is often used in classification problems, such as: processing multi-class classification tasks.

[0124] In step 403, the affine transformation matrix and displacement field corresponding to the second brain tomography image are determined, and the second brain tomography image is registered based on the affine transformation matrix and the displacement field to obtain a spatially standardized third brain tomography image.

[0125] For example, after the brain tomography image is processed by the backbone convolutional neural network, an output image with a corrected direction is obtained, and the second brain tomography image is resampled according to the standardized brain image space (MNI space), and the resampling result is input into the affine voxel transformation model (Affine Voxel Morph).

[0126] Before the affine voxel transformation model processes the second brain tomography image, the second brain tomography image is subjected to intensity normalization and a slight random affine transformation to obtain an input image of the affine voxel transformation model. The above operation makes the size of the input image 96×128×96 to accommodate subsequent affine transformation and elastic registration operations.

[0127] The affine voxel transformation model includes an affine model and a voxel transformation model. The affine model uses a five-layer convolutional neural regression network with 16, 32, 64, 128, and 256 channels and a step size of 2, which is used to predict all 12 parameters of the affine matrix. The network structure includes multiple convolutional layers and fully connected layers, and the number of channels gradually increases from 16 to 256, which is used to learn and predict the transformation parameters used in the affine transformation operation from the input image. The input image is preliminarily transformed according to the affine matrix predicted by the affine model, and the displacement field is predicted by the voxel transformation model. The image after affine transformation is finely elastically registered according to the displacement field. The registration can be achieved in the following way: each voxel in the image after affine transformation is multiplied by each vector in the displacement field to obtain a third brain tomography image that is spatially aligned with the standard template. Reference Figure 6B , Figure 6BIt is a schematic diagram of affine transformation and elastic transformation provided in an embodiment of the present application; it is a schematic diagram of affine transformation and elastic transformation of an input PET image in an embodiment of the present application, and the arrow is the transformation direction of the pixel point.

[0128] In some embodiments, the mean square error loss function (MSE Loss Function) is used in the training process of the voxel transformation model to optimize the accuracy of the affine transformation and the elastic displacement field, supplemented by the bending loss to ensure the smoothness of the displacement field. The weight of the deformation loss is set to 0.1. The optimization target is set to the spatial normalized output of the standard SPM software package to align with the Centiloid scale. The mean square error loss function is a loss function commonly used for regression problems. In machine learning and deep learning, the loss function is used to measure the difference between the model prediction value and the true value. MSE calculates the square of the difference between the predicted value and the true value of each sample, and then takes the average of these square differences. When processing images or geometric shapes, deformation loss is used to measure the difference between predicted deformation and actual deformation.

[0129] In step 404, the tracer is identified based on the third brain tomography image, and the standardized uptake value ratio corresponding to the tracer is determined, and the standardized uptake value ratio is converted into the amount of brain amyloid beta protein deposition.

[0130] For example, the third brain tomography image is a registered PET image. The third brain tomography image is processed with the preconfigured mask provided by the Centiloid scale method to obtain a PET image of a specific region in the brain. The Standardized Uptake Value ratio (SUVr) is calculated based on the image. The Standardized Uptake Value ratio is represented by meanVOI / meanRef. The Standardized Uptake Value ratio is an indicator used to quantify the uptake of radioactive tracers in a specific region in a PET image. In a PET scan, a patient is injected with a radioactive tracer that accumulates in specific metabolically active regions in the body, such as tumor tissue. The Centiloid algorithm provides a standardized mask (also called a template or a region of interest (VOI)) that defines a specific region in the brain for analyzing the distribution of amyloid protein. meanVOI refers to the average uptake value of the radioactive tracer in the PET image within the region of interest (VOI) defined by the Centiloid mask. meanRef generally refers to the average uptake value of the radioactive tracer in a PET image within a reference region (Ref). This reference area is usually an area of ​​the brain that is not affected by the disease and is used as a normal level of uptake. Depending on the PET tracer used (such as PiB, AV45, etc.), the SUVr value is converted using the corresponding Centiloid calculation formula to obtain the final Centiloid value (the amount of beta-amyloid protein deposits in the brain).

[0131] refer to Figure 5 , Figure 5 It is a schematic diagram of the principle of image processing provided in an embodiment of the present application; the fully automatic Centiloid calculation system based on a cascade neural network provided in an embodiment of the present application is compared with the Centiloid calculation process based on statistical parameter mapping (Statistical Parametric Mapping, SPM), hereinafter referred to as SPM12. The image registration result of SPM12 can be used to train a neural network. The Centiloid calculation system based on a cascade neural network provided in an embodiment of the present application only needs to input a PET image during reasoning. The PET image is processed by the following models in sequence to obtain a standard spatialized image: an affine voxel transformation model composed of a rigid deformation model 501 (that is, a backbone convolutional neural network for rigid deformation), an affine model 502, and a voxel transformation model 503. The Centiloid value of amyloid protein is calculated based on the standard spatialized image.

[0132] Compared with the SPM12 software in the related art, the method for processing brain tomography images provided in the embodiment of the present application does not need to rely on magnetic resonance imaging technology, can improve the registration efficiency and save computing resources.

[0133] In some embodiments, the backbone convolutional neural network for rigid transformation and the affine voxel transformation model constitute a cascade neural network model, which can be trained using PyTorch and MONAI frameworks (medical open source AI frameworks). By iteratively optimizing the neural network parameters, it can accurately learn and predict the spatial normalization and Centiloid calculation tasks of PET images. The trained model is exported to the Open Neural Network Exchange (ONNX) format, and reasoning is performed using the ONNX runtime and Insight Segmentation and Registration Toolkit (ITK) in C++ or PyTorch and MONAI based on the Python platform to ensure efficient operation on different computing platforms. The Insight Segmentation and Registration Toolkit is an open source, cross-platform medical image processing library. It provides a series of building blocks for medical image analysis, which can be used to implement image segmentation, registration, transformation, image processing algorithms, etc.

[0134] The beneficial effects of the brain tomography image processing method provided in the embodiment of the present application are explained below in combination with experimental results.

[0135] refer to Fig. 7A , Fig. 7A Schematic diagram of linear correlation provided by the embodiment of the present application; the embodiment of the present application is compared with the Centiloid calculated value of the related art, which has good consistency and meets the linear correlation (R 2 >0.99)

[0136] refer to Figure 7B , Figure 7B is a first schematic diagram for comparing experimental results provided in the embodiments of the present application; Figure 7B The SUVr error obtained in the embodiment of the present application is compared with the error generated by using SPM12, and compared with the calculation results of SPM12, indicating that the embodiment of the present application has good generalization performance on a variety of imaging agents. Among them, Florbetaben and Flutemetamol, two imaging agents, do not appear in the neural network training process.

[0137] refer to Figure 7C , Figure 7C is a second schematic diagram for comparing experimental results provided in the embodiments of the present application; Figure 7C The relative error between the embodiment of the present application and the SPM12 calculation method is characterized.

[0138] refer to Fig.7D , Fig.7D is a third schematic diagram for comparing experimental results provided in the embodiments of the present application; Fig.7D Characterization, in the embodiment of the present application, when the graphics processing unit is enabled, compared with the time-consuming SPM12 calculation method of the related technology, while there is no need for researchers to perform rough alignment, the calculation speed is increased by 80 to 100 times, which can greatly improve the Centiloid calculation efficiency.

[0139] The embodiment of the present application does not rely on magnetic resonance imaging images, and can handle changes such as rotation, translation, and flipping of images without manual intervention, ensuring the accurate alignment of the input PET image to the MNI space. A cascade neural network method that can quickly and accurately complete coarse and fine registration can greatly reduce computing time and resource consumption, and achieve efficient automatic Centiloid calculation. After verification by the above experiments, it has good generalization performance for the four common Aβ imaging agents in clinical practice. The Centiloid calculation process is greatly simplified, and it is easy to use in conjunction with various medical image processing systems.

[0140] In actual application scenarios, the system integrated with the processing method of brain tomography images provided in the embodiment of the present application can be easily connected to general medical imaging software such as 3DSlicer, and can also rely on existing hardware and software systems to perform calculations. The system configuration includes high-performance computing and storage devices of the graphics card, and the graphics processing unit (GPU) is used to accelerate the Centiloid calculation process to improve computing efficiency and processing speed. Develop a C++ green program to implement Centiloid calculations, so that it is easy to convert into medical imaging software plug-ins such as 3D slicers (3Dslicer), so that ordinary users can easily use the method designed in the embodiment of the present application for PET image processing and Centiloid calculations. Or use the hardware and software system provided in the embodiment of the present application for Centiloid calculations.

[0141] refer to Figure 8 , Figure 8801 is a schematic diagram of the interface of the application provided in the embodiment of the present application. The application interface 801 is the interface of the 3D Slicer. The 3D Slicer medical image processing software is used to open the file in the medical image file format to be analyzed (for example, DICOM or Nifti file). The brain tomography image processing method provided in the embodiment of the present application is run through the plug-in. After the execution is completed, the calculated Centiloid result is displayed. The brain tomography image processing method provided in the embodiment of the present application can be simply combined with other medical image processing systems, such as Figure 8 As shown, it is used in conjunction with the medical image processing software 3DSlicer to provide nuclear medicine researchers with one-click Centiloid calculation results.

[0142] The following is a description of an exemplary structure of a brain tomography image processing device 455 provided in an embodiment of the present application implemented as a software module. In some embodiments, Figure 2 As shown, the software modules in the brain tomography image processing device 455 stored in the memory 450 may include: a position adjustment module 4551, used to adjust the position of the first brain tomography image based on the brain reference direction to obtain a second brain tomography image after position adjustment; a registration module 4552, used to predict the registration parameters based on the second brain tomography image and the standard template; the registration module 4552 is also used to perform registration processing on the second brain tomography image based on the registration parameters to obtain a third brain tomography image; an identification module 4553, used to perform tracer identification on the third brain tomography image to obtain a tracer uptake value corresponding to the brain area, wherein the tracer uptake value is used to determine the amount of amyloid protein deposition in the brain.

[0143] In some embodiments, the position adjustment module 4551 is used to call the convolutional neural network model to perform the following processing on the first brain tomography image: perform feature extraction processing on the first brain tomography image to obtain a first image feature; predict a brain reference position in the first brain tomography image based on the first image feature, wherein the brain reference position includes at least an anterior commissure position; determine a brain reference direction corresponding to the first brain tomography image based on the brain reference position; and adjust the placement direction of the first brain tomography image according to the brain reference direction to obtain the second brain tomography image.

[0144] In some embodiments, the position adjustment module 4551 is used to obtain a first sample set before the position adjustment of the first brain tomography image based on the brain reference direction to obtain the second brain tomography image after the position adjustment, wherein the first sample set includes a first sample image and an actual reference position and an actual reference direction corresponding to each of the first sample images; based on each of the first sample images, calling the initialized convolutional neural network model for prediction processing to obtain a predicted reference position, and determining a predicted reference direction based on the predicted reference position; based on the difference between the predicted reference position and the actual reference position, determining a first loss function; based on the difference between the predicted reference direction and the actual reference direction, determining a second loss function; based on the first loss function and the second loss function, updating the parameters of the initialized convolutional neural network model to obtain the trained convolutional neural network model.

[0145] In some embodiments, the registration module 4552 is used to call the affine transformation model to perform the following processing on the second brain tomography image: perform feature extraction processing on the second brain tomography image to obtain second image features; perform prediction processing based on the second image features to obtain a predicted affine matrix; perform affine transformation on the second brain tomography image based on the predicted affine matrix to obtain a fourth brain tomography image; determine the mapping relationship between each first voxel in the fourth brain tomography image and each second voxel in the standard template; perform prediction processing based on the mapping relationship to obtain a registration vector corresponding to each first voxel, wherein the registration vector represents the displacement required to map the first voxel to the standard space corresponding to the second voxel; combine the registration vectors corresponding to each first voxel into a displacement field, and use the displacement field as the registration parameter.

[0146] In some embodiments, the registration module 4552 is used to obtain a second sample set before predicting the registration parameters based on the second brain tomography image and the standard template, wherein the second sample set includes second sample images and the actual affine matrix, actual registration parameters and actual registration images corresponding to each of the second sample images; based on each of the second sample images, call the initialized affine transformation model for prediction processing to obtain a predicted affine matrix and predicted registration parameters; based on the predicted affine matrix and the predicted registration parameters, perform registration processing on the second sample images to obtain a predicted registered image; based on the difference between the predicted affine matrix and the actual affine matrix, and the difference between the predicted registration parameters and the actual registration parameters, determine a third loss function; based on the deformation difference between the predicted registration image and the actual registration image, determine a fourth loss function; based on the third loss function and the fourth loss function, update the parameters of the initialized affine transformation model to obtain the trained affine transformation model.

[0147] In some embodiments, the registration parameters include multiple displacement vectors, each of which corresponds one-to-one to a first voxel in the second brain tomography image; the registration module 4552 is used to perform displacement processing on the corresponding first voxel in the second brain tomography image according to each displacement vector to obtain a displaced third voxel; each of the third voxels is combined into the third brain tomography image.

[0148] In some embodiments, the identification module 4553 is used to extract the target brain area image in the third brain tomography image based on a preconfigured mask; perform tracer identification on the target brain area image to obtain the actual tracer uptake average value in the target brain area image, wherein the tracer is used to trace brain beta-amyloid protein; and determine the tracer uptake value corresponding to the brain area based on the actual uptake average value and the reference uptake average value.

[0149] In some embodiments, the position adjustment module 4551 is used to perform image enhancement processing on the initial brain tomography image to obtain a data enhanced image before adjusting the position of the first brain tomography image based on the brain reference direction to obtain the second brain tomography image after the position adjustment, wherein the image enhancement processing includes at least one of the following processing: intensity normalization, affine transformation, flipping, pixel erasing and adding Gaussian noise; scaling the data enhanced image to a first preconfigured size to obtain the first brain tomography image.

[0150] The embodiment of the present application provides a computer program product, which includes a computer program or a computer executable instruction, and the computer program or the computer executable instruction is stored in a computer-readable storage medium. The processor of the electronic device reads the computer program or the computer executable instruction from the computer-readable storage medium, and the processor executes the computer program or the computer executable instruction, so that the electronic device executes the brain tomography image processing method described in the embodiment of the present application.

[0151] The present application embodiment provides a computer-readable storage medium storing computer executable instructions, wherein the computer executable instructions or computer programs are stored. When the computer executable instructions or computer programs are executed by a processor, the processor will be caused to execute the brain tomography image processing method provided by the present application embodiment, for example, Figure 3A A method for processing brain tomography images is shown.

[0152] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or it may be various devices including one or any combination of the above memories.

[0153] In some embodiments, computer executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0154] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file storing other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0155] As an example, computer executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed at multiple sites and interconnected by a communication network.

[0156] In summary, the embodiment of the present application adjusts the position of the brain tomography image so that the brain tomography image matches the reference direction, reduces errors in the subsequent registration process, and predicts the registration parameters based on the brain tomography image and the standard template. Compared with the manual registration scheme in the related art, it can improve the accuracy and efficiency of the registration, and then detect the tracer uptake value based on the registered brain tomography image, thereby improving the accuracy and efficiency of determining the amount of amyloid protein deposition in the brain.

[0157] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A method for processing a brain tomography image, characterized in that: The method comprises: Adjusting the position of the first brain tomography image based on the brain reference direction to obtain a second brain tomography image after the position adjustment; predicting registration parameters based on the second brain tomography image and a standard template; performing registration processing on the second brain tomography image based on the registration parameters to obtain a third brain tomography image; The third brain tomography image is subjected to tracer identification to obtain a tracer uptake value corresponding to the brain region, wherein the tracer uptake value is used to determine the amount of amyloid protein deposition in the brain.

2. The method according to claim 1, characterized in that The method of adjusting the position of the first brain tomography image based on the brain reference direction to obtain the second brain tomography image after the position adjustment includes: The convolutional neural network model is called to perform the following processing on the first brain tomography image: Performing feature extraction processing on the first brain tomography image to obtain a first image feature; Predicting a brain reference position in the first brain tomography image based on the first image feature, wherein the brain reference position at least includes an anterior commissure position; determining a brain reference direction corresponding to the first brain tomography image based on the brain reference position; The placement direction of the first brain tomography image is adjusted according to the brain reference direction to obtain the second brain tomography image.

3. The method according to claim 2, characterized in that Before adjusting the position of the first brain tomography image based on the brain reference direction to obtain the second brain tomography image after the position adjustment, the method further includes: Acquire a first sample set, wherein the first sample set includes first sample images and an actual reference position and an actual reference direction corresponding to each of the first sample images; Based on each of the first sample images, the initialized convolutional neural network model is called to perform prediction processing to obtain a prediction reference position, and a prediction reference direction is determined based on the prediction reference position; determining a first loss function based on a difference between the predicted reference position and the actual reference position; determining a second loss function based on a difference between the predicted reference direction and the actual reference direction; Based on the first loss function and the second loss function, the parameters of the initialized convolutional neural network model are updated to obtain the trained convolutional neural network model.

4. The method according to claim 1, characterized in that The predicting of registration parameters based on the second brain tomography image and the standard template includes: The affine transformation model is called to perform the following processing on the second brain tomography image: performing feature extraction processing on the second brain tomography image to obtain second image features; Performing prediction processing based on the second image feature to obtain a prediction affine matrix; Performing an affine transformation on the second brain tomography image based on the predicted affine matrix to obtain a fourth brain tomography image; Determining a mapping relationship between each first voxel in the fourth brain tomography image and each second voxel in the standard template; Performing prediction processing based on the mapping relationship to obtain a registration vector corresponding to each of the first voxels, wherein the registration vector represents a displacement required to map the first voxel to a standard space corresponding to the second voxel; The registration vectors corresponding to each first voxel are combined into a displacement field, and the displacement field is used as the registration parameter.

5. The method according to claim 4, characterized in that Before predicting the registration parameters based on the second brain tomography image and the standard template, the method further includes: Acquire a second sample set, wherein the second sample set includes second sample images and an actual affine matrix, actual registration parameters, and actual registered images corresponding to each of the second sample images; Based on each of the second sample images, the initialized affine transformation model is called to perform prediction processing to obtain a prediction affine matrix and a prediction registration parameter; Performing registration processing on the second sample image based on the predicted affine matrix and the predicted registration parameters to obtain a predicted registered image; Determining a third loss function based on a difference between the predicted affine matrix and the actual affine matrix, and a difference between the predicted registration parameters and the actual registration parameters; Determining a fourth loss function based on a deformation difference between the predicted registered image and the actual registered image; The parameters of the initialized affine transformation model are updated based on the third loss function and the fourth loss function to obtain the trained affine transformation model.

6. The method according to claim 1, characterized in that The registration parameters include a plurality of displacement vectors, each of which corresponds one-to-one to a first voxel in the second brain tomography image; The step of performing registration processing on the second brain tomography image based on the registration parameters to obtain a third brain tomography image comprises: According to each of the displacement vectors, a displacement process is performed on the corresponding first voxel in the second brain tomography image to obtain a third voxel after displacement; Each of the third voxels is combined into the third brain tomography image.

7. The method according to claim 1, characterized in that The step of performing tracer identification on the third brain tomography image to obtain a tracer uptake value corresponding to a brain region includes: extracting a target brain region image in the third brain tomography image based on a preconfigured mask; Performing tracer identification on the target brain region image to obtain an actual average value of tracer uptake in the target brain region image, wherein the tracer is used to trace brain amyloid beta protein; Based on the actual uptake average value and the reference uptake average value, a tracer uptake value corresponding to the brain region is determined.

8. The method according to any one of claims 1 to 7, characterized in that: Before adjusting the position of the first brain tomography image based on the brain reference direction to obtain the second brain tomography image after the position adjustment, the method further includes: Performing image enhancement processing on the initial brain tomography image to obtain a data enhanced image, wherein the image enhancement processing includes at least one of the following processing: intensity normalization, affine transformation, flipping, pixel erasing, and adding Gaussian noise; The data augmented image is scaled to a first preconfigured size to obtain the first brain tomography image.

9. A brain tomography image processing device, characterized in that: The device comprises: A position adjustment module, used for adjusting the position of the first brain tomography image based on the brain reference direction to obtain a second brain tomography image after position adjustment; A registration module, used for predicting registration parameters based on the second brain tomography image and a standard template; The registration module is further used to perform registration processing on the second brain tomography image based on the registration parameters to obtain a third brain tomography image; The identification module is used to perform tracer identification on the third brain tomography image to obtain a tracer uptake value corresponding to the brain region, wherein the tracer uptake value is used to determine the amount of amyloid protein deposition in the brain.

10. An electronic device, characterized in that: The electronic device comprises: A memory for storing computer executable instructions or computer programs; A processor, configured to implement the method for processing a brain tomography image according to any one of claims 1 to 8 when executing the computer executable instructions or computer program stored in the memory.

Images