CN118365632B - Image processing method and device and computer-aided diagnosis method of brain diseases - Google Patents

Abstract

The embodiments of this specification provide an image processing method and device, and a computer-aided diagnosis method for brain diseases, wherein the image processing method includes: receiving an image processing task, wherein the image processing task carries a three-dimensional brain image; inputting the three-dimensional brain image into an anatomical segmentation model, obtaining gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on the contrast information of the three-dimensional brain image; generating a target cerebral cortex according to the gray matter information, and determining the curvature characteristic information of the vertices on the target cerebral cortex; determining the cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and the curvature characteristic information of the vertices on the target cerebral cortex; determining the whole brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, white matter information, and subcortical structure information. This method provides a two-step segmentation method from coarse-grained to fine-grained, which speeds up data processing and improves the accuracy of image processing.

Description

Image processing method and device, computer-aided diagnosis method for brain diseases

Technical Field

The embodiments of the present specification relate to the field of computer technology, and in particular to an image processing method.

Background Art

In the analysis of brain magnetic resonance imaging (MRI) in the medical field, the segmentation of whole brain structure is a basic and critical step. It assigns a semantic label corresponding to a specific neuroanatomical structure to each voxel, so that quantitative attribute values such as volume, thickness, and area of various regions of interest can be obtained. These attribute values are crucial for many clinical applications and scientific research.

The current whole-brain segmentation algorithms are generally divided into two categories. The first category is based on brain maps, and the second category is to develop an end-to-end segmentation network using deep learning technology. The first method is difficult to align different areas when subdividing gray matter into smaller areas, and the algorithm is more complex and time-consuming. The second method has a faster processing speed, but its accuracy is poor. Therefore, a new whole-brain segmentation method is urgently needed to solve the above problems.

Summary of the invention

In view of this, an embodiment of this specification provides an image processing method. One or more embodiments of this specification also relate to a computer-aided diagnosis method for brain diseases, an image processing device, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

According to a first aspect of an embodiment of this specification, there is provided an image processing method, including:

receiving an image processing task, wherein the image processing task carries a three-dimensional brain image;

Inputting the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image;

Generate a target cerebral cortex according to the gray matter information, and determine curvature characteristic information of at least one vertex on the target cerebral cortex;

Determine a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and curvature feature information of at least one vertex on the target cerebral cortex;

A whole-brain segmentation result corresponding to the three-dimensional brain image is determined according to the cortical surface segmentation result, the white matter information and the subcortical structure information.

According to a second aspect of an embodiment of this specification, there is provided an image processing method, which is applied to a cloud-side device, and includes:

An image processing task sent by a receiving end, wherein the image processing task carries a three-dimensional brain image;

Inputting the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image;

Generate a target cerebral cortex according to the gray matter information, and determine curvature characteristic information of at least one vertex on the target cerebral cortex;

Determine a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and curvature feature information of at least one vertex on the target cerebral cortex;

Determining a whole-brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information, and the subcortical structure information;

The whole-brain segmentation result is sent to the client-side device.

According to a third aspect of the embodiments of this specification, a computer-aided diagnosis method for brain diseases is provided, comprising:

receiving a brain disease detection task, wherein the brain disease detection task carries a three-dimensional brain image and a target brain structure identifier;

Inputting the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image;

Generate a target cerebral cortex according to the gray matter information, and determine curvature characteristic information of at least one vertex on the target cerebral cortex;

Determine a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and curvature feature information of at least one vertex on the target cerebral cortex;

Determining a whole-brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information, and the subcortical structure information;

The target brain structure is determined in the whole-brain segmentation result according to the target brain structure identifier, and the target brain structure is detected to obtain the segmentation and detection results corresponding to the target brain structure.

According to a fourth aspect of the embodiments of this specification, there is provided an image processing device, including:

A receiving module, configured to receive an image processing task, wherein the image processing task carries a three-dimensional brain image;

A first segmentation module is configured to input the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image;

A generating module, configured to generate a target cerebral cortex according to the gray matter information, and determine curvature characteristic information of at least one vertex on the target cerebral cortex;

A second segmentation module is configured to determine a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and curvature feature information of at least one vertex on the target cerebral cortex;

The determination module is configured to determine the whole brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information and the subcortical structure information.

According to a fifth aspect of an embodiment of this specification, a computing device is provided, including:

Memory and processor;

The memory is used to store computer programs/instructions, and the processor is used to execute the computer programs/instructions. When the computer programs/instructions are executed by the processor, the steps of the above method are implemented.

According to a sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores a computer program/instruction, and the steps of the above method are implemented when the computer program/instruction is executed by a processor.

According to a seventh aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program/instruction, which implements the steps of the above method when executed by a processor.

An embodiment of the present specification provides an image processing method, comprising: receiving an image processing task, wherein the image processing task carries a three-dimensional brain image; inputting the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image; generating a target cerebral cortex according to the gray matter information, and determining curvature feature information of at least one vertex on the target cerebral cortex; determining a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and the curvature feature information of at least one vertex on the target cerebral cortex; and determining a whole-brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information, and the subcortical structure information.

The image processing method provided by the embodiment of the present application adopts a two-step segmentation method from coarse-grained to fine-grained, and the three-dimensional brain image is input into the anatomical segmentation model. The anatomical segmentation model only segments the gray matter information, white matter information and subcortical structure information based on the contrast information. Further, the cerebral cortex is generated based on the gray matter information, and the curvature characteristics of each vertex on the cerebral cortex are calculated. The curvature characteristics represent the structural information of the cerebral cortex. The gray matter information representing the curvature characteristics of the structural information is used to determine the cortical surface segmentation result corresponding to the cerebral cortex, which solves the problem of low brain segmentation accuracy. Finally, based on the white matter information, subcortical structure information and cortical surface segmentation results, the whole brain segmentation result corresponding to the three-dimensional brain image is determined, which not only saves model processing time, speeds up data processing speed, but also improves the accuracy of image processing.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of an image processing method provided by an embodiment of the present specification;

FIG2 is a schematic diagram of a processing structure of an image processing method provided by an embodiment of this specification;

FIG3 is an architecture diagram of an image processing system provided by an embodiment of the present specification;

FIG4 is a schematic diagram of a flow chart of an image processing method applied to a cloud-side device provided by an embodiment of the present specification;

FIG5 is a flow chart of a computer-aided diagnosis method for brain diseases provided in one embodiment of the present specification;

FIG6 is a schematic diagram of the structure of an image processing device provided by an embodiment of this specification;

FIG. 7 is a structural block diagram of a computing device provided by an embodiment of the present specification.

DETAILED DESCRIPTION

Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and/or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this manual are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

First, the terms involved in one or more embodiments of this specification are explained.

MRI: Magnetic resonance imaging, magnetic resonance imaging technology or often refers to magnetic resonance images collected by magnetic resonance imaging technology.

WM: White matter, refers to the area in the central nervous system that is mainly composed of nerve axons covered by myelin (which can form nerve bundles), which usually appears white in structural MRI.

GM: Graymatter, also known as the cortex, is an important part of the central nervous system and usually appears gray in structural MRI.

CorticalParcellation: Cortical parcellation, which is the division of the cortex into different regions, is a prerequisite step in brain imaging analysis.

Whole-brain segmentation: Whole-brain segmentation is an important neuroimaging task, which is to subdivide brain MRI images into different regions, including WM, GM, cerebellum, brainstem, and more detailed cortical partitions, etc. It is a prerequisite for brain image analysis.

In the analysis of brain magnetic resonance imaging (MRI) in the medical field, the segmentation of whole brain structure is a basic and critical step. Whole brain segmentation of brain structure MRI images divides the entire brain into multiple anatomically related regions of interest, which is a prerequisite for the subsequent quantitative analysis of different brain structures. It assigns a semantic label corresponding to a specific neuroanatomical structure to each voxel, so that quantitative attribute values such as volume, thickness, and area of various regions of interest can be obtained. These attribute values can be used for brain morphology analysis, surgical planning, and postoperative evaluation.

Traditional whole-brain segmentation algorithms usually include two categories:

The first category is the brain atlas-based method, which uses deformable registration technology to migrate the manually segmented brain atlas labels to each image, such as FreeSurfer and BrainSuite. Although this method is relatively accurate in segmenting high-contrast tissue types (such as white matter, gray matter, and cerebrospinal fluid) and subcortical structures (such as ventricles, cerebellum, and amygdala), it generally performs poorly when subdividing gray matter (ie, cortex) into smaller areas. This is because the cerebral cortex is a highly wrinkled and thin structure, and the tissue density it contains is similar. Traditional grayscale-based registration algorithms have difficulty aligning different regions. Moreover, the complexity and time-consuming nature of such algorithms further hinder their routine use in clinical applications.

The second type of whole-brain segmentation methods use deep learning technology to develop end-to-end segmentation networks. This type of method directly takes the whole brain image as input and outputs the label classification results for each voxel. These networks successfully solve the speed problem of the first type of algorithms and meet the requirements of clinical real-time applications. However, these end-to-end deep learning networks only rely on the grayscale features of the input image, ignoring the complex geometric details of the brain's anatomical structure, and therefore often sacrifice accuracy.

Based on this, in this specification, an image processing method is provided. This specification also involves a computer-aided diagnosis method for brain diseases, an image processing device, a computing device, a computer-readable storage medium and a computer program product, which are described in detail one by one in the following embodiments.

Referring to FIG. 1 , FIG. 1 shows a flow chart of an image processing method provided according to an embodiment of the present specification, which specifically includes the following steps.

Step 102: receiving an image processing task, wherein the image processing task carries a three-dimensional brain image.

In practical applications, the image processing method provided in the embodiments of this specification can be applied to a server or a client. In this implementation, the execution subject of the image processing method is not limited.

Specifically, the image processing task can be understood as a whole-brain segmentation task for three-dimensional brain images, and its purpose is to segment the three-dimensional brain images. The image processing task carries three-dimensional brain images, and the three-dimensional brain images can be understood as three-dimensional images generated by brain magnetic resonance imaging technology.

In practical applications, a person takes a three-dimensional brain image of the brain through a nuclear magnetic resonance instrument, and needs to perform a whole brain segmentation task on the three-dimensional brain image. An image processing task is triggered based on the three-dimensional brain image, and the execution terminal receives the image processing task, which carries the three-dimensional brain image.

Furthermore, when the method provided in this specification is applied to a client, the user can trigger an image processing task based on a three-dimensional brain image in the client, and the client can receive the image processing task. When the method provided in this specification is applied to a server, the user can generate an image processing task based on a three-dimensional brain image in the client, and send the image processing task to the server, and the server can receive the image processing task.

Step 104: input the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image.

Among them, the anatomical segmentation model can be understood as a pre-trained machine learning model, which is trained to distinguish different anatomical structures in three-dimensional brain images based on MRI grayscale values and contrast.

In practical applications, magnetic resonance imaging is a medical diagnostic technology that uses the principle of nuclear magnetic resonance to perform tomographic imaging of human tissue. In MRI images, grayscale values are represented according to the signal intensity of different tissues. The examination results of magnetic resonance imaging usually have two signals, namely T1-weighted image signals and T2-weighted image signals. In typical MRI images, since white matter, gray matter and subcortical structures have relatively high contrast, rough segmentation of gray matter information, white matter information and subcortical structure information can be achieved based on contrast information.

In this embodiment, the subcortical structure information includes masks of various subcortical brain tissue structures, and the subcortical structures include ventricles, cerebellum, thalamus, caudate nucleus, putamen, and the like.

In practical applications, the 3D brain image is input into the anatomical segmentation model for preliminary segmentation, and the 3D brain image is roughly segmented into white matter, gray matter and subcortical structures (such as ventricles, cerebellum, thalamus, caudate nucleus, putamen, etc.). In the anatomical segmentation model, only the grayscale value and contrast features are emphasized, without considering the geometric structure features in the brain structure, which can speed up the data processing speed of the anatomical segmentation model to process 3D brain images, thereby meeting the requirements of clinical implementation and application.

In a specific embodiment provided in this specification, the anatomical segmentation model is obtained by training through the following steps:

Acquire a sample three-dimensional brain image and sample gray matter information, sample white matter information, and sample subcortical structure information corresponding to the sample three-dimensional brain image;

Inputting the sample three-dimensional brain image into an anatomical segmentation model to obtain predicted gray matter information, predicted white matter information, and predicted subcortical structure information output by the anatomical segmentation model based on contrast information in the sample three-dimensional brain image;

Calculating a gray matter loss value according to the sample gray matter information and the predicted gray matter information, calculating a white matter loss value according to the sample white matter information and the predicted white matter information, and calculating a subcortical structure loss value according to the sample subcortical structure information and the predicted subcortical structure information;

The model parameters of the anatomical segmentation model are adjusted according to the gray matter loss value, the white matter loss value and the subcortical structure loss value, and the anatomical segmentation model is continuously trained until a model training stop condition is reached.

In the method provided in the embodiment of the present specification, the anatomical segmentation model is trained in a supervised manner, which includes a plurality of training sample pairs, wherein the training sample pairs include a sample three-dimensional brain image and a sample label corresponding to the sample three-dimensional brain image. Furthermore, the sample label specifically includes sample gray matter information, sample white matter information, and sample subcortical structure.

In practical applications, the sample 3D brain image and the sample 3D brain image can be manually annotated. After authorization, the sample 3D brain image can be obtained, and the technicians can annotate the sample 3D brain image, or the sample 3D brain image can be input into the sample annotation model for processing, and the sample annotation information output by the sample annotation model can be obtained. The sample 3D brain image and the sample annotation information form a training sample pair for subsequent model training of the anatomical segmentation model.

Specifically, the cortical region of interest labels can be merged into the sample gray matter information. Furthermore, the sample gray matter information can be further divided into the sample left gray matter information and the sample right gray matter information according to the positional relationship in the three-dimensional brain image. In practical applications, it can be determined whether the sample gray matter information needs to be split into the sample left gray matter information and the sample right gray matter information according to the actual situation. This is not limited in the present embodiment.

The sample three-dimensional brain image is input into the anatomical segmentation model for processing. The anatomical segmentation model performs image recognition on the sample three-dimensional brain image based on contrast information and grayscale values. Due to the high contrast of gray matter, white matter and subcortical structures in MRI images, the sample three-dimensional brain image can be divided into predicted gray matter information, predicted white matter information and predicted subcortical structure information.

The anatomical segmentation model at this time is a machine learning model that has not been trained yet. It is necessary to compare the predicted results with the sample labels in the sample pair to determine the gap between the predicted results of the anatomical segmentation model and the actual results. Specifically, the predicted results include predicted gray matter information, predicted white matter information, and predicted subcortical structure information; the sample labels include sample gray matter information, sample white matter information, and sample subcortical structure information. The gray matter loss value can be calculated based on the sample gray matter information and the predicted gray matter information, the white matter loss value can be calculated based on the sample white matter information and the predicted white matter information, and the subcortical structure loss value can be calculated based on the sample subcortical structure information and the predicted subcortical structure information.

There are many methods for calculating the model loss value, such as cross entropy loss function, maximum loss function, average loss function, etc. In the implementation method provided in this specification, the specific method of the loss function is not limited and is subject to actual application.

After calculating the gray matter loss value, white matter loss value and subcortical structure loss value, the model parameters of the anatomical segmentation model can be adjusted based on these three loss values. Specifically, the three loss values can be spliced and fused to obtain the model loss value, and the model loss value can be back-propagated to update the model parameters in the anatomical segmentation model.

After adjusting the model parameters of the anatomical segmentation model, the above steps can be repeated to continue training the anatomical segmentation model until the model training stop condition is reached. In practical applications, the training stop condition of the anatomical segmentation model includes:

The model loss value is less than a preset threshold; and/or

The training rounds reach the preset training rounds.

Specifically, during the training of the anatomical segmentation model, the training stop condition of the model can be set to that the model loss value is less than a preset threshold. Specifically, the gray matter loss value, the white matter loss value, and the subcortical structure loss value can be less than their respective corresponding preset thresholds, or the sum of the three can be less than the preset threshold, which is not limited in the method provided in this specification.

The training stop condition of the model can also be set to a preset training round number, for example, the model training is 20 rounds. When the model training rounds reach 20 times, it can be considered that the model training stop condition is met.

In practical applications, the loss value can be used as the training stop condition, the training round can be used as the training stop condition, or the loss value and the training round can be used as the training stop condition at the same time.

The method provided in the embodiments of this specification trains the anatomical segmentation model so that the anatomical segmentation model can complete the segmentation of gray matter, white matter and subcortical structure based only on the contrast information of the three-dimensional brain image, and can complete the segmentation using only the grayscale information in the image, and obtain gray matter information, white matter information, and subcortical structure information such as ventricles, cerebellum, thalamus, caudate nucleus, putamen, etc. The high efficiency of the machine learning model is utilized to improve the processing speed of whole brain segmentation.

Step 106: Generate a target cerebral cortex according to the gray matter information, and determine curvature characteristic information of at least one vertex on the target cerebral cortex.

After the above steps, the rough anatomical structure segmentation is completed, and the gray matter information, white matter information and masks of each subcortical structure are obtained. In the subsequent processing, the geometric features of the cerebral cortex are needed to perform more detailed cortical division. Furthermore, the curvature feature information of fixed points on the cerebral cortex is calculated.

In the method provided in the embodiment of this specification, only the necessary key step of calculating curvature is retained, that is, reconstructing the cerebral cortex. Specifically, after the above processing, gray matter information is obtained, and the target cerebral cortex corresponding to the three-dimensional brain image can be generated according to the gray matter information.

The target cerebral cortex is generated based on three-dimensional reconstruction. There are many gullies on the target cerebral cortex, and the target cerebral cortex after three-dimensional reconstruction is composed of many triangular facets. In the method provided in this specification, the cerebral cortex is further segmented according to the structural information of the cerebral cortex, wherein the structural information of the cerebral cortex can be understood as curvature feature information. Therefore, in the method provided in the embodiment of this specification, the target cerebral cortex is generated according to the gray matter information, and the curvature information of each vertex on the target cerebral cortex is determined for subsequent segmentation of the cerebral cortex.

In a specific embodiment provided in this specification, generating a target cerebral cortex according to the gray matter information includes:

Perform three-dimensional reconstruction based on the gray matter information to generate an initial cerebral cortex;

An iterative smoothing process is performed on the initial cerebral cortex to generate a target cerebral cortex.

In the method provided in the embodiment of this specification, in order to speed up the processing speed of the overall process, the steps of topological correction, spherical mapping, surface registration, etc. in the traditional processing method are abandoned, and only the step of cerebral cortex reconstruction is retained. Specifically, brain tissue structures such as cerebellum and brainstem are first removed, and other subcortical areas are filled. Three-dimensional reconstruction is performed based on gray matter information to generate the initial cerebral cortex.

The Marching Cubes Algorithm is a computer graphics algorithm used to extract isosurfaces from three-dimensional volume data. It is often used for three-dimensional shape reconstruction in fields such as medical imaging, geological exploration data, and fluid dynamics simulation.

The gray matter information includes the grayscale value corresponding to each voxel. Voxel is the abbreviation of volume element and is the smallest unit in three-dimensional space segmentation. The volume containing voxels can be expressed by stereo rendering or extracting polygonal isosurfaces with given threshold contours. Voxels are used in the fields of three-dimensional imaging, scientific data and medical impact. Its concept is similar to the smallest unit pixel in two-dimensional space.

A surface that is equal to or close to a preset threshold is extracted from the grayscale value of each voxel. This surface is called an isosurface. The algorithm divides the entire volume space into multiple cubic units. The center of each cube corresponds to a sampling point, and the eight vertices of the cube determine their scalar values through interpolation based on the values of the adjacent sampling points. For each cube, the algorithm checks which edges have the quality that crosses the isovalue threshold, and determines the intersection points of these edges through interpolation, thereby forming the contour line of the isosurface passing through the cube. When all intersection points are determined, the algorithm uses a pre-configured connection pattern to connect these intersection points into a series of triangular facets to form a smooth surface. These smooth surfaces constitute the initial cerebral cortex.

The initial cerebral cortex generated at this time still has an uneven surface, and the initial cerebral cortex can be further iteratively smoothed. Specifically, iterative smoothing is used to reduce the angularity of the initial cerebral cortex, making it look smoother and more natural. After iterative smoothing, the target cerebral cortex is generated.

Through the processing method provided in the embodiment of this specification, the entire reconstruction process takes less than 1 second, which simplifies the reconstruction process of the cerebral cortex and greatly reduces the processing time of reconstructing the cerebral cortex.

The target cerebral cortex is composed of a plurality of triangular facets, each of which includes three vertices, and the curvature information corresponding to each vertex represents the structural characteristic information of the target cerebral cortex. In a specific embodiment provided in this specification, determining the curvature characteristic information of at least one vertex on the target cerebral cortex includes:

Calculate the adjacent vertex curvature between each vertex and the corresponding adjacent vertex;

Calculate the vertex curvature corresponding to each vertex according to the curvature of the adjacent vertices corresponding to each vertex;

The vertex curvature corresponding to each vertex is mapped to the three-dimensional voxel space to determine the curvature feature information corresponding to each vertex.

In the traditional curvature calculation method, it is necessary to use methods such as polynomial fitting to model the surface, and then calculate the maximum curvature and minimum curvature of the surface based on the first-order and second-order derivatives. Finally, the maximum curvature and minimum curvature are averaged to obtain the average curvature. This process generally takes 2-3 minutes and is time-consuming and labor-intensive.

In the method provided in the embodiment of the present specification, the traditional calculation method is simplified. First, the adjacent vertex curvature between each vertex and the adjacent vertex is calculated, and the vertex curvature of each vertex is calculated through the adjacent vertex curvature.

At this time, the vertex curvature calculated for each vertex is the vertex on the three-dimensional visualization model. There will be voxels corresponding to each vertex in the three-dimensional voxel space. In order to perform a more refined partition of the cortex later, it is also necessary to map the curvature information of each vertex to the three-dimensional voxel space. Specifically, according to the correspondence between the three-dimensional visualization model and the voxels in the three-dimensional voxel space, the curvature corresponding to each vertex is added to the voxel feature information of the voxel corresponding to each vertex to represent the curvature feature information of each voxel.

In the method provided in the embodiment of this specification, a more simplified curvature calculation method is adopted. Specifically, the curvature of adjacent vertices between each vertex and the adjacent vertices corresponding to each vertex is calculated, including:

Determine a target vertex and an adjacent vertex corresponding to the target vertex, wherein the target vertex is any one of the at least one vertex;

Calculate the adjacent vertex curvatures corresponding to the target vertex and each adjacent vertex;

Accordingly, the vertex curvature corresponding to each vertex is calculated according to the curvature of the adjacent vertices corresponding to each vertex, including:

The target vertex curvature corresponding to the target vertex is determined according to the average of the curvatures of the adjacent vertices.

In practical applications, a target vertex and adjacent vertices corresponding to the target vertex are determined among multiple vertices, wherein the adjacent vertices can be understood as other vertices on the triangle where the target vertex is located.

The adjacent vertex curvatures between the target vertex and each adjacent vertex are calculated respectively, and the vertex curvature of the target vertex is determined according to the average value of the curvatures of each adjacent vertex.

For example, if a target vertex has 6 adjacent vertices, there are 6 adjacent vertex curvatures between the target vertex and each adjacent vertex. The average of the 6 adjacent vertex curvatures is calculated and used as the target vertex curvature corresponding to the target vertex.

Specifically, calculating the adjacent vertex curvatures corresponding to the target vertex and each adjacent vertex includes:

Determine a normal vector of the target vertex, wherein the normal vector is determined according to normal vectors of adjacent triangles corresponding to the target vertex;

Determine a target circle center corresponding to the target adjacent vertex according to a perpendicular bisector between the target vertex and the target adjacent vertex and the normal vector, wherein the target adjacent vertex is any one of the adjacent vertices corresponding to the target vertex;

The adjacent vertex curvature corresponding to the target adjacent vertex is determined according to the target circle center and the target vertex.

In practical applications, we take one of the target vertices as an example to explain, first determine the normal vector of the target vertex. The normal vector of the target vertex is calculated from the normal vectors of its adjacent triangles. In practical applications, first determine the adjacent triangles corresponding to the target vertex, and then perform unit normalization on the result based on the cross product of any two edges of each adjacent triangle to obtain the normal vector of the target vertex.

After determining the normal vector of each vertex, connect the target vertex with each adjacent vertex to obtain a local two-dimensional curve at the target vertex passing through each of its adjacent vertices. Draw the perpendicular bisector of the line connecting the target vertex and the target adjacent vertex. The intersection of the perpendicular bisector and the normal vector is the target circle center passing through the target adjacent vertex and the target vertex. Based on the target circle center and the target vertex, the radius r corresponding to the target circle center and the target vertex can be calculated, and then the adjacent vertex curvature 1/r corresponding to the target adjacent vertex can be determined.

All adjacent vertices corresponding to the target vertex are conveniently obtained, and the adjacent vertex curvature corresponding to each adjacent vertex is calculated by the above method. Finally, the average of the curvatures of each adjacent vertex is taken to obtain the target vertex curvature corresponding to the target vertex.

In the curvature calculation method provided in the embodiment of this specification, the curvature calculation method is simplified. There is no need to calculate the maximum curvature and minimum curvature of the surface based on the first-order and second-order derivatives. It is only necessary to calculate the adjacent vertex curvature between the target vertex and the adjacent vertex, and then determine the target vertex curvature of the target vertex based on the average of the adjacent vertex curvature. After experiments, this processing method only takes about 1.5 seconds, which greatly reduces the calculation time of curvature, saves computing resources, and further meets the requirements of clinical real-time applications.

Step 108: Determine a cortical surface segmentation result corresponding to the target cerebral cortex based on the gray matter information and the curvature feature information of at least one vertex on the target cerebral cortex.

After obtaining the curvature feature information of each voxel, the calculated curvature feature information can be spliced with the original input grayscale image in the voxel space, and the curvature feature information can be used to provide a reference for geometric features for the cortical surface segmentation of the cerebral cortex. Thus, the cortical surface segmentation result is determined based on the gray matter information and the curvature feature information of each vertex on the target cerebral cortex.

In a specific embodiment provided in this specification, determining a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and the curvature characteristic information of at least one vertex on the target cerebral cortex includes:

Cropping the three-dimensional brain image according to the gray matter information to obtain a three-dimensional gray matter brain image;

The three-dimensional gray matter brain image and curvature feature information of at least one vertex on the target cerebral cortex are input into a cortical segmentation model to obtain a cortical surface segmentation result output by the cortical segmentation model.

In practical applications, in order to better segment the cerebral cortex and eliminate the influence of the white matter and subcortical structures that have been segmented in the previous stage, in this embodiment, the three-dimensional brain image is cropped using the gray matter information to retain the three-dimensional gray matter brain image.

After the three-dimensional gray matter brain image and the curvature feature information of each vertex are spliced, they are input into the cortical segmentation model for processing to obtain the cortical surface segmentation result output by the cortical segmentation model.

The three-dimensional gray matter brain image includes the grayscale value corresponding to each voxel point, and the curvature feature information of each vertex represents the curvature corresponding to each voxel point. The three-dimensional gray matter brain image and the curvature feature information of each vertex are spliced to obtain the splicing feature information corresponding to each voxel point, and the splicing feature information corresponding to each voxel point is input into the cortical segmentation model to obtain the cortical surface segmentation result output by the cortical segmentation model.

The cortical segmentation model is trained to predict the brain structure region representation corresponding to each voxel point based on the splicing feature information of the voxel points. According to the brain structure region identification corresponding to each voxel point, the cerebral cortex can be segmented into different cortical surface segmentation results. The splicing feature information retains both the grayscale information of each voxel point and the geometric structure features corresponding to each voxel point, so that the cortical segmentation model can pay attention to the structural curvature features at the same time, solving the problem of low recognition accuracy in previous deep learning models.

In a specific embodiment provided in this specification, the cortical segmentation model is obtained by training through the following steps:

Acquire a sample three-dimensional brain image, sample gray matter information corresponding to the sample three-dimensional brain image, and a sample cortical surface segmentation result corresponding to the sample three-dimensional brain image;

Generate a sample cerebral cortex according to the sample gray matter information, and determine sample curvature characteristic information of at least one vertex on the sample cerebral cortex;

Cropping the sample three-dimensional brain image according to the gray matter information to obtain a sample three-dimensional gray matter brain image;

Inputting the sample three-dimensional gray matter brain image and sample curvature feature information of at least one vertex on the sample cerebral cortex into a cortical segmentation model to obtain a predicted cortical surface segmentation result output by the cortical segmentation model;

Calculating a model loss value according to the sample cortical surface segmentation result and the predicted cortical surface segmentation result;

The model parameters of the cortical segmentation model are adjusted according to the model loss value, and the cortical segmentation model is continuously trained until a model training stop condition is reached.

In the method provided in the embodiments of this specification, the cortical segmentation model is also trained in a supervised manner, which includes a sample three-dimensional brain image, sample gray matter information corresponding to the sample three-dimensional brain image, and a sample cortical surface segmentation result.

First, a three-dimensional reconstruction is performed based on the sample gray matter information to generate a sample cerebral cortex, and the sample curvature characteristic information of each vertex on the cerebral cortex is calculated. The method of performing three-dimensional reconstruction based on the sample gray matter information to generate a sample cerebral cortex and calculate the sample curvature characteristic information of each vertex is described above and will not be repeated here.

Please refer to the above description for the implementation method of inputting the sample curvature feature information corresponding to each vertex on the sample cerebral cortex into the cortical segmentation model for processing to obtain the predicted cortical surface segmentation result.

At this time, the cortical segmentation model is an untrained model. The model loss value can be calculated based on the predicted cortical surface segmentation results and the sample cortical surface segmentation results output by the model. The specific method for calculating the model loss value can be cross entropy loss function, maximum loss function, average loss function, etc. In the implementation mode provided in this specification, the specific method of the loss function is not limited and is subject to actual application.

After calculating the model loss value, the model parameters of the cortical segmentation model can be adjusted according to the model loss value, and the cortical segmentation model can continue to be trained with the sample data until the model training stop condition is reached.

In the method provided in the embodiment of this specification, in the fine cortical subdivision stage, the curvature feature information representing the structural features is input into the deep neural network, so that the deep neural network model can refer to the structural features, solving the problem of low recognition accuracy in previous deep learning solutions and improving the accuracy of model recognition.

Step 110: Determine a whole-brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information and the subcortical structure information.

The cortical surface segmentation results can be understood as the labels of the regions of interest of the cerebral cortex. The cerebral cortex is cropped using the gray matter mask output by the cortical segmentation model, and a corresponding label is assigned to each cropped region. This is then combined with the previously predicted white matter and subcortical structures to obtain the final whole-brain segmentation results.

The method provided in the embodiment of this specification provides two segmentation stages. In the first segmentation stage, a rough anatomical structure is segmented, and only a coarse filtering segmentation is performed based on the contrast information of the three-dimensional brain image. Gray matter information, white matter information and subcortical structure information are determined. In the second stage, the cerebral cortex is further generated based on the gray matter information, and the curvature information of each vertex on the cerebral cortex is calculated. The curvature information representing the structural features is input into the cortical segmentation model, so that the cortical segmentation model can refer to the structural information to further segment the cerebral cortex, and divide the cerebral cortex into multiple different subdivision areas, thereby obtaining the whole brain segmentation result. The whole operation process only takes about 10 seconds to obtain the whole brain segmentation result of a brain structure, which improves processing efficiency, reduces processing time, and improves processing precision and accuracy at the same time. Experiments on public data sets show that in the method provided in the embodiment of this specification, compared with other whole brain segmentation methods, the overall segmentation accuracy can be improved by about 7% at comparable speeds.

In a specific implementation provided in this specification, it also includes:

determining a target brain structure in the whole-brain segmentation result;

Detection is performed on the target brain structure to obtain segmentation and detection results corresponding to the target brain structure.

In a specific implementation provided in the embodiments of this specification, after obtaining the whole brain segmentation result, a subsequent diagnosis of the brain structure can be performed based on the whole brain segmentation result. Specifically, the target brain structure can be determined in the whole brain segmentation result, and the target brain structure can be tested to obtain the segmentation and test results corresponding to the target brain structure, and obtain the corresponding test conclusion.

For example, taking the target brain structure as the hippocampus as an example, after obtaining the whole-brain segmentation result based on the magnetic resonance image of a certain user, the hippocampus structure of the user can be determined from the whole-brain segmentation result, and by testing the hippocampus structure, it can be determined whether it has atrophied, thereby determining whether the user suffers from Alzheimer's disease.

For another example, if a user has symptoms of unstable gait, we can take a magnetic resonance image and perform whole-brain segmentation on the magnetic resonance image to identify the user's cerebellum structure and determine whether the cerebellum structure has atrophied, thereby determining whether the user's unstable gait symptoms are caused by cerebellar atrophy.

Through the method provided in the embodiments of this specification, a two-stage model processing method from coarse-grained to fine-grained is provided, which divides different types of segmentation tasks into two different stages, and uses different models to process their corresponding tasks respectively, thereby improving the processing speed of whole-brain segmentation and at the same time improving the accuracy of each model.

In the second stage of fine segmentation of the cerebral cortex, the curvature feature information and grayscale value of the voxel points are combined, which essentially improves the model's ability to understand the cortical segmentation task and obtains better cortical segmentation results.

FIG2 shows a schematic diagram of a processing structure of an image processing method provided in an embodiment of the present specification. As shown in FIG2 , in the method provided in the embodiment of the present specification, two processing stages are provided, namely a coarse segmentation stage and a fine segmentation stage.

In the rough segmentation stage, the three-dimensional brain image is input into the anatomical segmentation model for processing. The anatomical segmentation model performs segmentation according to the gray value and contrast information of each point in the three-dimensional brain image to obtain the anatomical segmentation result.

The anatomical segmentation results include gray matter information, white matter information, and subcortical structure information.

In the fine segmentation stage, the cerebral cortex is reconstructed based on the gray matter information. After the reconstruction is completed, the curvature information of each vertex on the cerebral cortex is calculated, and then the curvature information of each vertex is mapped back to the voxel space. After the curvature feature information and grayscale information in the voxel space are spliced, they are input into the cortical segmentation model for processing to obtain the cortical surface segmentation result.

Finally, the whole brain segmentation result is generated based on the cortical surface segmentation result and the white matter information and subcortical structure information generated in the rough segmentation stage.

The method provided in the embodiment of this specification provides two segmentation stages. In the first segmentation stage, a rough segmentation of the anatomical structure is performed, and a coarse filtering segmentation is performed only based on the contrast information of the three-dimensional brain image. Gray matter information, white matter information and subcortical structure information are determined. In the second stage, the cerebral cortex is further generated based on the gray matter information, and the curvature information of each vertex on the cerebral cortex is calculated. The curvature information representing the structural characteristics is input into the cortical segmentation model, so that the cortical segmentation model can refer to the structural information to further segment the cerebral cortex, and divide the cerebral cortex into multiple different subdivision areas, thereby obtaining the whole brain segmentation result. The whole operation process only takes about 10 seconds to obtain the whole brain segmentation result of a brain structure, which improves processing efficiency, reduces processing time, and improves processing precision and accuracy at the same time.

FIG3 shows an architecture diagram of an image processing system provided by an embodiment of the present specification. The image processing system may include a client 100 and a server 200;

The client 100 is used to send an image processing task to the server 200, wherein the image processing task carries a three-dimensional brain image;

The server 200 is used to input the three-dimensional brain image into the anatomical segmentation model, obtain the gray matter information, white matter information and subcortical structure information output by the anatomical segmentation model based on the contrast information of the three-dimensional brain image; generate a target cerebral cortex according to the gray matter information, and determine the curvature feature information of at least one vertex on the target cerebral cortex; determine the cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and the curvature feature information of at least one vertex on the target cerebral cortex; determine the whole brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information and the subcortical structure information; and send the whole brain segmentation result to the client 100;

The client 100 is also used to receive the whole brain segmentation result sent by the server 200.

The image processing system may include multiple clients 100 and a server 200, wherein the client 100 may be referred to as a client-side device and the server 200 may be referred to as a cloud-side device. A communication connection may be established between multiple clients 100 through the server 200. In the image processing scenario, the server 200 is used to provide image processing services between multiple clients 100. Multiple clients 100 may serve as a sender or a receiver respectively, and realize communication through the server 200.

The user can interact with the server 200 through the client 100 to receive data sent by other clients 100, or send data to other clients 100, etc. In the image processing scenario, the user can publish a data stream to the server 200 through the client 100, and the server 200 generates a whole brain segmentation result based on the data stream, and pushes the whole brain segmentation result to other clients that have established communication.

The client 100 and the server 200 are connected via a network. The network provides a medium for a communication link between the client 100 and the server 200. The network may include various connection types, such as wired or wireless communication links or optical fiber cables, etc. The data transmitted by the client 100 may need to be encoded, transcoded, compressed, etc. before being released to the server 200.

The client 100 can be a browser, an APP (Application), or a web application such as an H5 (HyperText Markup Language 5, Hypertext Markup Language Version 5) application, or a light application (also known as a mini-program, a lightweight application) or a cloud application, etc. The client 100 can be based on the software development kit (SDK, Software Development Kit) of the corresponding service provided by the server 200, such as based on the real-time communication (RTC, Real Time Communication) SDK development and acquisition. The client 100 can be deployed in an electronic device and needs to rely on the device to run or some APPs in the device to run. For example, the electronic device can have a display screen and support information browsing, such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications can also be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

The server 200 may include servers that provide various services, such as servers that provide communication services to multiple clients, servers for background training that support models used on clients, and servers that process data sent by clients. It should be noted that the server 200 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server of a distributed system, or a server combined with a blockchain. The server can also be a cloud server for 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 distribution networks (CDN, Content Delivery Network), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

It is worth noting that the image processing method provided in the embodiments of this specification is generally executed by the server, but in other embodiments of this specification, the client may also have similar functions to the server, thereby executing the image processing method provided in the embodiments of this specification. In other embodiments, the image processing method provided in the embodiments of this specification may also be executed jointly by the client and the server.

FIG4 is a flow chart of an image processing method applied to a cloud-side device provided in an embodiment of the present specification, which specifically includes:

Step 402: receiving an image processing task sent by the end side, wherein the image processing task carries a three-dimensional brain image.

Step 404: input the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image.

Step 406: Generate a target cerebral cortex according to the gray matter information, and determine curvature characteristic information of at least one vertex on the target cerebral cortex.

Step 408: Determine a cortical surface segmentation result corresponding to the target cerebral cortex based on the gray matter information and the curvature feature information of at least one vertex on the target cerebral cortex.

Step 410: Determine a whole-brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information and the subcortical structure information.

Step 412: Send the whole-brain segmentation result to the client-side device.

It should be noted that the line of sight method of step 402 to step 410 is the same as the line of sight method of step 102 to step 110 described above, and will not be described in detail in this specification.

Optionally, the method further includes:

A target detection task sent by a receiving end, wherein the target detection task carries a target brain structure identifier;

Determining a target brain structure in the whole-brain segmentation result according to the target brain structure identifier;

Performing detection on the target brain structure to obtain segmentation and detection results corresponding to the target brain structure;

The segmentation and detection results corresponding to the target brain structure are sent to the end-side device.

In actual applications, the cloud server can deploy the neural network model on the cloud and provide an API calling interface for the end-side device. Users can send image processing tasks to the cloud side through the AIP calling interface. The above image processing method is executed in the cloud-side device, and the final whole-brain segmentation result is returned to the end-side device. This method does not require the deployment of a neural network model on the end-side, realizes lightweight deployment of the end-side device, and reduces the hardware requirements for the end-side device. It improves the user experience.

FIG5 is a flowchart of a computer-aided diagnosis method for brain diseases provided in an embodiment of this specification, which specifically includes:

Step 502: Receive a brain disease detection task, wherein the brain disease detection task carries a three-dimensional brain image and a target brain structure identifier.

Step 504: input the three-dimensional brain image into the anatomical segmentation model to obtain gray matter information, white matter information and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image.

Step 506: Generate a target cerebral cortex according to the gray matter information, and determine curvature characteristic information of at least one vertex on the target cerebral cortex.

Step 508: Determine a cortical surface segmentation result corresponding to the target cerebral cortex based on the gray matter information and the curvature feature information of at least one vertex on the target cerebral cortex.

Step 510: Determine a whole-brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information and the subcortical structure information.

Step 512: Determine the target brain structure in the whole-brain segmentation result according to the target brain structure identifier, perform detection on the target brain structure, and obtain the segmentation and detection results corresponding to the target brain structure.

The brain disease detection method provided in the embodiments of this specification is suitable for detecting brain diseases that require examination of brain structures due to pathological changes in brain structures. Brain diseases include the detection of neurodegenerative diseases, such as Alzheimer’s disease (AD), Parkinson’s disease (PD), cerebellar atrophy, etc.; brain diseases also include mental illnesses, such as schizophrenia, depression, etc.; brain diseases also include neurodevelopmental diseases, such as autism, etc. After obtaining the whole-brain segmentation structure by accurately performing whole-brain segmentation on the three-dimensional brain image, the target brain structure is determined from the whole-brain segmentation result, and the target brain structure is further detected to determine whether the user suffers from the corresponding brain disease.

The method provided in the embodiment of this specification provides two segmentation stages. In the first segmentation stage, a rough segmentation of the anatomical structure is performed, and a coarse filtering segmentation is performed only based on the contrast information of the three-dimensional brain image. Gray matter information, white matter information and subcortical structure information are determined. In the second stage, the cerebral cortex is further generated based on the gray matter information, and the curvature information of each vertex on the cerebral cortex is calculated. The curvature information representing the structural characteristics is input into the cortical segmentation model, so that the cortical segmentation model can refer to the structural information to further segment the cerebral cortex, and divide the cerebral cortex into multiple different subdivision areas, thereby obtaining the whole brain segmentation result. The whole operation process only takes about 10 seconds to obtain the whole brain segmentation result of a brain structure, which improves processing efficiency, reduces processing time, and improves processing precision and accuracy at the same time.

Corresponding to the above method embodiment, this specification also provides an image processing device embodiment. FIG6 shows a schematic diagram of the structure of an image processing device provided by an embodiment of this specification. As shown in FIG6, the device includes:

The receiving module 602 is configured to receive an image processing task, wherein the image processing task carries a three-dimensional brain image.

The first segmentation module 604 is configured to input the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image.

The generation module 606 is configured to generate a target cerebral cortex according to the gray matter information, and determine curvature characteristic information of at least one vertex on the target cerebral cortex.

The second segmentation module 608 is configured to determine a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and curvature feature information of at least one vertex on the target cerebral cortex.

The determination module 610 is configured to determine the whole brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information and the subcortical structure information.

The device provided in the embodiment of this specification provides two segmentation stages. In the first segmentation stage, a rough anatomical structure segmentation is performed, and a coarse filtering segmentation is performed only based on the contrast information of the three-dimensional brain image. Gray matter information, white matter information and subcortical structure information are determined. In the second stage, the cerebral cortex is further generated based on the gray matter information, and the curvature information of each vertex on the cerebral cortex is calculated. The curvature information representing the structural characteristics is input into the cortical segmentation model, so that the cortical segmentation model can refer to the structural information to further segment the cerebral cortex, and divide the cerebral cortex into multiple different subdivision areas, thereby obtaining the whole brain segmentation result. The whole operation process only takes about 10 seconds to obtain the whole brain segmentation result of a brain structure, which improves processing efficiency, reduces processing time, and improves processing precision and accuracy at the same time.

The above is a schematic scheme of an image processing device of this embodiment. It should be noted that the technical scheme of the image processing device and the technical scheme of the above-mentioned image processing method belong to the same concept, and the details not described in detail in the technical scheme of the image processing device can be referred to the description of the technical scheme of the above-mentioned image processing method.

Fig. 7 shows a block diagram of a computing device 700 according to an embodiment of the present specification. The components of the computing device 700 include but are not limited to a memory 710 and a processor 720. The processor 720 is connected to the memory 710 via a bus 730, and the database 750 is used to store data.

The computing device 700 also includes an access device 740 that enables the computing device 700 to communicate via one or more networks 760. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 740 may include one or more of any type of network interface (e.g., a network interface card (NIC)) of wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a world-wide interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, and a near field communication (NFC).

In one embodiment of the present specification, the above components of the computing device 700 and other components not shown in FIG. 7 may also be connected to each other, for example, through a bus. It should be understood that the computing device structure block diagram shown in FIG. 7 is only for illustrative purposes and is not intended to limit the scope of the present specification. Those skilled in the art may add or replace other components as needed.

The computing device 700 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 700 may also be a mobile or stationary server.

The processor 720 is used to execute the following computer program/instructions, which implement the steps of the above-mentioned image processing method when executed by the processor.

Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computing device embodiment, since it is basically similar to the image processing method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the image processing method embodiment.

An embodiment of the present specification further provides a computer-readable storage medium storing a computer program/instruction, which implements the steps of the above-mentioned image processing method when executed by a processor.

Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computer-readable storage medium embodiment, since it is basically similar to the image processing method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the image processing method embodiment.

An embodiment of the present specification further provides a computer program product, including a computer program/instruction, which implements the steps of the above-mentioned image processing method when executed by a processor.

The above is a schematic solution of a computer program product of this embodiment. It should be noted that the technical solution of the computer program product and the technical solution of the above-mentioned image processing method belong to the same concept, and the details not described in detail in the technical solution of the computer program product can be referred to the description of the technical solution of the above-mentioned image processing method.

The above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

It should be noted that the above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of the present specification.

In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can understand and use this specification well. This specification is only limited by the claims and their full scope and equivalents.

Claims (14)

1. An image processing method, comprising: receiving an image processing task, wherein the image processing task carries a three-dimensional brain image; Inputting the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image; Generate a target cerebral cortex according to the gray matter information, and determine a target vertex and an adjacent vertex corresponding to the target vertex, wherein the target vertex is any one of at least one vertex; determine a normal vector of the target vertex, wherein the normal vector is determined according to a normal vector of an adjacent triangle corresponding to the target vertex; determine a target center point corresponding to the target adjacent vertex according to a perpendicular bisector between the target vertex and the target adjacent vertex and the normal vector, wherein the target adjacent vertex is any one of the adjacent vertices corresponding to the target vertex; determine an adjacent vertex curvature corresponding to the target adjacent vertex according to the target center point and the target vertex; calculate a vertex curvature corresponding to each vertex according to the adjacent vertex curvature corresponding to each vertex; map the vertex curvature corresponding to each vertex to a three-dimensional voxel space, and determine curvature feature information corresponding to each vertex; Determine a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and curvature feature information of at least one vertex on the target cerebral cortex; A whole-brain segmentation result corresponding to the three-dimensional brain image is determined according to the cortical surface segmentation result, the white matter information and the subcortical structure information. 2. The method according to claim 1, generating a target cerebral cortex according to the gray matter information, comprising: Perform three-dimensional reconstruction based on the gray matter information to generate an initial cerebral cortex; An iterative smoothing process is performed on the initial cerebral cortex to generate a target cerebral cortex. 3. The method according to claim 1, wherein the vertex curvature corresponding to each vertex is calculated according to the curvature of the adjacent vertices corresponding to each vertex, comprising: The target vertex curvature corresponding to the target vertex is determined according to the average of the curvatures of the adjacent vertices. 4. The method according to claim 1, determining a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and the curvature characteristic information of at least one vertex on the target cerebral cortex, comprising: Cropping the three-dimensional brain image according to the gray matter information to obtain a three-dimensional gray matter brain image; The three-dimensional gray matter brain image and curvature feature information of at least one vertex on the target cerebral cortex are input into a cortical segmentation model to obtain a cortical surface segmentation result output by the cortical segmentation model. 5. The method of claim 1, further comprising: determining a target brain structure in the whole-brain segmentation result; Detection is performed on the target brain structure to obtain segmentation and detection results corresponding to the target brain structure. 6. The method according to claim 1, wherein the anatomical segmentation model is obtained by training through the following steps: Acquire a sample three-dimensional brain image and sample gray matter information, sample white matter information, and sample subcortical structure information corresponding to the sample three-dimensional brain image; Inputting the sample three-dimensional brain image into an anatomical segmentation model to obtain predicted gray matter information, predicted white matter information, and predicted subcortical structure information output by the anatomical segmentation model based on contrast information in the sample three-dimensional brain image; Calculating a gray matter loss value according to the sample gray matter information and the predicted gray matter information, calculating a white matter loss value according to the sample white matter information and the predicted white matter information, and calculating a subcortical structure loss value according to the sample subcortical structure information and the predicted subcortical structure information; The model parameters of the anatomical segmentation model are adjusted according to the gray matter loss value, the white matter loss value and the subcortical structure loss value, and the anatomical segmentation model is continuously trained until a model training stop condition is reached. 7. The method according to claim 4, wherein the cortical segmentation model is obtained by training through the following steps: Acquire a sample three-dimensional brain image, sample gray matter information corresponding to the sample three-dimensional brain image, and a sample cortical surface segmentation result corresponding to the sample three-dimensional brain image; Generate a sample cerebral cortex according to the sample gray matter information, and determine sample curvature characteristic information of at least one vertex on the sample cerebral cortex; Cropping the sample three-dimensional brain image according to the gray matter information to obtain a sample three-dimensional gray matter brain image; Inputting the sample three-dimensional gray matter brain image and sample curvature feature information of at least one vertex on the sample cerebral cortex into a cortical segmentation model to obtain a predicted cortical surface segmentation result output by the cortical segmentation model; Calculating a model loss value according to the sample cortical surface segmentation result and the predicted cortical surface segmentation result; The model parameters of the cortical segmentation model are adjusted according to the model loss value, and the cortical segmentation model is continuously trained until a model training stop condition is reached. 8. An image processing method, applied to a cloud-side device, comprising: An image processing task sent by a receiving end, wherein the image processing task carries a three-dimensional brain image; Inputting the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image; Generate a target cerebral cortex according to the gray matter information, and determine a target vertex and an adjacent vertex corresponding to the target vertex, wherein the target vertex is any one of at least one vertex; determine a normal vector of the target vertex, wherein the normal vector is determined according to a normal vector of an adjacent triangle corresponding to the target vertex; determine a target center point corresponding to the target adjacent vertex according to a perpendicular bisector between the target vertex and the target adjacent vertex and the normal vector, wherein the target adjacent vertex is any one of the adjacent vertices corresponding to the target vertex; determine an adjacent vertex curvature corresponding to the target adjacent vertex according to the target center point and the target vertex; calculate a vertex curvature corresponding to each vertex according to the adjacent vertex curvature corresponding to each vertex; map the vertex curvature corresponding to each vertex to a three-dimensional voxel space, and determine curvature feature information corresponding to each vertex; Determine a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and curvature feature information of at least one vertex on the target cerebral cortex; Determining a whole-brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information, and the subcortical structure information; The whole-brain segmentation result is sent to the client-side device. 9. The method of claim 8, further comprising: A target detection task sent by a receiving end, wherein the target detection task carries a target brain structure identifier; Determining a target brain structure in the whole-brain segmentation result according to the target brain structure identifier; Performing detection on the target brain structure to obtain segmentation and detection results corresponding to the target brain structure; The segmentation and detection results corresponding to the target brain structure are sent to the end-side device. 10. A computer-aided diagnosis method for brain diseases, comprising: receiving a brain disease detection task, wherein the brain disease detection task carries a three-dimensional brain image and a target brain structure identifier; Inputting the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image; Generate a target cerebral cortex according to the gray matter information, and determine a target vertex and an adjacent vertex corresponding to the target vertex, wherein the target vertex is any one of at least one vertex; determine a normal vector of the target vertex, wherein the normal vector is determined according to a normal vector of an adjacent triangle corresponding to the target vertex; determine a target center point corresponding to the target adjacent vertex according to a perpendicular bisector between the target vertex and the target adjacent vertex and the normal vector, wherein the target adjacent vertex is any one of the adjacent vertices corresponding to the target vertex; determine an adjacent vertex curvature corresponding to the target adjacent vertex according to the target center point and the target vertex; calculate a vertex curvature corresponding to each vertex according to the adjacent vertex curvature corresponding to each vertex; map the vertex curvature corresponding to each vertex to a three-dimensional voxel space, and determine curvature feature information corresponding to each vertex; Determine a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and curvature feature information of at least one vertex on the target cerebral cortex; Determining a whole-brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information, and the subcortical structure information; The target brain structure is determined in the whole-brain segmentation result according to the target brain structure identifier, and the target brain structure is detected to obtain the segmentation and detection results corresponding to the target brain structure. 11. An image processing device, comprising: A receiving module, configured to receive an image processing task, wherein the image processing task carries a three-dimensional brain image; A first segmentation module is configured to input the three-dimensional brain image into an anatomical segmentation model to obtain gray matter information, white matter information, and subcortical structure information output by the anatomical segmentation model based on contrast information of the three-dimensional brain image; A generation module is configured to generate a target cerebral cortex according to the gray matter information, and determine a target vertex and an adjacent vertex corresponding to the target vertex, wherein the target vertex is any one of at least one vertex; determine a normal vector of the target vertex, wherein the normal vector is determined according to a normal vector of an adjacent triangle corresponding to the target vertex; determine a target center point corresponding to the target adjacent vertex according to a perpendicular bisector between the target vertex and the target adjacent vertex and the normal vector, wherein the target adjacent vertex is any one of the adjacent vertices corresponding to the target vertex; determine an adjacent vertex curvature corresponding to the target adjacent vertex according to the target center point and the target vertex; calculate a vertex curvature corresponding to each vertex according to the adjacent vertex curvature corresponding to each vertex; map the vertex curvature corresponding to each vertex to a three-dimensional voxel space, and determine curvature feature information corresponding to each vertex; A second segmentation module is configured to determine a cortical surface segmentation result corresponding to the target cerebral cortex according to the gray matter information and curvature feature information of at least one vertex on the target cerebral cortex; The determination module is configured to determine the whole brain segmentation result corresponding to the three-dimensional brain image according to the cortical surface segmentation result, the white matter information and the subcortical structure information. 12. A computing device comprising: Memory and processor; The memory is used to store computer programs/instructions, and the processor is used to execute the computer programs/instructions. When the computer programs/instructions are executed by the processor, the steps of the method described in any one of claims 1 to 10 are implemented. 13. A computer-readable storage medium storing a computer program/instruction, wherein the computer program/instruction, when executed by a processor, implements the steps of the method according to any one of claims 1 to 10. 14. A computer program product, comprising a computer program/instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 10.