Medical image registration method, device and storage medium based on artificial intelligence

Abstract

The present invention provides a medical image registration method, device and storage medium based on artificial intelligence, obtaining a medical image group, wherein each medical image group includes a floating image and a fixed image; identifying the first feature point of the medical image group according to the description information of the medical image group and the feature point extraction algorithm; wherein the first feature point is a pixel point in the fixed image whose difference in feature value with surrounding pixel points is greater than a preset difference; calculating the complexity of the first region where each first feature point is located; selecting a target registration model corresponding to each first feature point from a plurality of pre-established registration models according to the complexity; registering the medical image group according to the first feature point and the target registration model corresponding to each first feature point, and obtaining a registered image. By dividing the two images to be registered, and then using different registration models for registration according to the complexity of each region, that is, adapting a more suitable registration model to each region, the fuzzy characteristics of the registration model are effectively reduced, and the registration effect is improved.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and in particular relates to a medical image registration method, device and storage medium based on artificial intelligence. Background Art

[0002] In the process of medical image acquisition and disease prediction, medical images usually need to be registered to assist in disease prediction. Image registration refers to seeking a (or a series of) geometric space transformations between two (or more) medical images at different times, different fields of view, different modalities, or even different subjects, so that the image after spatial transformation can match another (or more) medical image. Specifically, the medical images obtained at different time points or under different imaging conditions are spatially aligned and fused. The registered images can more accurately reflect the anatomical structure and pathological changes of the human body, so that doctors can compare patient images from multiple times or multiple angles, which is of great value in disease diagnosis, treatment planning, and efficacy evaluation. This matching makes anatomical points or points of interest with the same anatomical structure have a one-to-one correspondence in the geometric space of two (or more) medical images. The image that remains fixed during the registration process is called the reference image, and the image that undergoes spatial transformation is called the floating image.

[0003] Existing registration methods include traditional iterative optimization registration algorithms and deep learning methods. However, iterative optimization registration algorithms require manual design of complex image features or setting of control points to control registration accuracy. Specific registration models and optimization targets need to be designed for specific applications. The process requires manual intervention and is greatly affected by human subjectivity, making it difficult to automate. Although deep learning methods can achieve fast registration, their registration accuracy and generalization ability are affected by many factors such as model structure, training data, training strategy, and optimization objective function. There are problems such as low registration accuracy and blurred deformation fields in local areas, areas with complex anatomical structures, and boundary areas. Therefore, the image registration effect of conventional registration methods is poor. Summary of the invention

[0004] In view of this, the present invention provides a medical image registration method, device and storage medium based on artificial intelligence, aiming to solve the problem of poor image registration effect of conventional registration methods.

[0005] A first aspect of an embodiment of the present invention provides a medical image registration method based on artificial intelligence, comprising:

[0006] Acquire a medical image group, wherein each medical image group includes a floating image and a fixed image;

[0007] According to the description information of the medical image group and the feature point extraction algorithm, a first feature point of the medical image group is identified; wherein the first feature point is a pixel point in the fixed image whose feature value difference with surrounding pixel points is greater than a preset difference value;

[0008] Calculating the complexity of the first area where each first feature point is located;

[0009] Selecting a target registration model corresponding to each first feature point from a plurality of pre-established registration models according to the complexity;

[0010] The medical image group is registered according to the first feature points and the target registration model corresponding to each first feature point to obtain a registered image.

[0011] A second aspect of an embodiment of the present invention provides a medical image registration method and device based on artificial intelligence, comprising:

[0012] An acquisition module, used for acquiring a medical image group, wherein each medical image group includes a floating image and a fixed image;

[0013] An identification module, used to identify a first feature point of the medical image group according to the description information of the medical image group and a feature point extraction algorithm; wherein the first feature point is a pixel point in the fixed image whose feature value difference with surrounding pixel points is greater than a preset difference value;

[0014] A calculation module, used for calculating the complexity of the first area where each first feature point is located;

[0015] A selection module, used for selecting a target registration model corresponding to each first feature point from a plurality of pre-established registration models according to complexity;

[0016] The registration module is used to register the medical image group according to the first feature points and the target registration model corresponding to each first feature point to obtain a registered image.

[0017] A third aspect of an embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the artificial intelligence-based medical image registration method of the first aspect above when executing the computer program.

[0018] The fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the artificial intelligence-based medical image registration method of the first aspect above are implemented.

[0019] The artificial intelligence-based medical image registration method, device and storage medium provided by the embodiment of the present invention obtain a medical image group, wherein each medical image group includes a floating image and a fixed image; according to the description information of the medical image group and the feature point extraction algorithm, the first feature point of the medical image group is identified; wherein the first feature point is a pixel point in the fixed image whose difference in feature value with the surrounding pixel points is greater than a preset difference; the complexity of the first area where each first feature point is located is calculated; according to the complexity, the target registration model corresponding to each first feature point is selected from multiple pre-established registration models; the medical image group is registered according to the first feature point and the target registration model corresponding to each first feature point to obtain a registered image. By dividing the two images to be registered, and then using different registration models for registration according to the complexity of each area, that is, adapting a more suitable registration model to each area, the fuzzy characteristics of the registration model are effectively reduced, and the registration effect is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 is a flowchart of an implementation of an artificial intelligence-based medical image registration method provided in an embodiment of the present invention;

[0022] Figure 2 It is a structural schematic diagram of a medical image registration method and device based on artificial intelligence provided by an embodiment of the present invention;

[0023] Figure 3 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0025] Figure 1 FIG. 1 is a flowchart of an implementation of an artificial intelligence-based medical image registration method provided by an embodiment of the present invention. Figure 1 As shown, in some embodiments, the medical image registration method based on artificial intelligence includes:

[0026] S110, obtaining a medical image group, wherein each medical image group includes a floating image and a fixed image.

[0027] In the embodiment of the present invention, the medical image may be a CT image, an MRI image, a PET image, etc., which is not limited here. The two images to be registered in the medical image group may be two medical images of the same type but different positions, or two images of different types but the same position, which is not limited here. Specifically, the floating image and the fixed image in the image group may be customized by medical personnel, which is not limited here.

[0028] S120, identifying a first feature point of the medical image group according to the description information of the medical image group and a feature point extraction algorithm; wherein the first feature point is a pixel point in the fixed image whose feature value difference with surrounding pixel points is greater than a preset difference.

[0029] In the embodiment of the present invention, the optical flow field refers to the apparent motion of the grayscale pattern of the image. It is a two-dimensional vector field, which is a two-dimensional instantaneous velocity field composed of all pixels in the image, in which the two-dimensional velocity vector is the projection of the three-dimensional velocity vector of the visible point in the scene on the imaging surface. The optical flow field not only contains the motion information of the observed object, but also contains rich information about the three-dimensional structure of the scene. Optical flow field analysis can identify and track feature points in the image sequence, which remain unchanged or have a stable motion pattern in different image frames. These feature points can be used as reference points for image registration, and the correspondence between different images can be found by matching these feature points.

[0030] In an embodiment of the present invention, the characteristic value may include but is not limited to at least one of the following: grayscale, curvature, optical flow of the optical flow field, and displacement value of the displacement deformation field. Among them, the recognition of grayscale, the recognition of curvature, the calculation of optical flow, and the calculation of displacement value can all be calculated by conventional methods and are not limited here. Since the first feature point is significantly different from the surrounding pixels, it is easier to be identified during registration, so it can be used as a key point in the image, and the pixels in the area are registered with the key point as the center.

[0031] S130, calculating the complexity of the first area where each first feature point is located;

[0032] In the embodiment of the present invention, the first region may be a rectangular region of a preset size or a circular region of a preset size, which is not limited here. The complexity of the first region may be determined by the average of the difference between the characteristic values ​​of each pixel point and its adjacent pixel points in the region. For example, if the characteristic value is grayscale, the complexity of the first region is:

[0033] (1)

[0034] in, η For complexity, a i For the i The gray level of each pixel, a j For the i Pixels adjacent to each other j The grayscale of the pixel, n is the total number of pixels. a i minus a j The sum of the absolute values ​​of n ! / 2, which is the complexity of the first region.

[0035] For example, when the eigenvalues ​​include grayscale, curvature, optical flow of the optical flow field, and displacement value of the displacement deformation field, the complexity of the first region is:

[0036] (2)

[0037] in, η For complexity, a i For the i The gray level of each pixel, a j For the i Pixels adjacent to each other j The grayscale of the pixel, b i For the i The curvature of a pixel, b j For the i Pixels adjacent to each other j The curvature of the pixel, c i For the i The optical flow of pixels, c j For the i Pixels adjacent to each other j The optical flow of the pixel, d i For the i The displacement value of each pixel, d j For the i Pixels adjacent to each other j The displacement value of the pixel point, n is the total number of pixels. k 1. k 2. k 3. k4 is the preset coefficient corresponding to each type of eigenvalue, k 1+ k 2+ k 3+ k 4=1. Subtract the adjacent eigenvalue from the eigenvalue of each pixel, multiply the calculation result corresponding to each eigenvalue by its own preset coefficient, and then divide it by the total number of pixel combinations. n ! / 2, which is the complexity of the first region.

[0038] S140, selecting a target registration model corresponding to each first feature point from a plurality of pre-established registration models according to the complexity.

[0039] In an embodiment of the present invention, the target registration model may specifically be a neural network model, a deep learning model, an adversarial network model, etc., which are not limited here. The specific type of model to be selected needs to be set according to the image to be registered. The adversarial network can usually generate very high-quality and diverse data through the confrontation between the generator and the discriminator, so it can be applied to the registration with higher complexity, otherwise a convolutional neural network is used. However, the adversarial network training process itself is unstable and is easily affected by problems such as mode collapse, gradient disappearance or explosion. When processing multimodal data, the training difficulty will be further increased because multiple generators and discriminators need to be optimized at the same time, and the optimization goals between different modalities may conflict, and the convolutional neural network is not affected by this factor. Therefore, the adversarial network can be used when the types of two medical images in the medical image group are the same, and the convolutional neural network is not used at the same time. If the image to be registered is highly complex and not of the same type, an adversarial convolutional neural network is used. When training the model, medical images of normal people in a healthy state can be used as references for training.

[0040] S150, registering the medical image group according to the first feature points and the target registration model corresponding to each first feature point to obtain a registered image.

[0041] In some embodiments, S130 may include: determining a first image parameter based on description information of the medical image group; calculating a second image parameter based on feature values ​​of each pixel point in the first area; and calculating the complexity of the first feature point based on the first image parameter and the second image parameter value.

[0042] In the embodiment of the present invention, the values ​​calculated in the above formulas (1) and (2) can be η As the second image parameter, multiplying it by the first image parameter can obtain the final complexity.

[0043] The description information of the medical image group, that is, the type of fixed image and the type of floating image. Different medical images have different advantages. For example, CT images have a high spatial resolution and can clearly display various tissues and organs in the human body, especially the best display effect for bones and calcified tissues. MRI images have a high resolution for soft tissues and can clearly display soft tissue structures such as muscles, fat, nerves and blood vessels. They have unique advantages for imaging the central nervous system, spine and joints.

[0044] Therefore, in the embodiment of the present invention, the first image parameter can be introduced to reflect the influence of the difference between the two images to be registered on the registration. For example, in the registration between two CT images, the registered images belong to the same type, and the first image parameter can be set to 1, that is, the type of medical effect has no influence on the registration. For the registration between MRI images and CT images, the first complexity is λ 1 η , λ 1 is the first imaging parameter, which is a value greater than or equal to 1, obtained based on expert evaluation and drawn into a table. When actually used, the table can be looked up based on the description information of the medical imaging group.

[0045] In an embodiment of the present invention, the description information of the medical image group can be obtained from a business system or device. The description information of the medical image group may also include image type, imaging time, and shooting posture. The corresponding first image parameter is selected according to the image type, imaging time, shooting posture, etc. For example, the greater the difference in image type, the longer the time interval of imaging time, the greater the angle between the two images, and the greater the first image parameter. The shooting posture is specifically the incident angle of the shooting ray relative to the human body part.

[0046] In some embodiments, S150 may include: calculating the corresponding mapping function of the first region where each first feature point is located in the floating image according to the target registration model corresponding to each first feature point, registering the floating image with the fixed image, and obtaining a registered image.

[0047] In an embodiment of the present invention, the target registration model is a convolutional neural network model. Due to the complexity of the human body structure, medical images of different parts of the human body have great differences. These differences often lead to the problem of partial pixel mapping errors when the same neural network is used as a registration model for registration, resulting in certain distortion of the registered image. After the present invention divides the fixed image into multiple first areas, different registration models are assigned to each first area, so as to more accurately calculate the mapping relationship between the floating image and the fixed image.

[0048] Registration between medical images is a challenging task because they may be affected by many factors, resulting in differences between images. Among them, the images of the same patient taken at different imaging times can well reflect the patient's pathological changes after registration, but the position of the organs may change due to breathing or heartbeat, and even the same patient has different anatomical structures at different times. Different image types (CT, MRI, etc.) will have different information such as image contrast, brightness, and resolution, which will have a certain impact on registration. After registering different types of images, more detailed information about the patient's various parts can be provided as a reference for doctors' diagnosis. Registering medical images taken in different shooting postures can convert the patient's 2D image into a 3D medical image and display the patient's information in a three-dimensional manner. In medical imaging, in addition to the anteroposterior (front-to-back) and lateral (left-right) positions, there are many other shooting postures to facilitate the observation of the internal structure of the body from different angles and directions. For example: head CT scan: including sagittal, coronal, frontal, etc., for comprehensive observation of the head and brain structure. Spine CT scan: may include oblique and tangential positions to evaluate abnormalities or injuries of the spine.

[0049] In an embodiment of the present invention, after selecting the corresponding first image parameter according to the above-mentioned image type, imaging time, shooting posture, etc. and changing the calculated complexity, a suitable registration model can be selected according to the image type, imaging time, and shooting posture, so as to calculate a more suitable mapping relationship for the two images. That is, the mapping between images of the same patient at different angles, the mapping between the same angle at different times under the same imaging conditions, and the mapping between different types of medical images. By considering the influence of information such as image type, imaging time, and shooting posture on the registration mapping, a more accurate mapping relationship can be obtained to complete the registration of floating images and fixed images.

[0050] In an embodiment of the present invention, multiple registration models can be the same model, the difference is that the training set and the test set of the model are different. For example, multiple registration images, original floating images, and original fixed images annotated by doctors can be obtained in advance, and then the above three images are divided into regions, and the images of the area with complexity less than the preset complexity are used as the training set and test set of one registration model, and the images of the area with complexity not less than the preset complexity are used as the training set and test set of another registration model. The two models are trained separately to obtain two different registration models. One of the two models has high recognition and high registration effect for low-complexity images, and the other has high recognition and high registration effect for high-complexity images.

[0051] In an embodiment of the present invention, the network loss function is a conventional neural network loss function, the difference being that the feedforward value is different, and it usually includes three parts: one is the similarity loss of calculating the registered image and the fixed image, the second is the intra-region smoothness loss to ensure the continuity and smoothness of each first region after registration, and the third is the interval smoothness loss, which is the continuity and smoothness of each first region and the adjacent first region, and between the adjacent first regions after registration.

[0052] Among them, the similarity loss is calculated based on the difference between the fixed image and the floating image. The similarity loss function is used to measure the matching degree of two images after spatial transformation. By minimizing the similarity loss, accurate registration of the floating image to the fixed image can be achieved. This process is crucial for tasks such as image stitching, stereo matching, and motion tracking. The loss function can be specifically least squares loss, cross-correlation loss, etc., which are not limited here.

[0053] Among them, smoothness loss is an important concept in image registration, which is used to measure the continuity and smoothness of the deformation field. A smooth deformation field helps to avoid local distortion and unnatural transformations in the image registration process, thereby improving the overall quality of the registration. By minimizing the smoothness loss, it can be ensured that the entire image range is continuous and smooth, which helps to avoid unnatural distortions or discontinuities in image registration. The loss function can be specifically the sum of squared gradients, the sum of squared second-order derivatives, etc., which are not limited here.

[0054] Since the present invention performs regional division, the smoothing loss can be divided into intra-region smoothing loss and interval smoothing loss. The intra-region smoothing loss is the operation of the smoothing loss function of a single neural network, which will not be described here. The various registration models of the present invention are not completely isolated when running, but have certain loss sharing characteristics, that is, each adjacent first region will pass its smoothing loss function, the loss value calculated at the region boundary, to the registration model of the adjacent first region on the boundary. The loss functions between adjacent regions are transmitted to each other, which can avoid the boundary differences between two adjacent regions being too large after registration, resulting in obvious boundary splitting.

[0055] In the embodiment of the present invention, after the floating image and the image of the first region are input into the registration model corresponding to the first region, the model identifies the pixel points corresponding to each pixel point of the first region in the floating image, and maps the pixel points in the floating image to the fixed image, completing the registration process corresponding to the first region. By traversing each first region, the registration of the entire fixed image can be completed.

[0056] In some embodiments, S120 may include: identifying the initial first feature points of the medical image group according to a scale-invariant feature conversion algorithm; calculating the evaluation value of each initial first feature point according to the description information of the medical image group; and taking the initial first feature point whose evaluation value is greater than a preset threshold as the first feature point.

[0057] In an embodiment of the present invention, the scale-invariant feature conversion algorithm is a commonly used feature extraction method. After extracting a large number of pixels whose feature values ​​differ from those of surrounding pixels by a value greater than a preset value, due to the differences in the body parts and the types of medical images, the feature points that best represent the key features of the image are necessarily different, so it is necessary to further screen these extracted pixels. Specifically, different evaluation indicators can be set for each body part and each type of medical image. For example, after the fixed image is a head CT image, the extracted descriptive information is "head" and "CT", and the evaluation indicators corresponding to these two information items in the basic query database are used in turn to evaluate the first feature point, and the weighted sum of the evaluation values ​​of each evaluation indicator is used as the final evaluation value. The descriptive information of the specific medical image group may also include patient information such as the patient's age, the type of population to which he belongs, and gender. According to the different conditions of the patients, the above-mentioned indicators are evaluated in combination with the type and angle of the medical image to obtain the evaluation value.

[0058] In some embodiments, S120 may include: identifying the initial first feature point of the medical image group according to a scale-invariant feature conversion algorithm; determining the constraint function according to the description information of the medical image group; optimizing each initial first feature point according to a particle swarm optimization algorithm with the goal of maximizing the difference of eigenvalues ​​to obtain the first feature point.

[0059] In the embodiment of the present invention, in addition to the above-mentioned index screening method, it can also be optimized by particle swarm algorithm.

[0060] In some embodiments, multiple registration models include a first deep learning model and a second deep learning model; S140 may include: when the complexity of the area where the first feature point is located is greater than the preset complexity, using the first deep learning model as the target registration model for the first feature point; when the complexity of the area where the first feature point is located is not greater than the preset complexity, using the second deep learning model as the target registration model for the first feature point.

[0061] The beneficial effects of the present invention are as follows: by dividing the two images to be registered, and then using different registration models for registration according to the complexity of each area, that is, adapting a more appropriate registration model to each area, the fuzzy characteristics of the registration model are effectively reduced, and the registration effect is improved.

[0062] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0063] Figure 2 FIG. 1 is a schematic diagram of the structure of a medical image registration method and device based on artificial intelligence provided by an embodiment of the present invention. Figure 2 As shown, in some embodiments, an artificial intelligence-based medical image registration method and device 2 includes:

[0064] An acquisition module 210, for acquiring a medical image group, wherein each medical image group includes a floating image and a fixed image;

[0065] The identification module 220 is used to identify the first feature point of the medical image group according to the description information of the medical image group and the feature point extraction algorithm; wherein the first feature point is a pixel point in the fixed image whose feature value difference with the surrounding pixel points is greater than a preset difference value;

[0066] A calculation module 230, configured to calculate the complexity of the first area where each first feature point is located;

[0067] A selection module 240, configured to select a target registration model corresponding to each first feature point from a plurality of pre-established registration models according to complexity;

[0068] The registration module 250 is used to register the medical image group according to the first feature points and the target registration model corresponding to each first feature point to obtain a registered image.

[0069] Optionally, the calculation module 230 is specifically used to determine the first image parameter according to the description information of the medical image group; calculate the second image parameter according to the characteristic value of each pixel point in the first area; and calculate the complexity of the first feature point according to the first image parameter and the second image parameter.

[0070] Optionally, a registration module 250 is used to calculate a corresponding mapping function of the first region where each first feature point is located in the floating image according to a target registration model corresponding to each first feature point;

[0071] According to the mapping function, the floating image is registered with the fixed image to obtain a registered image.

[0072] Optionally, the identification module 220 is used to identify the initial first feature points of the medical image group based on the scale-invariant feature conversion algorithm; calculate the evaluation value of each initial first feature point based on the description information of the medical image group; and take the initial first feature point with an evaluation value greater than a preset threshold as the first feature point.

[0073] Optionally, the identification module 220 is used to identify the initial first feature point of the medical image group according to the scale-invariant feature transformation algorithm; determine the constraint function according to the description information of the medical image group; optimize each initial first feature point according to the particle swarm optimization algorithm with the goal of maximizing the difference of the eigenvalues ​​to obtain the first feature point.

[0074] Optionally, the multiple registration models include a first deep learning model and a second deep learning model; a selection module 240 is used to use the first deep learning model as the target registration model for the first feature point when the complexity of the area where the first feature point is located is greater than a preset complexity; when the complexity of the area where the first feature point is located is not greater than the preset complexity, use the second deep learning model as the target registration model for the first feature point.

[0075] The artificial intelligence-based medical image registration method and device provided in this embodiment can be used to execute the above method embodiments. The implementation principles and technical effects are similar and will not be described in detail in this embodiment.

[0076] Figure 3 Schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an electronic device 3 provided by one embodiment of the present invention includes: a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30. When the processor 30 executes the computer program 32, the steps in the above-mentioned medical image registration method based on artificial intelligence are implemented, for example Figure 2 Alternatively, when the processor 30 executes the computer program 32, the functions of each module / unit in the above-mentioned system embodiments are implemented, for example Figure 3 The functions of each module are shown.

[0077] Exemplarily, the computer program 32 may be divided into one or more modules / units, one or more modules / units are stored in the memory 31, and are executed by the processor 30 to implement the present invention. One or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program 32 in the electronic device 3.

[0078] The electronic device 3 may be a terminal or a server. The terminal may be a mobile phone, MCU, ECU, industrial computer, etc., which are not limited here. The server may be a physical server, a cloud server, etc., which are not limited here. The electronic device 3 may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will understand that Figure 3It is only an example of the electronic device 3 and does not constitute a limitation of the electronic device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0079] The processor 30 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0080] The memory 31 may be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 31 may also be an external storage device of the electronic device 3, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 3. Further, the memory 31 may also include both an internal storage unit and an external storage device of the electronic device 3. The memory 31 is used to store computer programs and other programs and data required by the electronic device. The memory 31 may also be used to temporarily store data that has been output or is to be output.

[0081] An embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in the above-mentioned embodiment of the medical image registration method based on artificial intelligence are implemented.

[0082] The computer readable storage medium stores a computer program 32, which includes program instructions. When the program instructions are executed by the processor 30, all or part of the processes in the above-mentioned embodiment method are implemented. The computer program 32 can also be used to instruct related hardware to complete. The computer program 32 can be stored in a computer readable storage medium. When the computer program 32 is executed by the processor 30, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program 32 includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium can include: any entity or device that can carry computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium.

[0083] The computer-readable storage medium may be an internal storage unit of the electronic device of any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0084] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0085] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0086] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0087] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0088] In the embodiments provided by the present invention, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0089] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0090] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0091] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. Computer-readable media can include: any entity or device that can carry computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0092] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A medical image registration method based on artificial intelligence, characterized in that: include: Acquire a medical image group, wherein each medical image group includes a floating image and a fixed image; According to the description information of the medical image group and the feature point extraction algorithm, a first feature point of the medical image group is identified; wherein the first feature point is a pixel point in the fixed image whose difference in feature value with surrounding pixel points is greater than a preset difference; Calculating the complexity of the first area where each first feature point is located; Selecting a target registration model corresponding to each first feature point from a plurality of pre-established registration models according to the complexity; Registering the medical image group according to the first feature point and the target registration model corresponding to each first feature point to obtain a registered image; When the eigenvalue is grayscale, the complexity of the first region is: Among them, η is the complexity, a i is the grayscale of the i-th pixel, a j is the grayscale of the j pixels adjacent to the ith pixel, n is the total number of pixels; all a i Subtract a j The sum of the absolute values ​​of and then divided by the total number n! / 2 is the complexity of the first region; When the eigenvalues ​​include grayscale, curvature, optical flow of the optical flow field, and displacement value of the displacement deformation field, the complexity of the first region is: Among them, η is the complexity, a i is the grayscale of the i-th pixel, a j is the grayscale of the j pixels adjacent to the ith pixel, b i is the curvature of the i-th pixel, b j is the curvature of the j pixels adjacent to the ith pixel, c i is the optical flow of the ith pixel, c j is the optical flow of the j pixels adjacent to the ith pixel, d i is the displacement value of the i-th pixel, d j is the displacement value of the jth pixel adjacent to the ith pixel, n is the total number of pixels, k1, k2, k3, k4 are the preset coefficients corresponding to each type of eigenvalue, k1+k2+k3+k4=1.

2. The medical image registration method based on artificial intelligence according to claim 1, characterized in that: The registering the medical image group according to the first feature point and the target registration model corresponding to each first feature point to obtain a registered image includes: Calculate, according to the target registration model corresponding to each first feature point, a mapping function corresponding to the first region where each first feature point is located in the floating image; The floating image is registered with the fixed image according to the mapping function to obtain the registered image.

3. The medical image registration method based on artificial intelligence according to claim 1, characterized in that: According to the description information of the medical image group and the feature point extraction algorithm, the first feature point of the medical image group is identified: According to a scale-invariant feature conversion algorithm, an initial first feature point of the medical image group is identified; Calculating the evaluation value of each initial first feature point according to the description information of the medical image group; An initial first feature point whose evaluation value is greater than a preset threshold is used as the first feature point.

4. The medical image registration method based on artificial intelligence according to claim 1, characterized in that: According to the description information of the medical image group and the feature point extraction algorithm, identifying the first feature point of the medical image group includes: According to a scale-invariant feature conversion algorithm, an initial first feature point of the medical image group is identified; Determine the constraint function according to the description information of the medical image group; With the maximum difference of the eigenvalues ​​as the goal, each initial first feature point is optimized according to the particle swarm optimization algorithm to obtain the first feature point.

5. The medical image registration method based on artificial intelligence according to any one of claims 1 to 4, characterized in that: The multiple registration models include a first deep learning model and a second deep learning model; selecting a target registration model corresponding to each first feature point from the pre-established multiple registration models according to the complexity, including: When the complexity of the area where the first feature point is located is greater than the preset complexity, using the first deep learning model as the target registration model of the first feature point; When the complexity of the area where the first feature point is located is not greater than the preset complexity, the second deep learning model is used as the target registration model of the first feature point.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the artificial intelligence-based medical image registration method as described in any one of claims 1 to 5 are implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the artificial intelligence-based medical image registration method as described in any one of claims 1 to 5 above are implemented.

Images