Rat brain magnetic resonance imaging guided focused ultrasound targeted localization method and system and application

Abstract

The invention relates to the crossing field of biomedicine and image processing technology, and discloses a focused ultrasound mouse brain tissue targeted positioning method and positioning system based on magnetic resonance guidance and application, the positioning system comprises a data layer, a model layer, an acquisition layer, an image layer and an algorithm layer, the data layer is used for processing a total data set J, the model layer is used for calculating and obtaining a brain tissue mark point prediction position, the acquisition layer is used for obtaining data for calculating a mark point prediction model M0, and the image layer is used for carrying out space correction on a current brain tissue magnetic resonance image A and aligning a corrected brain tissue magnetic resonance image A0 with an image template B. And the algorithm layer provides an algorithm basis for correcting the difference between brain tissue individuals and carrying out local coordinate adjustment on the predicted position of the mark point. According to the invention, magnetic resonance guidance is used for focused ultrasound mediation related operation, and target brain tissues can be accurately targeted and positioned.

Description

Technical Field

[0001] The present invention relates to the intersection of biomedicine and image processing technology, and in particular to a method, system and application of mouse brain magnetic resonance imaging-guided focused ultrasound targeting positioning. Background Art

[0002] Focused Ultrasound (FUS) is a non-invasive technology that can produce mechanical or thermal effects in the target area by focusing ultrasound on a specific target, and is used for tissue ablation, blood-brain barrier (BBB) ​​opening, drug delivery, etc. In recent years, the application of focused ultrasound technology in brain science research and neuroscience research has received widespread attention.

[0003] In the study of mouse brain science, the application of focused ultrasound faces many challenges. The mouse brain tissue structure is complex and the mouse model is small in size. Achieving high-precision targeted positioning of mouse brain tissue during focused ultrasound has become a key technical problem. Due to the small size and complex anatomical structure of the mouse brain, the focused ultrasound system is required to have extremely high spatial resolution and positioning accuracy. At the same time, the acoustic properties of the mouse head bones will cause attenuation and deviation of ultrasonic wave propagation, affecting the effect of focused ultrasound. However, existing research has mostly focused on large animals or human applications, and research on mouse brain models is still in the exploratory stage, especially in terms of positioning accuracy, system integration and individual differences, which still need to be optimized. Summary of the invention

[0004] In view of the above-mentioned deficiencies or defects in the prior art, the present invention provides a brain tissue targeted positioning method, positioning system and application based on magnetic resonance image guidance, which are used for accurately positioning brain tissue during focused ultrasound.

[0005] In order to achieve the above object, the present invention provides a brain tissue targeting method, comprising the following steps:

[0006] Obtain a mouse brain tissue magnetic resonance image dataset U and a labeled dataset V of brain tissue landmarks on the magnetic resonance image, and establish a corresponding mapping relationship between the magnetic resonance image dataset U and the labeled dataset V;

[0007] Construct a total data set J, the elements of which include magnetic resonance imaging data U i , the labeled data V on the magnetic resonance image i And magnetic resonance imaging data U i With the annotation data V on the magnetic resonance image i The corresponding mapping relationship between them;

[0008] Preprocess the total data set J and divide the total data set J into a training sample set J1 and a test sample set J2;

[0009] Constructing landmark prediction model M;

[0010] The landmark prediction model M is trained using the training sample set J1 and the evaluation test is performed using the test sample set J2 to obtain the trained landmark prediction model M0;

[0011] Acquire the current brain tissue magnetic resonance image A, generate target image data I, acquire the coordinates (x, y, z) of the brain tissue landmark in the current brain tissue magnetic resonance image A, and obtain the predicted position P of the brain tissue landmark in the current brain tissue magnetic resonance image A based on the trained landmark prediction model M0, the target image data I and the coordinates (x, y, z) of the landmark in the current brain tissue magnetic resonance image A;

[0012] Based on the acquired predicted position P of the brain tissue landmark, spatially correct the current brain tissue magnetic resonance target image A to obtain a corrected brain tissue magnetic resonance image A0, and align the corrected brain tissue magnetic resonance image A0 with the brain tissue magnetic resonance image template B;

[0013] The local coordinates are adjusted according to the predicted position P of the landmark point, and the transformation algorithm is used to correct the individual differences of brain tissues, so as to locate the landmark point position TP of the brain tissue in the current brain tissue magnetic resonance target image A.

[0014] In some embodiments, the step of preprocessing the total data set J includes: i Preprocessing is performed, and the preprocessing method includes size adjustment processing, normalization processing, and noise reduction and enhancement processing.

[0015] In some embodiments, the step of constructing the landmark point prediction model M includes: constructing a U-Net model, a ResNet model, and setting an optimizer and a loss function;

[0016] Among them: building a U-Net model includes defining model blocks; building a ResNet model includes defining basic blocks and bottleneck blocks, building layers, and initializing model weights.

[0017] In some embodiments, the step of training the landmark prediction model M using the training sample set J1 includes:

[0018] The landmark prediction model M is used to extract features from the training sample set J1, where the U-Net model extracts local and global features, and the ResNet model extracts deep features;

[0019] The features extracted by the U-Net model and the ResNet model are fused, and the local features extracted by the U-Net and the deep features extracted by the ResNet are used to jointly learn and predict the location of the landmark points;

[0020] The optimizer and loss function are used to calculate the results and update the parameters of the landmark prediction model M.

[0021] In some embodiments, the step of predicting the predicted position P of the brain tissue landmark in the current brain tissue magnetic resonance image A comprises:

[0022] Determine the type of the landmark point to be selected, and obtain the coordinates (x, y, z) of the landmark point to be selected in the current brain tissue magnetic resonance image A;

[0023] Input the target image data I of the current brain tissue magnetic resonance image A and the coordinates (x, y, z) of the proposed landmark point in the current brain tissue magnetic resonance image A into the trained landmark point prediction model M0;

[0024] The current brain tissue magnetic resonance image A is processed by a convolutional neural network to obtain a vector or a feature map;

[0025] The Softmax function is used to convert each element of the vector or feature map into a probability value to obtain the heat map H i (x, y, z), heat map H i The expression for (x, y, z) is:

[0026] H i (x, y, z) = σ(f(I));

[0027] Where: the heat map Hi(x, y, z) represents the probability of the landmark at each position in the image space; f(I) is the vector or feature map output by the current brain tissue magnetic resonance image A after being processed by the convolutional neural network; σ(f(I)) represents the Softmax function;

[0028] In the heat map H i The coordinate with the maximum probability value in the global coordinates of (x, y, z) is selected as the predicted position P of the landmark point.

[0029] In some embodiments, the step of performing spatial correction on the current brain tissue magnetic resonance image A comprises:

[0030] Calculate the similarity measure between the current brain tissue magnetic resonance image A and the brain tissue magnetic resonance image template B through a similarity function;

[0031] Optimize transformation parameters by minimizing similarity measure;

[0032] The similarity evaluation index is used to evaluate the accuracy of image registration to find the best matching position of the brain tissue magnetic resonance image template B in the current brain tissue magnetic resonance image A to achieve image alignment.

[0033] In some embodiments, the similarity function expression is:

[0034] S(x, y, z) = ∑ x′,v，,z， ω(x′,y′,z′)·|T(x′,y′,z′)-I(x+x′,y+y′,z+z′)|;

[0035] Wherein, ω(x′, y′, z′) is the weight within the template area, which is used to calculate the similarity measure between the current brain tissue magnetic resonance image A and the brain tissue magnetic resonance image template B at the coordinate point (x, y, z); T(x′, y′, z′) represents the pixel value or voxel value of the brain tissue magnetic resonance image template B at the position (x′, y′, z′); I(x+x′, y+y′, z+z′) represents the pixel value or voxel value of the current brain tissue magnetic resonance image A at the position (x+x′, y+y′, z+z′);

[0036] Similarity evaluation indicators include:

[0037] The normalized correlation coefficient is expressed as:

[0038]

[0039] Wherein, T(x, y) represents the pixel value of the brain tissue magnetic resonance image template B at the position (x, y); I(x, y) represents the pixel value of the current brain tissue magnetic resonance image A at the position (x, y); represents the average pixel value of brain tissue magnetic resonance image template B; Represents the average pixel value of the current brain tissue magnetic resonance image A.

[0040] In some embodiments, the transformation algorithm comprises a rigid transformation algorithm and / or a non-rigid transformation algorithm.

[0041] The present invention also provides a brain tissue targeting and positioning system based on the above-mentioned brain tissue targeting and positioning method.

[0042] The present invention also provides an application of the above-mentioned brain tissue targeting method or the above-mentioned brain tissue targeting system in hippocampus targeting.

[0043] A brain tissue targeting positioning method, positioning system and application using the above technical solution of the present invention have the following effects: through magnetic resonance guidance for focused ultrasound-mediated related operations, the target brain tissue can be accurately targeted and positioned.

[0044] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of a process of a brain tissue targeting method according to an embodiment of the present invention;

[0046] Figure 2 It is a structural block diagram of a brain tissue targeting and positioning system according to one embodiment of the present invention. DETAILED DESCRIPTION

[0047] The specific embodiments of the present invention are described in detail below. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0048] In the present invention, unless otherwise stated, the directional words such as "upper" and "lower" usually refer to the orientation in the assembled state. "Inside" and "outside" refer to the inside and outside relative to the outline of each component itself.

[0049] The first aspect of the present invention provides a method for targeting and localizing a mouse brain under the guidance of magnetic resonance imaging of the focused ultrasound, specifically a method for targeting and localizing a mouse brain tissue. Figure 1 As shown, the positioning method includes the following steps:

[0050] The first step is to obtain a mouse brain tissue magnetic resonance image dataset U and a labeled dataset V of brain tissue landmarks on the magnetic resonance images, and establish a corresponding mapping relationship between the magnetic resonance image dataset U and the labeled dataset V. Generally, when obtaining brain tissue magnetic resonance image data, several brain tissue magnetic resonance images are collected to establish the dataset U, that is, the dataset U can be expressed as [U1, U2, ..., U n ]. Any element in the dataset U is a brain tissue magnetic resonance image data, and n represents the number of brain tissue magnetic resonance image data in the dataset U. The annotated dataset V is a set of annotated position coordinates of brain tissue landmarks on different magnetic resonance images. That is, the annotated dataset V can be expressed as [V1, V2, ..., V n ]. Any element in the annotated data set V is the coordinate data of a brain tissue landmark point on the brain tissue magnetic resonance image, and n represents the number of brain tissue landmark point coordinate data in the annotated data set. It should be noted that the number of elements in the data set U is equal to the number of elements in the annotated data set V. In addition, a corresponding mapping relationship is established between the magnetic resonance imaging data set U and the annotated data set V. According to the source principle of the annotated data, it can be known that different annotated data in the annotated data set V come from different brain tissue magnetic resonance images. Specifically, the corresponding mapping relationship between the data set U and the annotated data set V is such that: U1 corresponds to V1, U2 corresponds to V 2、 … 、 U n Corresponding to V n That is to say: the labeled data V n is from the image data U nThe coordinate data of the brain tissue landmark points obtained on the corresponding brain tissue magnetic resonance image. The brain tissue landmark points can be selected but not limited to the hippocampus, olfactory bulb, corpus callosum, third ventricle and thalamus. The specific landmark point type can be selected according to the actual needs of use. Generally, the number of landmark points is one.

[0051] The second step is to construct a total data set J. The elements of the total data set J include magnetic resonance imaging data U i , the labeled data V on the magnetic resonance image i And magnetic resonance imaging data U i With the annotation data V on the magnetic resonance image i The corresponding mapping relationship between them. That is, the total data set J can be expressed as:

[0052]

[0053] In summary, each element of the total data set J is a matrix, and each matrix has three items. Specifically, the nth matrix includes: image data U n , label data V n and U n With V n The corresponding mapping relationship

[0054] The third step is to preprocess the total data set J and divide the total data set J into a training sample set J1 and a test sample set J2. The preprocessing of the total data set J is mainly to adapt to the input requirements of the landmark prediction model M in the subsequent process. In some embodiments of the present invention, the processing of the total data set J is mainly to process the magnetic resonance image data U in the total data set J. i Preprocessing is performed, and the preprocessing methods include but are not limited to resizing, normalization, and noise reduction and enhancement using Gaussian filtering. The division of the training sample set J1 and the test sample set J2 is based on machine deep learning to train the landmark prediction model M. Specifically, the training sample set J1 can be expressed as:

[0055]

[0056] The test sample set J2 can be expressed as:

[0057]

[0058] The sum of the number of elements in the training sample set J1 and the number of elements in the test sample set J2 is equal to the number of elements in the total data set J. That is, c+d=n. In deep machine learning, the number of training samples is generally much larger than the number of test samples, that is, c is much larger than d.

[0059] The fourth step is to construct a landmark prediction model M. It should be noted that the construction of the landmark prediction model M and the above-mentioned data acquisition process can be carried out simultaneously or successively. That is to say, there is no strict restriction on the order of the fourth step and the first step, the second step and the third step. Preferably, the construction of the landmark prediction model M and the processing and division of the total data set J are completed simultaneously, so that the time to start training the landmark prediction model M can be determined to ensure the efficiency of the entire method process. In some embodiments of the present invention, the step of constructing the landmark prediction model M includes: constructing a U-Net model, a ResNet model, and setting an optimizer and a loss function. Among them: constructing a U-Net model includes defining a model block. Constructing a ResNet model includes defining basic blocks and bottleneck blocks, constructing layers, and initializing model weights.

[0060] In the fifth step, after processing and dividing the total data set J and constructing the landmark prediction model M, the landmark prediction model M is trained using the training sample set J1 and the test sample set J2 is used for evaluation and testing to obtain the trained landmark prediction model M0. In some embodiments of the present invention, the step of training the landmark prediction model M using the training sample set J1 includes: extracting features from the training sample set J1 through the landmark prediction model M. The U-Net model extracts local and global features, and the ResNet model extracts deep features. The features extracted by the U-Net model and the ResNet model are fused, and the local features extracted by the U-Net and the deep features extracted by the ResNet are used to jointly learn and predict the landmark position. Then, the parameters of the landmark prediction model M are calculated and updated according to the optimizer and the loss function. For the test sample set J2, it is used to evaluate the performance of the landmark prediction model M and save the model parameters with the best performance during the test. Finally, the parameters used by the trained landmark prediction model M0 are the model parameters saved after the test and evaluation of the test sample set J2.

[0061] Step 6, obtain the current brain tissue magnetic resonance image A and generate the target image data I. Obtain the coordinates (x, y, z) of the brain tissue landmark points in the current brain tissue magnetic resonance image A. Based on the trained landmark prediction model M0, the target image data I and the coordinates (x, y, z) of the landmark points in the current brain tissue magnetic resonance image A, obtain the predicted position P of the brain tissue landmark points in the current brain tissue magnetic resonance image A. Specifically, in this step: first determine the type of landmark points to be selected, and obtain the coordinates (x, y, z) of the landmark points to be selected in the current brain tissue magnetic resonance image A. Secondly, input the target image data I of the current brain tissue magnetic resonance image A and the coordinates (x, y, z) of the landmark points to be selected in the current brain tissue magnetic resonance image A into the trained landmark prediction model M0. Then, the current brain tissue magnetic resonance image A is processed by a convolutional neural network to obtain a vector or feature map. Finally, each element of the vector or feature map is converted into a probability value by the Softmax function to obtain a heat map H. i (x, y, z), heat map H i The expression for (x, y, z) is:

[0062] H i (x, y, z) = σ(f(I));

[0063] Among them: Heat map H i (x, y, z) represents the probability of the landmark at each position in the image space; f(I) is the vector or feature map output by the current brain tissue magnetic resonance image A after being processed by the convolutional neural network; σ(f(I)) represents the Softmax function. i The coordinate with the maximum probability value in the global coordinates of (x, y, z) is selected as the predicted position P of the landmark point.

[0064] Step 7: Based on the predicted position P of the acquired brain tissue landmark, the current brain tissue magnetic resonance target image A is spatially corrected to obtain the corrected brain tissue magnetic resonance image A0. The corrected brain tissue magnetic resonance image A0 is aligned with the brain tissue magnetic resonance image template B. The local coordinates are adjusted according to the predicted position P of the landmark, and the transformation algorithm is used to correct the individual differences of the brain tissue to locate the landmark position TP of the brain tissue landmark in the current brain tissue magnetic resonance target image A.

[0065] In some embodiments of the present invention, a similarity function is used to calculate a similarity measure between the current brain tissue magnetic resonance image A and the brain tissue magnetic resonance image template B. Then, the transformation parameters are optimized by minimizing the similarity measure. The accuracy of image registration is evaluated using a similarity evaluation index to find the best matching position of the brain tissue magnetic resonance image template B in the current brain tissue magnetic resonance image A to achieve image alignment.

[0066] In some embodiments of the present invention, the similarity function expression is:

[0067] S(x, y, z) = ∑ x′,v′,z′ ω(x′,y′,z′)·|T(x′,y′,z′)-I(x+x′,y+y′,z+z′)|.

[0068] Among them, ω(x′, y′, z′) is the weight in the template area, which is used to calculate the similarity measure between the current brain tissue magnetic resonance image A and the brain tissue magnetic resonance image template B at the coordinate point (x, y, z). T(x′, y′, z′) represents the pixel value or voxel value of the brain tissue magnetic resonance image template B at the position (x′, y′, z′). I(x+x′, y+y′, z+z′) represents the pixel value or voxel value of the current brain tissue magnetic resonance image A at the position (x+x′, y+y′, z+z′).

[0069] Similarity evaluation indicators include: normalized mutual correlation coefficient, which is expressed as:

[0070]

[0071] Wherein, T(x, y) represents the pixel value of the brain tissue magnetic resonance image template B at the position (x, y). I(x, y) represents the pixel value of the current brain tissue magnetic resonance image A at the position (x, y). Represents the average pixel value of brain tissue magnetic resonance image template B. Represents the average pixel value of the current brain tissue magnetic resonance image A.

[0072] In some embodiments of the present invention, the transformation algorithm includes a rigid transformation algorithm and / or a non-rigid transformation algorithm.

[0073] Rigid transformation includes two operations: translation and rotation to keep the shape and size of the object unchanged. Assume that the original coordinates of a point in space are After rigid transformation, the new coordinates The mathematical expression of rigid transformation is: Q′=R·Q+T. Where R is a 3*3 rotation matrix that satisfies R T R = I, ensuring rigidity. For example, the rotation matrix for counterclockwise rotation around the x-axis by an angle θ is:

[0074] Similar matrices can define the rotation matrices of the v-axis and z-axis respectively:

[0075]

[0076] T is the translation vector, which indicates the translation in three directions. The translation matrix of T is:

[0077] Non-rigid transformation is used to handle more complex deformations. It allows local nonlinear deformation on the basis of rigid transformation. The mathematical expression of non-rigid transformation is: Q′=R·Q+T+D(Q). Among them, R and T are the same as rigid transformation. D(Q) refers to: the local deformation field describing point Q, usually expressed by polynomial or basis function expansion, such as bicubic interpolation or B-spline function.

[0078] The combination of rigid and non-rigid transformations can effectively complete spatial correction from global alignment to local refinement. Rigid transformation quickly aligns the image position and direction, while non-rigid transformation achieves high-precision registration through bending deformation models. Through the optimization process of landmark points and template registration of brain tissue magnetic resonance images, a corrected image consistent with the standard coordinate system can be generated.

[0079] Magnetic Resonance Imaging (MRI) is a high-resolution imaging technology that can provide anatomical and functional information of brain tissue, providing support for the targeted positioning of focused ultrasound. Magnetic Resonance-guided Focused Ultrasound (MRgFUS) combines the high resolution of MRI with the high precision of focused ultrasound, enabling real-time imaging, target positioning, and monitoring of brain tissue.

[0080] The second aspect of the present invention provides a brain tissue targeting system based on the above-mentioned brain tissue targeting method. Figure 2As shown, the positioning system includes a data layer, a model layer, an acquisition layer, an image layer and an algorithm layer. The data layer is used to acquire, integrate, store, process and divide the total data set J. The total data set J can be updated regularly or in real time according to the use requirements, and the update content includes but is not limited to the number of elements of the total data set J, the source of brain tissue, the type of brain tissue landmarks, and the training sample set J1 and the test sample set J2. The model layer is used to construct and store the landmark prediction model M and the landmark prediction model M0 after running training, and the landmark prediction model M0 is used to calculate the data collected in the acquisition layer to obtain the predicted position of the brain tissue landmark. In addition, the model layer adopts an editable type, which is convenient for timely updating and optimizing the model parameters of the landmark prediction model M. The acquisition layer is used to obtain the data used for the calculation of the landmark prediction model M0, specifically including the current brain tissue magnetic resonance image A, generating target image data I and obtaining the coordinates of the brain tissue landmarks in the current brain tissue magnetic resonance image A. In some embodiments of the present invention, the acquisition layer is associated with the magnetic resonance imaging device or other medical devices that can be used to obtain brain tissue magnetic resonance images to establish image data transmission. The image layer is used to perform spatial correction on the current brain tissue magnetic resonance image A and to align the corrected brain tissue magnetic resonance image A0 with the image template B. The algorithm layer provides an algorithm basis for correcting individual brain tissues and adjusting the local coordinates of the predicted positions of the landmarks.

[0081] The third aspect of the present invention provides an application of the above-mentioned brain tissue targeting method or the above-mentioned mouse brain tissue targeting system in hippocampus targeting. Specifically, the above-mentioned method or system can be applied to brain tissue targeting during magnetic resonance-guided focused ultrasound.

[0082] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0083] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0084] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A brain tissue targeting method, characterized in that: The steps include: Obtain a mouse brain tissue magnetic resonance image dataset U and a labeled dataset V of brain tissue landmarks on the magnetic resonance image, and establish a corresponding mapping relationship between the magnetic resonance image dataset U and the labeled dataset V; Construct a total data set J, the elements of which include magnetic resonance imaging data U i , the labeled data V on the magnetic resonance image i And magnetic resonance imaging data U i With the annotation data V on the magnetic resonance image i The corresponding mapping relationship between them; Preprocess the total data set J and divide the total data set J into a training sample set J1 and a test sample set J2; Constructing landmark prediction model M; The landmark prediction model M is trained using the training sample set J1 and the evaluation test is performed using the test sample set J2 to obtain the trained landmark prediction model M0; Acquire the current brain tissue magnetic resonance image A, generate target image data I, acquire the coordinates (x, y, z) of the brain tissue landmark in the current brain tissue magnetic resonance image A, and obtain the predicted position P of the brain tissue landmark in the current brain tissue magnetic resonance image A based on the trained landmark prediction model M0, the target image data I and the coordinates (x, y, z) of the landmark in the current brain tissue magnetic resonance image A; Based on the acquired predicted position P of the brain tissue landmark, spatially correct the current brain tissue magnetic resonance target image A to obtain a corrected brain tissue magnetic resonance image A0, and align the corrected brain tissue magnetic resonance image A0 with the brain tissue magnetic resonance image template B; The local coordinates are adjusted according to the predicted position P of the landmark point, and the transformation algorithm is used to correct the individual differences of brain tissues, so as to locate the landmark point position TP of the brain tissue in the current brain tissue magnetic resonance target image A.

2. The brain tissue targeting method according to claim 1, characterized in that: The step of preprocessing the total data set J includes: i Preprocessing is performed, and the preprocessing method includes size adjustment processing, normalization processing, and noise reduction and enhancement processing.

3. The brain tissue targeting method according to claim 1, characterized in that: The steps of constructing the landmark point prediction model M include: constructing a U-Net model, a ResNet model, and setting an optimizer and a loss function; Among them: building a U-Net model includes defining model blocks; building a ResNet model includes defining basic blocks and bottleneck blocks, building layers, and initializing model weights.

4. The brain tissue targeting method according to claim 3, characterized in that: The steps of training the landmark prediction model M using the training sample set J1 include: The landmark prediction model M is used to extract features from the training sample set J1, where the U-Net model extracts local and global features, and the ResNet model extracts deep features; The features extracted by the U-Net model and the ResNet model are fused, and the local features extracted by the U-Net and the deep features extracted by the ResNet are used to jointly learn and predict the location of the landmark points; The optimizer and loss function are used to calculate the results and update the parameters of the landmark prediction model M.

5. The brain tissue targeting method according to claim 1, characterized in that: The step of predicting the predicted position P of the brain tissue landmark in the current brain tissue magnetic resonance image A comprises: Determine the type of the landmark point to be selected, and obtain the coordinates (x, y, z) of the landmark point to be selected in the current brain tissue magnetic resonance image A; Input the target image data I of the current brain tissue magnetic resonance image A and the coordinates (x, y, z) of the proposed landmark point in the current brain tissue magnetic resonance image A into the trained landmark point prediction model M0; The current brain tissue magnetic resonance image A is processed by a convolutional neural network to obtain a vector or a feature map; The Softmax function is used to convert each element of the vector or feature map into a probability value to obtain the heat map H i (x, y, z), heat map H i The expression for (x, y, z) is: H i (x,y,z)=σ(f(I)); Where: the heat map Hi(x, y, z) represents the probability of the landmark at each position in the image space; f(I) is the vector or feature map output by the current brain tissue magnetic resonance image A after being processed by the convolutional neural network; σ(f(I)) represents the Softmax function; In the heat map H i The coordinate with the maximum probability value in the global coordinates of (x, y, z) is selected as the predicted position P of the landmark point.

6. The brain tissue targeting method according to claim 1, characterized in that: The steps of performing spatial correction on the current brain tissue magnetic resonance image A include: Calculate the similarity measure between the current brain tissue magnetic resonance image A and the brain tissue magnetic resonance image template B through a similarity function; Optimize transformation parameters by minimizing similarity measure; The similarity evaluation index is used to evaluate the accuracy of image registration to find the best matching position of the brain tissue magnetic resonance image template B in the current brain tissue magnetic resonance image A to achieve image alignment.

7. The brain tissue targeting method according to claim 6, characterized in that: The similarity function expression is: S(x,y,z)=∑ x′,y′,z′ ω(x′,y′,z′)·|T(x′,y′,z′)-I(x+x′,y+y′,z+z′)|; Among them, ω(x′, y′, z′) is the weight in the template area, which is used to calculate the similarity measure between the current brain tissue magnetic resonance image A and the brain tissue magnetic resonance image template B at the coordinate point (x, y, z); T(x′, y′, z′) represents the pixel value or voxel value of the brain tissue magnetic resonance image template B at the position (x′, y′, z′); I(x+x′, y+y′, z+z′) represents the pixel value or voxel value of the current brain tissue magnetic resonance image A at the position (x+x′, y+y′, z+z′); Similarity evaluation indicators include: The normalized correlation coefficient is expressed as: Wherein, T(x,y) represents the pixel value of the brain tissue magnetic resonance image template B at the position (x,y); I(x,y) represents the pixel value of the current brain tissue magnetic resonance image A at the position (x,y); represents the average pixel value of brain tissue magnetic resonance image template B; Represents the average pixel value of the current brain tissue magnetic resonance image A.

8. The brain tissue targeting method according to claim 1, characterized in that: The transformation algorithm includes a rigid transformation algorithm and / or a non-rigid transformation algorithm.

9. A brain tissue targeting and positioning system based on the brain tissue targeting and positioning method according to any one of claims 1 to 8.

10. Use of the brain tissue targeting method according to any one of claims 1 to 8 or the brain tissue targeting system according to claim 9 in hippocampus targeting.

Images