CN116310479A

CN116310479A - Alzheimer's disease early identification system based on multi-center structure magnetic resonance image

Info

Publication number: CN116310479A
Application number: CN202211510960.3A
Authority: CN
Inventors: 涂丽云
Original assignee: Beijing University of Posts and Telecommunications
Current assignee: Beijing University of Posts and Telecommunications
Priority date: 2022-11-29
Filing date: 2022-11-29
Publication date: 2023-06-23
Anticipated expiration: 2042-11-29
Also published as: CN116310479B

Abstract

The application provides an early Alzheimer's disease identification system based on multi-center structure magnetic resonance image, and the system includes: the first data processing module is used for acquiring clinical index data of a user and magnetic resonance images of the multi-center structure and constructing target skeleton models of a plurality of brain structures; the first feature extraction module is used for respectively converting each target skeleton model into Euclidean features to obtain Euclidean geometric object features of each brain structure, and extracting corresponding volume features and thickness features; and the model identification module is used for generating an early Alzheimer's disease identification result by applying the early Alzheimer's disease identification model based on clinical index data, euclidean geometric object characteristics, volume characteristics and the like. The method can effectively improve the comprehensiveness and diversity of the expression characteristics of the skeleton model, can perform early identification of Alzheimer's disease aiming at the multi-modal characteristics of a plurality of brain structures, and can effectively improve the degree of automation, the effectiveness, the generalization capability of the model and the like of the identification.

Description

Alzheimer's disease early identification system based on multi-center structure magnetic resonance image

Technical Field

The application relates to the technical field of image recognition, in particular to an early Alzheimer disease recognition system based on a multi-center structure magnetic resonance image.

Background

Alzheimer's Disease (AD) is a progressive, irreversible, neurolethal degenerative disease with a 6-8 year post-diagnosis survival time, high mortality rate, long disease course, and great psychological and economic burden to the home and society. With the emergence of research hotspot problems such as big data, precise identification, early diagnosis, interpretable AI and the like in clinical science, neural networks and brain structure magnetic resonance imaging (sMRI) have become indispensable tools for brain science and neuroscience research, and particularly have extremely important clinical application values for early precise identification and prognosis evaluation of brain diseases. The current methods for researching brain structural morphology change and assisting AD diagnosis by utilizing sMRI are numerous, and the methods for extracting image histology feature fusion multi-center data set based on a third-party tool and fusion multi-mode data based on deep learning are continuously emerging.

At present, the existing Alzheimer disease identification mode is used for constructing a skeleton model of a brain structure, so that the bottleneck of modeling only the boundary (boundary) of the shape of the brain structure can not fully express the local characteristics of the interior of an object for a long time, and the effectiveness and the reliability of an identification result output by the model are poor; and, the existing method is usually only trained on a single and small sample data set, so that the learned model has the problems of over fitting, limited generalization capability, difficult reproduction of research results on a new data set and the like. Some methods, while using multiple data sets, still train on a single data set, with other data sets only serving as independent cross-validation data sets.

Disclosure of Invention

In view of this, embodiments herein provide an early identification system for alzheimer's disease based on multi-centered structural magnetic resonance imaging to obviate or ameliorate one or more of the disadvantages of the prior art.

The application provides an early Alzheimer's disease identification system based on multi-center structure magnetic resonance image, which comprises:

the first data processing module is used for acquiring clinical index data of a user and multi-center structure magnetic resonance images and constructing target skeleton models of a plurality of brain structures corresponding to the multi-center structure magnetic resonance images;

the first feature extraction module is used for respectively converting each target skeleton model into Euclidean features to obtain Euclidean geometric object features corresponding to each brain structure, and respectively extracting volume and thickness features corresponding to each brain structure based on each target skeleton model;

the model identification module is used for generating an early Alzheimer's disease identification result aiming at the user by applying a preset early Alzheimer's disease identification model based on the clinical index data and the Euclidean geometric object characteristics, volume and thickness characteristics of each brain structure.

In some embodiments of the present application, further comprising:

the second data processing module is used for acquiring clinical index data and multi-center structure magnetic resonance images corresponding to a plurality of historical users respectively and constructing target skeleton models of a plurality of brain structures corresponding to the multi-center structure magnetic resonance images;

the second feature extraction module is used for respectively converting each target skeleton model corresponding to each historical user into Euclidean features to obtain Euclidean geometric object features corresponding to each brain structure of each historical user, and respectively extracting volume and thickness features corresponding to each brain structure based on each target skeleton model;

the model training module is used for forming a data set from the clinical index data corresponding to each historical user and the Euclidean geometric object characteristics, volume and thickness characteristics of each brain structure, and collecting the data set to train a preset neural network model so as to obtain an early Alzheimer disease identification model for outputting early Alzheimer disease identification results;

the model training module is further used for sending the early identification model of the Alzheimer's disease obtained through training to the model identification module.

In some embodiments of the present application, further comprising:

the feature fusion module is used for carrying out data fusion on Euclidean geometric object features, volume and thickness features of each brain structure of the same multi-center structure magnetic resonance image and the clinical index data respectively to obtain brain structure feature vectors corresponding to each brain structure of the multi-center structure magnetic resonance image, and carrying out data fusion on each brain structure feature vector to obtain corresponding fusion feature vectors;

correspondingly, the model identification module is specifically configured to: inputting the fusion feature vector into a preset early identification model of the Alzheimer's disease, so that the early identification model of the Alzheimer's disease outputs an early identification result of the Alzheimer's disease aiming at the user.

In some embodiments of the present application, the first data processing module and the second data processing module each comprise:

the image segmentation unit is used for segmenting the acquired multi-center structure magnetic resonance image to obtain a plurality of brain structure magnetic resonance images corresponding to the multi-center structure magnetic resonance image;

the shape modeling unit is used for respectively carrying out skeleton modeling on the magnetic resonance images of the brain structures to obtain initial skeleton models corresponding to the brain structures;

And the registration unit is used for carrying out registration processing on the initial skeleton models of the brain structures corresponding to the magnetic resonance images of the same multi-center structure to obtain target skeleton models corresponding to the brain structures.

In some embodiments of the present application, the shape modeling unit includes:

the modeling module is used for respectively carrying out skeleton modeling on the magnetic resonance images of the brain structures to obtain discrete skeleton models corresponding to the brain structures, wherein the discrete skeleton models comprise discrete skeleton points and spokes pointing to the brain structure surface from the skeleton points;

and the interpolation subunit is used for respectively carrying out interpolation processing on the discrete skeleton models of each brain structure according to the similarity transformation and the non-rigid deformation of the thin plate spline to obtain initial skeleton models corresponding to each brain structure.

In some embodiments of the present application, the registration unit includes:

and the group registration subunit is used for carrying out group registration processing on the initial skeleton models of the brain structures of the same multi-center structure magnetic resonance image by adopting an entropy-based registration method to obtain target skeleton models corresponding to the brain structures of the multi-center structure magnetic resonance image.

In some embodiments of the present application, the first feature extraction module and the second feature extraction module each include:

the Euclidean transformation unit is used for carrying out main component nested sphere decomposition treatment on the unit hyper sphere corresponding to the skeleton point and the spoke in each target skeleton model to obtain Euclidean geometric object characteristics corresponding to each brain structure, wherein the Euclidean geometric object characteristics comprise: a point distribution model of skeleton points, a scaling factor, the length and the direction of spokes;

and the local feature extraction unit is used for extracting the direction histogram features of the bone points in each target skeleton model and performing size transformation processing on the direction histogram features.

In some embodiments of the present application, the early recognition model of alzheimer's disease comprises: the coding unit and the decoding unit are connected;

the coding unit comprises a multi-layer perceptron and a supervised variation self-coder which are connected, the decoding unit comprises a generation model and a classification model, so that the early Alzheimer disease identification model correspondingly outputs the early Alzheimer disease identification result according to the fusion feature vector input into the decoding unit, and simultaneously, the fusion feature vector is also reconstructed, and the early Alzheimer disease identification result is constrained based on the reconstructed fusion feature vector.

In some embodiments of the present application, the brain structure comprises: at least two of the left ventricle, right ventricle, left hippocampus, right hippocampus, left caudate nucleus and right caudate nucleus.

In some embodiments of the present application, the clinical index data comprises: at least one of user age, user gender, and mental state detection data.

In another aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the functions of the early identification system for alzheimer's disease based on multi-center structure magnetic resonance imaging when the computer program is executed.

A third aspect of the present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs the functions of the multi-central structure magnetic resonance image based early alzheimer's disease identification system.

According to the Alzheimer's disease early-stage identification system based on the multi-center structure magnetic resonance image, a first data processing module, a first feature extraction module and a model identification module which are sequentially connected are arranged, wherein the first data processing module is used for acquiring clinical index data of a user and the multi-center structure magnetic resonance image, and constructing target skeleton models of a plurality of brain structures corresponding to the multi-center structure magnetic resonance image; the first feature extraction module is used for converting each target skeleton model into Euclidean features respectively to obtain Euclidean geometric object features corresponding to each brain structure, and extracting the features such as volume, thickness and the like corresponding to each brain structure respectively based on each target skeleton model; the model identification module is used for generating an early Alzheimer's disease identification result aiming at the user by applying a preset early Alzheimer's disease identification model based on the clinical index data and the characteristics of the European geometric object characteristics, the volumes, the thicknesses and the like of each brain structure, can effectively improve the comprehensiveness and diversity of skeleton model expression characteristics, can carry out early Alzheimer's disease identification aiming at the multi-modal characteristics of a plurality of brain structures, can effectively improve the automation degree of early Alzheimer's disease identification, the effectiveness and the reliability of the identification result, and can also improve the generalization capability of the early Alzheimer's disease identification model and the like.

Additional advantages, objects, and features of the application will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and drawings.

It will be appreciated by those skilled in the art that the objects and advantages that can be achieved with the present application are not limited to the above-detailed description, and that the above and other objects that can be achieved with the present application will be more clearly understood from the following detailed description.

Drawings

The accompanying drawings are included to provide a further understanding of the application, and are incorporated in and constitute a part of this application. The components in the figures are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the application. Corresponding parts in the drawings may be exaggerated, i.e. made larger relative to other parts in an exemplary system actually manufactured according to the present application, for convenience in showing and describing some parts of the present application. In the drawings:

fig. 1 is a schematic flow chart of an early identification system for alzheimer's disease based on a multi-central structure magnetic resonance image in an embodiment of the present application.

Fig. 2 is a schematic diagram of a second structure of an early identification system for alzheimer's disease based on a multi-central structure magnetic resonance image in an embodiment of the present application.

Fig. 3 is a schematic diagram of a third structure of an early identification system for alzheimer's disease based on a multi-central structure magnetic resonance image in an embodiment of the present application.

Fig. 4 is a schematic flow chart of an implementation of an early-stage interpretable intelligent recognition system for alzheimer's disease based on a multi-center structure magnetic resonance image provided in an application example of the present application.

Fig. 5 is a technical roadmap of an early-stage interpretable intelligent recognition system for alzheimer's disease based on multi-central structure magnetic resonance images provided in an application example of the present application.

Fig. 6 (a) is a schematic diagram comparing the tail nuclei S-reps before and after registration provided in the application example of the present application.

FIG. 6 (b) is a schematic diagram showing an example of parallel processing of an object in three parts by S-rep provided in the application example of the present application.

FIG. 6 (c) shows the subdivision scheme provided in the application example of the present application

An exemplary schematic diagram with optimal computational speed and accuracy is shown.

FIG. 6 (d) shows an example of the application of the present application

Schematic of the spokes and boundaries implied by the s-rep of the ventricles of the lateral brain.

Fig. 7 is a schematic diagram showing a structural example of an early identification model of alzheimer's disease provided in an application example of the present application.

Detailed Description

For the purposes of making the objects, technical solutions and advantages of the present application more apparent, the present application will be described in further detail with reference to the embodiments and the accompanying drawings. The exemplary embodiments of the present application and their descriptions are used herein to explain the present application, but are not intended to be limiting of the present application.

It should be noted here that, in order to avoid obscuring the present application due to unnecessary details, only structures and/or processing steps closely related to the solution according to the present application are shown in the drawings, while other details not greatly related to the present application are omitted.

It should be emphasized that the term "comprises/comprising" when used herein is taken to specify the presence of stated features, elements, steps or components, but does not preclude the presence or addition of one or more other features, elements, steps or components.

It is also noted herein that the term "coupled" may refer to not only a direct connection, but also an indirect connection in which an intermediate is present, unless otherwise specified.

Hereinafter, embodiments of the present application will be described with reference to the drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

In one or more embodiments of the present application, alzheimer's disease is a progressive, irreversible, neurolethal degenerative disease. Multi-center refers to an abbreviation for one of a number of different hospitals, different facilities, as to the manner in which data is collected. Structural magnetic resonance imaging (sMRI) refers to Structural magnetic resonance imaging, a common medical imaging used clinically to examine the anatomy of the brain. Interpretable refers to the inference strategy of a machine learning process that can be understood by humans. Skeletal representation refers to Skeletal representation, a technique for modeling the shape of the surface and interior of a medical imaging subject. Shape registration refers to establishing a clear one-to-one correspondence between corresponding features in a shape model.

Alzheimer's Disease (AD) is a progressive, irreversible, neurolethal degenerative disease with a 6-8 year post-diagnosis survival time, high mortality rate, long disease course, and great psychological and economic burden to the home and society. According to the 2021 worldwide report on Alzheimer's disease, 2021 has only 25% of the worldwide diagnosis rate of AD. Mild cognitive impairment (mild cognitive impairment, MCI) is an intermediate state between normal aging and AD, with a patient population of 3877 tens of thousands, and more than 50% of MCI patients will progress to AD within 5 years, with a risk of about 10 times that of normal elderly. In China, there are currently about 1000 tens of thousands of AD patients, the number of which is the first worldwide. With the acceleration of the aging process of the population in China, the number is continuously increased, the AD patient is expected to break through 4000 ten thousand in 2050, and the economic burden is up to 126433 hundred million yuan. The huge ill population will be a significant challenge for medical health services and long-term care services. Thus, early accurate diagnosis and intervention in the MCI phase can effectively delay AD progression, tailor disease management, and plan future care.

With the emergence of research hotspot problems such as big data, precise identification, early diagnosis, interpretable AI and the like in clinical science, neural networks and brain structure magnetic resonance imaging (sMRI) have become indispensable tools for brain science and neuroscience research, and particularly have extremely important clinical application values for early precise identification and prognosis evaluation of brain diseases. According to incomplete statistics, 14412 documents have been developed for the last 20 years for the brain structure and brain function of AD patients using magnetic resonance imaging of brain structures (scmr) and functional magnetic resonance imaging (fMRI). Current methods of analyzing sMRI find that patients with AD and MCI often accompany morphological changes in brain structures, either on a voxel (voxel), slice (slide), region (region), square (patch) level, or on a three-dimensional shape model based study of morphological analysis of brain structures, at the input scale of the image, but efforts have been made to explore what image histology features can be used as effective markers for AD diagnosis.

Numerous methods are currently available for studying morphological changes in brain structures using sMRI to aid in diagnosis of AD, and Table 1 summarizes key technologies and input-output dataset information for some representative methods. The existing method is usually only trained on a single and small sample data set, so that the learned model has the problems of over fitting, limited generalization capability, difficulty in reproduction of research results on a new data set and the like. Some methods, while using multiple data sets, still train on a single data set, with other data sets only serving as independent cross-validation data sets. How to combine multiple data sets to train a unified model so as to improve the generalization capability of the model, and aiming at the problem, a method for extracting image histology characteristics based on a third party tool to fuse multiple center data sets and based on deep learning to fuse multi-mode data continuously emerge.

TABLE 1 overview of part of the study for diagnosis of AD using sMRI

Feature scale, brain structure and model core adopted in the prior art	Data set, training set number, AD vs. HC classification accuracy
		Voxel, cortical thickness, incremental learning	ADNI，491，-
Block, patch-Nets, attention mechanism	ADNI，954，92.63
		Single frame image, wavelet entropy+multi-layer perceptron	OASIS，126，92.40
Partial frame image, VGG-16+ attention mechanism	OASIS+P，196，97.76
		Square, multitasking multichannel convolutional neural network	ADNI-1，797，93.70
Square, multi-layer full convolution network (H-FCN)	ADNI-1，821，90.30
		Complete MRI image, convolutional self-encoder (CAE)	ADNI，694，86.60
51 frame full size image, resNet-50	OASIS-1，416，98.99
		Complete MRI image + black and white matter, DCAE	ADNI，479，>80.0
Voxel, cortical thickness, CNN+RNN	ADNI，830，91.33
		Voxel, cortical thickness, SVM	ADNI，255，94.99
Shape, hippocampus, SVM	ADNI，120，99.30

Where P in table 1 refers to a private data set.

Firstly, the method for extracting the image histology feature fusion multi-center data set based on the third-party tool extracts the interesting features such as the shape of the core brain region, the statistical image histogram and the like by the third-party tool such as Radiomics, freeSurfer, brain-Voyager, FSL, SPM and the like. Taking the most widely applied Freesurfer as an example, the method can complete the segmentation, registration and three-dimensional reconstruction of a high-resolution MRI image, the processing process mainly comprises the steps of removing skull, correcting B1 offset field, registering body data, carrying out gray matter segmentation, registering surface data, generating morphological characteristic data based on curved surfaces, including surface data files of cortex thickness, jacobian, sulcus, curvature, outer surface area, volume and the like, can conveniently process the brain MRI image, and can generate high-precision gray, white matter segmentation surface, gray matter and cerebrospinal fluid segmentation surface, the thickness of the cortex at any position and other surface data characteristics such as cortex outer surface area, curvature, gray matter volume and the like can be calculated according to the two surfaces, and the parameters can be mapped to the brain cortex surface obtained through a white matter expansion algorithm for visual display. In addition, the method also has the functions of inter-group difference analysis of characteristics and visualization of results. These tools generally provide simple and easy-to-handle graphical operation interfaces or standard interface instructions, and can automatically generate predefined multiple morphological features to characterize input data only by inputting own data in a fixed format. And analyzing the extracted features by combining a support vector machine, a K-means clustering method and the like to obtain the difference between the Alzheimer disease and a normal person.

However, these tools calculate the characteristics of coarse-grained voxels stacked together in the ROI of each frame image from the inputted sMRI volume representation, and it is difficult to perform fine-grained representation of complex brain structures because the calculation amount is too large; while these tools provide the ability to model shapes, they are only suitable for specific anatomies or limited topologies; these software also typically provide automatic or manual segmentation functions or accept brain structures segmented by other methods (such as U-Net, whose outstanding performance in image segmentation tasks has been verified over numerous data sets) as input, but suffer from poor overall activity, limited by the deficiencies of the technology used by the software, and adverse code reuse and technology upgrades; more importantly, these tools use only pairwise alignment (pair-wise alignment) to align each image to a predefined atlas or template, which has a relatively coarse consistency of sample shape, resulting in lesion-independent morphological changes between samples (due to equipment errors, handling sample errors, isotropic transformations of feature space, etc.) interfering with feature accuracy, which is detrimental to detection of early-stage fine morphological changes in AD. For example, a recent study has shown that the features extracted by freeprogressive have large head errors. The problem is more serious on multi-center high-dimensional heterogeneous big data, and early and accurate identification of AD is hindered.

Firstly, the method based on deep learning fusion multi-modal data automatically learns a group of features with lower dimensionality and higher discrimination capability than the original image from the image by means of various neural network models, such as a Bayesian network, a convolutional neural network, a cyclic neural network, a self-encoder, multi-core learning, recursively generating an countermeasure network and the like. In contrast to conventional neural networks, convolutional neural networks include a feature extractor consisting of a convolutional layer and a pooling layer. In the convolutional layer of a convolutional neural network, one neuron is connected with only a part of adjacent layer neurons. Without being fully connected as with a neural network. In a convolutional layer of CNN, a number of feature planes (feature maps) are usually included, each feature plane is composed of a number of neurons arranged in a rectangular shape, and the neurons of the same feature plane share weights, where the shared weights are convolution kernels. The convolution kernel is generally initialized in the form of a random decimal matrix, and learns to obtain reasonable weight values in the training process of the network. A direct benefit of sharing weights (convolution kernels) is to reduce the connections between layers of the network while reducing the risk of overfitting. Subsampling is also known as pooling (pooling), and typically takes two forms, mean subsampling (mean pooling) and maximum subsampling (max pooling). Sub-sampling can be seen as a special convolution process. The convolution and sub-sampling greatly simplify the complexity of the model and reduce the parameters of the model. The self-encoder and the recursion generation countermeasure network realize the generation capability and the prediction capability of the model by stacking and reorganizing CNNs of different layers.

However, the feature learning process of the deep neural network model is opaque, so that the deep learning feature is unexplained, and has an ambiguous meaning for clinical guidance. Interpreteability refers to the ability to interpret or express in human-understandable terms, which can be achieved either in post-hoc (interpretation of models using an interpretive method after modeling) or ad-hoc (modeling itself with interpretive capabilities). Methods for analyzing post-hoc model interpretability include deconvolution, pooling, and the like. Models of ad-hoc approaches include decision tree models, linear regression, logistic regression, multi-layer perceptrons (MLPs), bayesian instance models, etc., which are currently the most straightforward and efficient way to construct interpretable models. A recent study, using 6-layer CNNs, learns a probability map (probability maps) feature from randomly selected image blocks, and then classifies AD and normal populations using MLP. The probability map features do not adequately characterize the original image, firstly, each layer of CNN will discard some image details, and secondly, randomly selected image blocks may not cover the most informative areas, not be representative or overly generalized. Similarly, methods that utilize an interpretable feature but an unexplained classification model are not an interpretable model in the complete sense. Much research has begun focusing on interpretable models, but has so far remained one of the challenges to be addressed in the fields of artificial intelligence and clinical science.

Based on this, the embodiments of the present application solve at least one of the following key technical challenges:

1) Constructing a skeleton model of a brain structure to break through the bottleneck that only the boundary (boundary) of the shape of the brain structure can be modeled for a long time, fully expressing the local characteristics of the interior of an object, and tamping the basis for accurately identifying and longitudinally tracking early and fine morphological changes of AD;

2) Searching for a more sensitive and specific brain image marker (such as a certain local area of the tail end of a lateral ventricle), and exploring an interaction mode among a plurality of brain structures, so that the key problem that the traditional method can only analyze a single brain structure is overcome;

3) An interpretable intelligent recognition model capable of automatically processing multi-center heterogeneous sMRI big data is initially designed, the bottlenecks of small samples, overfitting, poor generalization capability and the like existing in a model which only uses a single data set for a long time are broken through, and a new way for eliminating the domain migration problem of the multi-source heterogeneous sMRI big data is provided.

The Alzheimer's disease early recognition system based on the multi-center structure magnetic resonance image provided by the application can assist in early diagnosis of AD, and the dilemma that the method of small sample and black box can not guide clinical intervention is broken through due to the interpretability of the characteristics and the model.

The following examples are provided to illustrate the invention in more detail.

The embodiment of the application provides an early identification system of Alzheimer's disease based on multi-center structure magnetic resonance image, referring to fig. 1, the early identification system of Alzheimer's disease based on multi-center structure magnetic resonance image specifically includes the following contents:

the first data processing module 1 is used for acquiring clinical index data of a user and multi-center structure magnetic resonance images, and constructing target skeleton models of a plurality of brain structures corresponding to the multi-center structure magnetic resonance images.

In the first data processing module 1, it may receive clinical index data and multi-center structure magnetic resonance images of a user to be identified early in alzheimer's disease sent by a client device, and may also obtain clinical index data and multi-center structure magnetic resonance images of a user to be identified early in alzheimer's disease directly entered locally. That is, the early Alzheimer's disease recognition system based on the multi-center structure magnetic resonance image can perform early Alzheimer's disease recognition on line, can also perform early Alzheimer's disease recognition off line, and can effectively improve the scene applicability and the application flexibility of early Alzheimer's disease recognition.

In one or more embodiments of the present application, the target skeleton model refers to an optimized skeleton model S-reps that may be formed after preprocessing such as interpolation and registration, and is used for inputting the first feature extraction module 2.

The first feature extraction module 2 is configured to convert each target skeleton model into euclidean features, obtain euclidean geometric object features corresponding to each brain structure, and extract features such as volume and thickness corresponding to each brain structure based on each target skeleton model.

In the first feature extraction module 2, the euclidean geometric object features (Euclideanized geometric object properties) may be abbreviated as GOPs. The volume feature may preferably employ a direction histogram feature of skeletal points in the skeletal model.

S-reps is a discrete topology, and the adjacent edges are curved lines with radians, which are located in a non-Euclidean space, so that the transformation of the features of the S-rep located in the non-Euclidean space into the Euclidean space (Euclidean) can only be fused with the features of other Euclidean spaces.

And the model identification module 3 is used for generating an early Alzheimer's disease identification result aiming at the user by applying a preset early Alzheimer's disease identification model based on the clinical index data, the Euclidean geometric object characteristics, the volume, the thickness and the like of each brain structure.

In the model identification module 3, the early identification model of the alzheimer's disease can be a neural network model, or other available models can be used.

In one or more embodiments of the present application, the early recognition result of the alzheimer's disease output by the early recognition model for the user refers to the early type of the alzheimer's disease to which the user belongs currently, where the early type of the alzheimer's disease at least includes: NC: normal people do not fall into the category of alzheimer's disease; MCI: mild cognitive impairment, about 50% -60% of which progress to alzheimer's disease in 3-5 years; AD: the clinical judgment is Alzheimer's disease.

From the above description, the early identification system for alzheimer's disease based on the magnetic resonance image with a multi-center structure provided by the embodiment of the application can effectively improve the comprehensiveness and diversity of the expression characteristics of the skeleton model, can perform early identification for alzheimer's disease aiming at the multi-modal characteristics of a plurality of brain structures, can effectively improve the automation degree of early identification for alzheimer's disease, the effectiveness and reliability of the identification result, and can also improve the generalization capability of the early identification model for alzheimer's disease.

In the early identification system for alzheimer's disease based on the magnetic resonance image with a multi-center structure provided in the embodiments of the present application, referring to fig. 2, the early identification system for alzheimer's disease based on the magnetic resonance image with a multi-center structure further specifically includes the following contents:

and the second data processing module 4 is used for acquiring clinical index data and multi-center structure magnetic resonance images corresponding to each of the plurality of historical users and constructing target skeleton models of a plurality of brain structures corresponding to each of the multi-center structure magnetic resonance images.

The second feature extraction module 5 is configured to convert each target skeleton model corresponding to each historical user into an euclidean feature, obtain an euclidean geometric object feature corresponding to each brain structure of each historical user, and extract the features such as volume, thickness, and the like corresponding to each brain structure based on each target skeleton model;

the model training module 6 is used for forming a dataset from the clinical index data corresponding to each historical user, the features of the Euclidean geometric object of each brain structure, the volume, the thickness and the like, and collecting the dataset to train a preset neural network model so as to obtain an early Alzheimer disease recognition model for outputting early Alzheimer disease recognition results;

Wherein the model training module 6 is further configured to send the trained early identification model of alzheimer's disease to the model identification module 3.

and the feature fusion module 7 is used for respectively carrying out data fusion on the features such as Euclidean geometric object features, volumes, thicknesses and the like of the brain structures of the magnetic resonance images with the same multi-center structure and clinical index data to obtain brain structure feature vectors corresponding to the brain structures of the magnetic resonance images with the multi-center structure, and carrying out data fusion on the brain structure feature vectors to obtain corresponding fusion feature vectors.

Correspondingly, the model identification module 3 is specifically configured to: inputting the fusion feature vector into a preset early identification model of the Alzheimer's disease, so that the early identification model of the Alzheimer's disease outputs an early identification result of the Alzheimer's disease aiming at the user.

In the early identification system for alzheimer's disease based on the multi-center magnetic resonance image provided in the embodiments of the present application, referring to fig. 3, the first data processing module 1 and the second data processing module 4 in the early identification system for alzheimer's disease based on the multi-center magnetic resonance image further specifically include the following:

the image segmentation unit 10 is configured to segment the acquired magnetic resonance images of the multi-central structure, and obtain magnetic resonance images of a plurality of brain structures corresponding to the magnetic resonance images of the multi-central structure.

And the shape modeling unit 20 is configured to perform skeleton modeling on the magnetic resonance images of the brain structures, so as to obtain initial skeleton models corresponding to the brain structures.

And the registration unit 30 is used for carrying out registration processing on the initial skeleton model of each brain structure corresponding to the magnetic resonance image of the same multi-center structure to obtain the target skeleton model corresponding to each brain structure.

In the early recognition system for alzheimer's disease based on the multi-central structure magnetic resonance image provided in the embodiment of the present application, referring to fig. 3, the shape modeling unit 20 in the early recognition system for alzheimer's disease based on the multi-central structure magnetic resonance image further specifically includes the following:

A modeling subunit 21, configured to perform skeleton modeling on the magnetic resonance images of the brain structures respectively, so as to obtain discrete skeleton models corresponding to the brain structures respectively, where the discrete skeleton models include discrete skeleton points and spokes pointing from the skeleton points to the brain structure surface respectively.

And the interpolation subunit 22 is configured to perform interpolation processing on the discrete skeleton models of each brain structure according to a similarity transformation and a non-rigid deformation of a thin-plate spline, so as to obtain initial skeleton models corresponding to each brain structure.

In particular, it has been shown that AD is a complex disease, often accompanied by morphological abnormalities of brain structures. The current method for quantitatively analyzing morphological abnormality of brain structure mostly adopts the characteristic based on third party software or shape boundary model, and can not describe the local characteristic inside the shape. The application builds a plurality of skeleton models of brain structures closely related to the course of AD. The S-reps construction process is to fit a reference model consisting of m x n discrete skeletal points (skeletons) and spokes from these points pointing to the surface of the object, one end of the spokes in a continuous model obtained by interpolating the reference model by a similarity transformation and non-rigid deformation based on thin plate splines, to the medial axis of the object, and the other end to the boundary of the object. Discrete S-reps is a simple, stable branch topology specifically designed for compact fitting to segmented objects. The pre-experiment results prove that the S-reps constructed by the method can accurately depict the global and local characteristics of the surfaces and the interiors of the sea horse, the lateral ventricle and the caudate nucleus, and lay a foundation for excavating the relationship between the structural shape change of the brain and the disease course of AD.

In the early identification system for alzheimer's disease based on multi-central structure magnetic resonance image provided in the embodiments of the present application, referring to fig. 3, the shape modeling unit 30 in the early identification system for alzheimer's disease based on multi-central structure magnetic resonance image further specifically includes the following:

the group registration subunit 31 is configured to perform a group registration process on the initial skeleton model of each brain structure of the same multi-center structure magnetic resonance image by using an entropy-based registration method, so as to obtain a target skeleton model corresponding to each brain structure of the multi-center structure magnetic resonance image.

Specifically, the process of modeling the brain structure as S-reps is mentioned, so that the disturbance of the shape change caused by the bumpy surface of the object to be described due to noise can be effectively avoided. To further eliminate noise present in the extracted skeletal features, the S-reps is first group registered.

In the early identification system for alzheimer's disease based on multi-central structure magnetic resonance image provided in the embodiments of the present application, referring to fig. 3, the first feature extraction module 2 and the second feature extraction module 4 in the early identification system for alzheimer's disease based on multi-central structure magnetic resonance image further specifically include the following:

The euclidean transformation unit 40 is configured to perform principal component nested sphere decomposition processing on unit hyper-spheres corresponding to bone points and spokes in each target skeleton model, so as to obtain euclidean geometric object features corresponding to each brain structure, where the euclidean geometric object features include: a point distribution model of skeletal points, a scaling factor, the length and direction of spokes.

A local feature extraction unit 50 for extracting a direction histogram feature of a bone point in each of the target skeleton models, and performing a size conversion process on the direction histogram feature.

S-reps is a discrete topology, and adjacent edges are curved lines with radians that lie in non-Euclidean space. Assuming that an s-rep is composed of n spokes and m skeleton points, the PDM formed by all skeleton points of the s-rep can be abstractly regarded as being positioned in a 3 m-4-dimensional unit hypersphere space after being subjected to mean value removal normalization

And a logarithmically converted scaling factor; the direction information of each spoke is abstracted and is positioned in a two-dimensional unit ball +.>

And a logarithmic conversion located in Euclidean space +.>

Is a scaling factor of (a). Thus, this discrete s-rep abstraction is located at

Manifold (manifold). For example, a 3X 13 lateral ventricle s-rep contains 39 bone spots and 72 spokes, which s-rep is located +.>

It would therefore involve converting the features of s-rep located in this non-euclidean space to euclidean space (euclidean) to be fused with features of other euclidean spaces. In addition, the shape characteristics of the direction histogram characteristics SHOT (Signature of Histogram of Orientation) near the s-rep skeleton points are extracted, and then the information is subjected to appropriate scale change and is originally positioned in the Euclidean space (hereinafter, all are abbreviated as Euclidean space)Features (spoke lengths) are spliced together to form a fused feature vector. The feature fusion process is performed on each brain structure respectively, and finally, a normalized feature vector for training a classification model, which is fused with a plurality of brain structures, is obtained.

In the early identification system of alzheimer's disease based on multi-center structure magnetic resonance image provided in the embodiment of the present application, the early identification model of alzheimer's disease includes: the coding unit and the decoding unit are connected;

the coding unit comprises a multi-layer perceptron and a supervised variation self-coder which are connected, the decoding unit comprises a generation model and a classification model, so that the early Alzheimer's disease identification model correspondingly outputs the early Alzheimer's disease identification result according to the fusion feature vector input into the decoding unit, and simultaneously, the fusion feature vector is also reconstructed, and the early Alzheimer's disease identification result is constrained based on the reconstructed fusion feature vector.

An interpretable neural network model is constructed, which receives multi-mode feature input, reconstructs input features while outputting classification results, and constrains the classification results by using reconstruction bias to ensure that the trained classification model is also an excellent generation model. Novel efficient domain adaptation (domain adaptation) or domain generalization (domain generalization) methods are explored to avoid domain migration (domain shift) problems when fusing multiple data sets. The pre-experiment result shows that the neural network model based on the multi-layer perceptron (MLP) and the supervised variational self-encoder (SVAE) can maintain the interpretability of the model and achieve satisfactory classification accuracy through KL constraint.

In the early identification system of alzheimer's disease based on multi-center structure magnetic resonance image provided in the embodiments of the present application, the brain structure includes: at least two of the left ventricle, right ventricle, left hippocampus, right hippocampus, left caudate nucleus and right caudate nucleus. Preferably, the left ventricle, right ventricle, left hippocampus, right hippocampus, left caudate nucleus and right caudate nucleus are employed.

In the early identification system for alzheimer's disease based on multi-center structure magnetic resonance image provided in the embodiment of the present application, the clinical index data includes: at least one of user age, user gender, and mental state detection data. Preferably, the user age, user gender and mental state detection data are used. Wherein, the mental state detection data can adopt a simple mental state examination scale MMSE (Mini-mental State Examination).

The part of the early identification system for the Alzheimer's disease based on the multi-center structure magnetic resonance image, which carries out the early identification of the Alzheimer's disease based on the multi-center structure magnetic resonance image, can be executed in a server, and in another practical application situation, all the operations can be completed in a client device. Specifically, the selection may be made according to the processing capability of the client device, and restrictions of the use scenario of the user. The present application is not limited in this regard. If all operations are performed in the client device, the client device may further comprise a processor for specific handling of early identification of alzheimer's disease based on multi-center structural magnetic resonance imaging.

The client device may have a communication module (i.e. a communication unit) and may be connected to a remote server in a communication manner, so as to implement data transmission with the server. The server may include a server on the side of the task scheduling center, and in other implementations may include a server of an intermediate platform, such as a server of a third party server platform having a communication link with the task scheduling center server. The server may comprise a single computer device, a server cluster formed by a plurality of servers, or a server structure of a distributed system.

Any suitable network protocol may be used for communication between the server and the client device, including those not yet developed at the filing date of this application. The network protocols may include, for example, TCP/IP protocol, UDP/IP protocol, HTTP protocol, HTTPS protocol, etc. Of course, the network protocol may also include, for example, RPC protocol (Remote Procedure Call Protocol ), REST protocol (Representational State Transfer, representational state transfer protocol), etc. used above the above-described protocol.

In order to further explain the scheme, the application also provides a specific application example of the early Alzheimer's disease identification system based on the multi-center structure magnetic resonance image, which can be also called as an early Alzheimer's disease interpretable intelligent identification system based on the multi-center structure magnetic resonance image. The key technologies and solutions involved in the application example implementation process of the present application include: fitting and interpolating a discrete skeleton model; group registration of skeleton models, skeleton feature extraction and multi-modal feature fusion, and interpretive intelligent recognition models.

Referring to fig. 4 and 5, the early-stage interpretable intelligent identification method for alzheimer's disease based on the multi-center structure magnetic resonance image, which is implemented by the early-stage interpretable intelligent identification system for alzheimer's disease based on the multi-center structure magnetic resonance image, specifically comprises the following steps:

construction of a skeletal model of a brain Structure

AD has been shown to be a complex disease, often accompanied by morphological abnormalities in brain structures. The current method for quantitatively analyzing morphological abnormality of brain structure mostly adopts the characteristic based on third party software or shape boundary model, and can not describe the local characteristic inside the shape. The application example constructs a plurality of skeleton models S-reps of brain structures closely related to the course of AD. The S-reps construction process is to fit a reference model consisting of m x n discrete skeletal points (skeletons) and spokes from these points pointing to the surface of the object, one end of the spokes in a continuous model obtained by interpolating the reference model by a similarity transformation and non-rigid deformation based on thin plate splines, to the medial axis of the object, and the other end to the boundary of the object. Discrete S-reps is a simple, stable branch topology specifically designed for compact fitting to segmented objects. The pre-experiment results prove that the S-reps constructed by the application example can accurately depict the global and local characteristics of the surfaces and the interiors of the sea horse, the lateral ventricle and the caudate nucleus, and lay a foundation for excavating the relationship between the structural shape change of the brain and the disease course of AD.

According to the application example, firstly, brain structures (such as hippocampus, lateral ventricle and caudate nucleus) related to memory are segmented from sMRI based on 3D U-Net, then a moving cube (Marching Cubes) algorithm is adopted to extract the iso-surface of the segmented binary image to obtain a point distribution model of the segmented binary image, then a discrete skeleton model is constructed based on thin-plate spline interpolation, and a continuous model is obtained through spherical linear interpolation. The more accurate segmentation method and the optimal fitting method enable the skeleton model to model a plurality of organs more widely and stably. The application example of the application is trained based on 80% of the data set 1, and the rest 20% are used for verification; after the stable model is obtained, cross-validation is performed on the

data sets

2 and 3; and finally, training all data sets together to obtain a new model, continuously verifying and adapting the new data to finally form a model with strong stability, sensitivity and generalization capability.

Specifically, the skeleton model is accessed by a group of discrete grids and spokes, the hidden continuous model is obtained by interpolation only when the model needs to be used, the required storage space is small, and the calculation speed is high. However, construction and interpolation of discrete S-reps has been a difficulty, especially for complex shapes.

The application example carries out skeleton modeling on a plurality of brain structures based on a fitting method of thin plate splines. The results of the preliminary experiments have verified the effectiveness of the method. However, we found in the laboratory that for a small number of samples with abnormal distortions or stenosis, the resulting s-rep would suffer from a non-smooth transition between adjacent polygons during interpolation (abrupt changes in normal vector direction are significant). The application example of the method adopts a new interpolation method based on the skeleton model to correct the initial model obtained based on the thin-plate spline interpolation so as to reduce the number of failed samples in the automatic construction process of the skeleton model, and reduce or even avoid manual correction.

(two) group registration

The system adopts the characteristics that: the European geometric object characteristics (Euclideanized geometric object properties, abbreviated as GOPs), the volume and the clinical index (age+sex+simple mental State examination scale MMSE (Mini-mental State Examination)).

The process of modeling the brain structure as S-reps can effectively avoid the disturbance of the surface of the object to be described on the shape change caused by the bumpy of noise. To further eliminate noise present in the extracted skeletal features, the S-reps is first group registered.

Most of the existing registration methods are only suitable for boundary models, but cannot register the volume, area, direction and other constituent elements in s-rep. The application example of the method adopts an entropy-based registration method to optimize the normalization and compactness of the spoke of the s-rep, and removes images of noise deformation on early and fine morphological changes of lesions.

The spokes and skeletal points of the S-reps of the same brain structure have initial consistency, and the group registration is adopted to further eliminate the noise difference between the shapes, so that the specificity, generalization and compactness of the S-reps shape characteristics are improved, as shown in fig. 6 (a). Each s-rep divides the shape into three separate regions for processing: upper, lower, waist, see fig. 6 (b), interpolation and registration for each part are independent, supporting multi-threading. FIG. 6 (c) shows the subdivision level

Calculating the optimal speed and precision; FIG. 6 (d) shows ∈>

Spokes and boundaries implicit by s-rep of the ventricles of the time-side.

Furthermore, because the relative position between all brain structures of the same patient is uncertain for the image of shape deformity detection, whether each brain structure of all individuals is registered individually or all brain structures are registered together when registering a plurality of brain structures is a problem that requires experimentation and specific assessment on the target data set.

Extraction of Euclidean geometric object features

And a logarithmic conversion located in Euclidean space +.>

Is a scaling factor of (a). Thus, this discrete s-rep abstraction is located at

It would therefore involve converting the features of s-rep located in this non-euclidean space to euclidean space (euclidean) to be fused with features of other euclidean spaces.

Besides, the shape characteristics of the direction histogram characteristics SHOT (Signature of Histogram of Orientation) near the s-rep skeleton points are extracted, and then the information is spliced with the characteristics (spoke length) originally positioned in Euclidean space (hereinafter, the Euclidean space) through proper scale change to form the fused characteristic vector. The feature fusion process is performed on each brain structure respectively, and finally, a normalized feature vector for training a classification model, which is fused with a plurality of brain structures, is obtained.

The technology adopted by the module further comprises: non-Euclidean spatial data (non-Euclidean data), euclidean (Euclidean), iterative Closest Point (ICP), principal Component Analysis (PCA), principal component nested spheres (principal nested sphere, PNS), composite principal component nested spheres (CPNS), multivariate gaussian distribution, minimized entropy, unconstrained optimization (NEWUOA), and the like.

Specifically, assume that

Super sphere space for waiting projection, where d ₁ ,d ₂ ,d ₃ And is more than or equal to 1.N is the number of samples. For->

Decomposing by PNS to obtain d ₂ The post-European feature (Euclideanized features) is projected. If d ₂ >N, then d of the vector ₂ The characteristics of the dimension N are all 0. For->

Decomposing the main component nested sphere PNS to obtain 2d ₃ And (5) the projected feature vector. These eigenvectors are further centered and normalized to a final length d=d ₁ +d ₂ +2d ₃ Is a feature tuple of (1). This feature tuple forms a composite Euclidean geometrical object feature (composite Euclideanized geometric object properties) matrix M of size d N _GOPs It is composed of four parts: (N-1) x N transformed PDM; a 1×n scaling factor; an n×n scaled spoke length and a 3n×n transformed spoke direction. The prior work of the applicant has studied the Euclidean process of the s-rep characteristic and connected The existing code is then based on a secondary development and optimization to enable joint processing of S-reps of multiple brain structures.

In addition, the SHOT geometrical characteristics based on s-rep bone points are extracted, a local coordinate system is established for each bone point, and the spherical neighborhood of the bone point is divided into 32 small areas along 8 directions, 2 pitching directions and 2 radial directions by default for histogram coding.

The application example of the application also relates to the simultaneous input of multiple types of characteristics (GOPs characteristics, volumes, ages, sexes and MMSEs) into the classification model, which belongs to the fusion problem of multi-mode input, and can be processed by adopting common normalization and regression.

(IV) construction of an interpretable Classification model

Exploring how to fuse the image histology characteristics of a plurality of brain structures with the characteristics of clinical observation of AD to obtain stable characteristics with universality; how to construct a reasonable learning model through the characteristics so as to detect early minor morphological changes of AD and MCI; how to open the deep learning black box makes the learning process of the model transparent and easy to understand, and makes the output result interpretable and reliable.

Specifically, the multicenter heterogeneous sMRI data adopted by the application example of the application provides a necessary condition for breaking through bottlenecks of small samples, overfitting, poor generalization capability and the like commonly faced by the current method, but also provides new challenges: how to solve the domain migration problem of large data sets, using non-EuropeanInvariance to rigid movements of the features of the genus (non-Euclidean feature) is a new approach to analyze longitudinal and multicentric large sample image databases. The S-reps constructed by the application example is located in an abstract hyper-sphere space, and skeleton features are typical non-Euclidean features and have robustness to rigid movement. In addition, the skeleton features designed by the application examples have clear geometric meaning, and provide advantages for constructing an interpretable intelligent recognition model. The most straightforward and efficient way to construct a fully interpretable smart recognition model is currently to employ interpretable features and interpretable reasoning models. As shown in fig. 7, the application example trains 3 networks simultaneously: encoder with a plurality of sensors

Decoder (p) _θ ) And classifier->

The input characteristics are encoded based on the generation model of the multi-layer perceptron and the supervised variation self-encoder, the decoder is composed of a generation model and a classification model, the loss function of the self-encoder is optimized as a main part, and the encoder is updated by combining with a second-order gradient, so that the encoder is properly deviated towards the direction of promoting the classification model while being matched with the generation model. The optimized objective function is shown in equation (1).

Wherein the parameter θ of the decoder and the parameter of the encoder

Is easy to solve, but the encoder parameters +.>

Is relatively complex because the lower bound of evidence (Evidence Lower Bound) is expected to be from +.>

A sample is taken at random, but we assume that we will add this random variable +.>

Expressed as given x and->

A transformation of the random variable e: />

Then the distribution of E will follow x and +.>

Irrespective of the fact that the first and second parts are. Thus (S)>

The gradient solution of (c) can be expressed as:

in fig. 7, x represents input data, μ and σ represent the mean and variance, respectively, of the spatial distribution of features learned by the encoder, s represents one distribution obtained by random sampling from gaussian distributions with mean μ and variance σ,

representing samples reconstructed by the decoder, +.>

Representing the probability that a sample predicted by a classifier belongs to a certain class, l _vae Representing the loss function of the variable self-encoder, l _cls Representing cross entropy loss between predictive and true labels, l=l _vae +l _cls The optimized objective function in equation (1) is represented. In summary, the early AD intelligent recognition model constructed by the application example of the application not onlyThe method adopts interpretable characteristics (being convenient for positioning, visualizing abnormal shapes corresponding to specific positions of brain structures and the like), and adopts interpretable reasoning process, thereby providing basis for guiding clinical intervention.

(fifth) evaluation method

The application example is intended to use data from a 10+ trimethyl hospital to construct a healthy brain morphological development model, and provides a reference for diagnosing brain morphological abnormalities of early AD patients. Changes in the morphological characteristics of the brain structures of individual AD patients as the course of the disease progresses, and the relationship between these patterns of changes and changes in brain cognitive ability, are constructed. Algorithms of great interest include: meta-learning, second order gradients, data enhancement, model generation, antagonism network, self-attention network (transformation), covariance shift (covariance shift), migration learning, domain adaptation, domain generalization, semi-supervised/self-supervised learning, etc.

The technical key point of the application example is as follows:

(1) Fitting and interpolating a discrete skeleton model;

(2) Performing group registration on a plurality of core brain structures in the multi-center structure magnetic resonance image based on the skeleton model;

(3) Extracting skeleton characteristics and effectively fusing multi-mode characteristics;

(4) And constructing an interpretable intelligent recognition model based on the generated model.

That is, the application example of the application firstly utilizes the skeleton model to describe global and local morphological changes of the shape boundary and the interior of the brain structure of the AD patient, and opens up a new path for intelligent and accurate identification in the early stage of AD; first joint analysis of a plurality of memory-related brain structures and exploration of their roles and cooperative relationships in the progression of AD; the problem of domain migration of the magnetic resonance image of the multi-center multi-scale brain structure is solved by using group registration and Euclidean geometric object features for the first time, and an interpretable AD early recognition model is constructed.

In summary, the application example of the application is oriented to multi-center large-scale clinical data, provides a new method (Euclidean geometric object characteristics and shape models of a plurality of brain structures are jointly analyzed), solves new technical problems (domain migration and interpretable intelligent recognition), is applied to auxiliary clinical early diagnosis of AD, opens up a new way for exploring an AD image marker and intelligent accurate recognition in the future, effectively delays the progress of AD illness for accurate intervention in the future, and helps to relieve economic and social pressures caused by AD patients in the aging process of population.

The embodiment of the application further provides an electronic device (i.e., an electronic device), for example, a central server, where the electronic device may include a processor, a memory, a receiver, and a transmitter, where the processor is configured to perform the functions of the early identification system for alzheimer's disease based on multi-central structure magnetic resonance imaging mentioned in the foregoing embodiment, and the processor and the memory may be connected by a bus or other means, for example, by a bus connection. The receiver may be connected to the processor, memory, by wire or wirelessly.

The processor may be a central processing unit (Central Processing Unit, CPU). The processor may also be any other general purpose processor, digital signal processor (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof.

The memory is used as a non-transitory computer readable storage medium and can be used for storing a non-transitory software program, a non-transitory computer executable program and a module, such as program instructions/modules corresponding to the early identification system of Alzheimer's disease based on multi-center structure magnetic resonance image in the embodiment of the application. The processor executes various functional applications and data processing of the processor by running non-transitory software programs, instructions and modules stored in the memory, i.e., the functions of the early identification system of alzheimer's disease based on multi-center structure magnetic resonance images in the above system embodiments are implemented.

The memory may include a memory program area and a memory data area, wherein the memory program area may store an operating system, at least one application program required for a function; the storage data area may store data created by the processor, etc. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory may optionally include memory located remotely from the processor, the remote memory being connectable to the processor through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The one or more modules are stored in the memory that, when executed by the processor, perform the functions of the multi-center structure magnetic resonance image based early Alzheimer's disease identification system in embodiments.

In some embodiments of the present application, the user equipment may include a processor, a memory, and a transceiver unit, where the transceiver unit may include a receiver and a transmitter, and the processor, the memory, the receiver, and the transmitter may be connected by a bus system, the memory storing computer instructions, and the processor executing the computer instructions stored in the memory to control the transceiver unit to transmit and receive signals.

As an implementation manner, the functions of the receiver and the transmitter in the present application may be considered to be implemented by a transceiver circuit or a dedicated chip for transceiver, and the processor may be considered to be implemented by a dedicated processing chip, a processing circuit or a general-purpose chip.

As another implementation manner, a manner of using a general-purpose computer may be considered to implement the server provided in the embodiments of the present application. I.e. program code for implementing the functions of the processor, the receiver and the transmitter are stored in the memory, and the general purpose processor implements the functions of the processor, the receiver and the transmitter by executing the code in the memory.

Embodiments of the present application also provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the aforementioned early identification system for alzheimer's disease based on multi-central structure magnetic resonance images. The computer readable storage medium may be a tangible storage medium such as Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disk, a removable memory disk, a CD-ROM, or any other form of storage medium known in the art.

Those of ordinary skill in the art will appreciate that the various illustrative components, systems, and systems described in connection with the embodiments disclosed herein can be implemented as hardware, software, or a combination of both. The particular implementation is hardware or software dependent on the specific application of the solution and the design constraints. Those skilled in the art may implement the described functionality using different systems for each particular application, but such implementation is not to be considered as outside the scope of this application. When implemented in hardware, it may be, for example, an electronic circuit, an Application Specific Integrated Circuit (ASIC), suitable firmware, a plug-in, a function card, or the like. When implemented in software, the elements of the present application are the programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine readable medium or transmitted over transmission media or communication links by a data signal carried in a carrier wave.

It should be clear that the present application is not limited to the particular arrangements and processes described above and illustrated in the drawings. A detailed description of known systems is omitted here for the sake of brevity. In the above embodiments, several specific steps are described and shown as examples. However, the system processes of the present application are not limited to the specific steps described and illustrated, and one skilled in the art can make various changes, modifications, and additions, or change the order between steps, after appreciating the spirit of the present application.

The features described and/or illustrated in this application for one embodiment may be used in the same way or in a similar way in one or more other embodiments and/or in combination with or instead of the features of the other embodiments.

The foregoing description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and variations may be made to the embodiment of the present application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. An early identification system for alzheimer's disease based on multi-central structure magnetic resonance imaging, comprising:

2. The early identification system for alzheimer's disease based on multi-central structure magnetic resonance imaging of claim 1, further comprising:

3. The early identification system for alzheimer's disease based on multi-central structure magnetic resonance imaging according to claim 1 or 2, further comprising:

4. The early identification system of alzheimer's disease based on multi-central structural magnetic resonance imaging of claim 2, wherein the first data processing module and the second data processing module each comprise:

5. The early identification system for alzheimer's disease based on multi-central structure magnetic resonance imaging according to claim 4, wherein said shape modeling unit comprises:

6. The early identification system for alzheimer's disease based on multi-central structure magnetic resonance imaging according to claim 4, wherein said registration unit comprises:

7. The early identification system of alzheimer's disease based on multi-central structure magnetic resonance imaging of claim 2, wherein the first feature extraction module and the second feature extraction module each comprise:

8. The early recognition system of alzheimer's disease based on multi-central structural magnetic resonance imaging of claim 3, wherein the early recognition model of alzheimer's disease comprises: the coding unit and the decoding unit are connected;

9. The early identification system for alzheimer's disease based on multi-central structural magnetic resonance imaging according to claim 1 or 2, wherein said brain structure comprises: at least two of the left ventricle, right ventricle, left hippocampus, right hippocampus, left caudate nucleus and right caudate nucleus.

10. The early identification system for alzheimer's disease based on multi-central structural magnetic resonance imaging according to claim 1 or 2, wherein said clinical index data comprises: at least one of user age, user gender, and mental state detection data.