Early cognitive impairment classification system, method and terminal

Abstract

The early cognitive dysfunction classification system, method and terminal of the present application obtain the early cognitive dysfunction classification result corresponding to the functional nuclear magnetic resonance image through the early cognitive dysfunction classification model based on the Transformer and the multi-level function connection; the present application extracts the most relevant features of the BOLD signal and the disease diagnosis based on the Transformer to obtain the classification result, greatly improving the accuracy and efficiency of the early cognitive dysfunction classification.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of early cognitive dysfunction recognition, and in particular to an early cognitive dysfunction classification system, method and terminal. BACKGROUND

[0002] With the growth of human age, the development of brain tissue myelin is damaged, nerve cells atrophy, synaptic contact and neurotransmitter decrease, which leads to a series of changes in brain structure and function, that is, the age-related decline in cognitive function. According to the rate and mode of changes in brain structure, function, metabolism and pathology, this process can be divided into physiological brain aging and pathological brain aging. After pathological brain aging, some people will suffer from AD (Alzheimer's disease). Broadly speaking, AD has the characteristics of gradual progression, which can be divided into three stages of disease development: preclinical AD (PCAD for short, at this time the patient has not shown clinical symptoms, but the pathology shows that there is abnormal deposition of Aβ protein in the brain), mild cognitive impairment (MCI for short, at this time the patient shows mild cognitive impairment), and AD dementia (at this time the early intervention time window has been missed).

[0003] At present, the stage division of the above AD and related cognitive impairment pathological process is not clear, especially when the brain aging process develops from physiological and pathological aging to AD, which has strong concealment. Research has confirmed that the neurodegenerative changes caused by AD can be decades earlier than the AD dementia stage, and only in the mild cognitive impairment (MCI) stage do they show clinical symptoms. In the last stage of preclinical AD (subjective cognitive decline stage, SCD), the proportion of patients transforming into MCI and AD dementia is as high as 54.2%, which is significantly higher than the conversion rate of 14.9% in the control group; however, SCD can be 15 years earlier than the clinical screening of cognitive impairment, which makes it extremely difficult to diagnose AD in the early stage.

[0004] How to achieve early diagnosis of AD is a difficult problem that has not been solved in the current research of cognitive impairment and AD, and is also a pain point in clinical diagnosis and treatment practice. In the early accurate diagnosis of AD, the abnormal deposition of Aβ protein is an important pathological marker; and Aβ-PET can achieve in vivo detection of Aβ protein, so it is regarded as the "gold standard" for early diagnosis of AD. As early as the PCAD stage (especially the SCD stage), although the patient has not yet appeared clinical symptoms, studies have found that Aβ abnormal deposition has already appeared in the right medial prefrontal cortex, anterior cingulate gyrus, posterior cingulate gyrus and right precuneus. Although PET imaging has the advantages of high sensitivity and high specificity in diagnosis, the cost of PET is high, the image acquisition and reading efficiency is general, and the acceptance of early patients is poor. At present, the neuropsychological test scale is widely used in clinical practice as the main examination method for cognitive function assessment and AD diagnosis. In the previous work, the project group systematically elaborated the application of the scale in AD staging diagnosis, control analysis and epidemiological research, especially in the selection of scales in the PCAD stage, which provided an important reference standard for the evaluation of early clinical cognitive function of AD. Compared with Aβ-PET, the cost of scale evaluation is significantly reduced, and it is suitable for popularization and use in early disease screening and diagnosis. However, the scale evaluation inevitably has subjective bias, and the process is time-consuming and laborious, with poor efficiency in clinical diagnosis and treatment. In addition to PET imaging, other modalities of brain imaging are also playing an increasingly important role in the clinical diagnosis of cognitive impairment. For example, the fMRI (functional magnetic resonance imaging) used in this patent can obtain the spatiotemporal characteristics of brain activity and construct a functional connectivity network. Studies have found that the brain function networks of normal people, MCI and AD patients all have small-world (Small-World) characteristics, but there are differences between the three groups, which suggests that the brain function information reflected by fMRI exists not only in local brain activity and functional connectivity, but also widely exists at a larger scale.

[0005] Deep learning models based on graph neural networks are widely used in network classification and have shown excellent performance. At the same time, there have been many GCN-based methods used in the classification of brain functional connectivity networks, such as GCN-LSTM, Multi-hierarchies Function network, etc.; but most of them only use Pearson correlation or other correlation calculation methods to establish functional connectivity networks based on fMRI BOLD (blood oxygen level dependent) signals, which ignores the hidden and disease-related features in BOLD signals. Meanwhile, studies have shown that the differences in functional connectivity networks between MCI patients and normal people exist in different levels of functional connectivity networks, and previous studies mostly use manually set fixed thresholds to set different levels of functional connectivity networks, but such thresholds cannot be optimized. SUMMARY

[0006] In view of the above-mentioned disadvantages of the prior art, the purpose of the present application is to provide an early cognitive dysfunction classification system, method and terminal for solving the above technical problems in the prior art.

[0007] To achieve the above-mentioned purpose and other related purposes, the present application provides an early cognitive dysfunction classification system, which comprises: a data acquisition module for acquiring a collected functional magnetic resonance image to be classified; and a classification module connected to the data acquisition module and configured to obtain an early cognitive dysfunction classification result corresponding to the functional magnetic resonance image according to an early cognitive dysfunction classification model based on a Transformer and a multi-level functional connection.

[0008] In an embodiment of the present application, the early cognitive dysfunction classification model training method comprises: the early cognitive dysfunction classification model comprises: a preprocessing and partitioning module configured to preprocess and partition the functional magnetic resonance image to obtain partitioned data; wherein the partitioned data comprises BOLD signals of one or more brain regions; a low-order classification module connected to the preprocessing and partitioning module and configured to construct a low-order functional connection network according to the partitioned data, construct a multi-level first brain functional connection network using a threshold learning network, and obtain features corresponding to each level to obtain a low-order classification result corresponding to each feature; a high-order classification module connected to the preprocessing and partitioning module and configured to construct a low-order functional connection network according to the partitioned data based on a Transformer, construct a multi-level second brain functional connection network using a threshold learning network, and obtain features corresponding to each level to obtain a high-order classification result corresponding to each feature; and a final classification result determination module connected to the low-order classification channel module and the high-order classification channel module and configured to obtain an early cognitive dysfunction classification result according to each low-order classification result and each high-order classification result.

[0009] In an embodiment of the present application, the low-order classification module comprises: a low-order functional connection construction submodule configured to calculate similarity data of BOLD signals corresponding to each brain region according to the partitioned data to construct a low-order functional connection network; a first threshold learning submodule connected to the low-order functional connection construction submodule and configured to obtain a plurality of thresholds and a first brain functional connection network corresponding to a plurality of levels according to an input low-order functional connection network based on a threshold learning network; and a low-order classification result acquisition submodule connected to the first threshold learning submodule and configured to extract features corresponding to each level from the first brain functional connection network of each level based on a classification network structure using a graph neural network and a fully connected layer, and obtain a low-order classification result corresponding to each feature.

[0010] In an embodiment of the present application, the high-order classification module comprises: a high-order functional connection construction submodule, configured to extract high-order disease features from the partitioned data based on a Transformer, and calculate similarity data of the corresponding high-order disease features to construct a high-order functional connection network; a second threshold learning submodule connected to the high-order functional connection construction submodule, configured to obtain multiple thresholds and obtain a multi-level second brain functional connection network according to an input high-order functional connection network based on a threshold learning network; and a high-order classification result acquisition submodule connected to the second threshold learning submodule, configured to extract features corresponding to each level according to each level of the second brain functional connection network based on a classification network structure using a graph neural network and a fully connected layer, and obtain high-order classification results corresponding to each feature.

[0011] In an embodiment of the present application, the high-order functional connection construction submodule comprises: a Transformer unit, configured to extract corresponding high-order disease features from the partitioned data; a high-order similarity calculation unit connected to the Transformer unit, configured to calculate a Pearson correlation coefficient of the high-order disease features; and a high-order functional connection network construction unit connected to the similarity calculation unit, configured to obtain a high-order functional connection network according to the Pearson correlation coefficient.

[0012] In an embodiment of the present application, the Transformer unit is configured to perform matrix mapping transformation on the partitioned data to obtain a Query matrix, a Key matrix and a Value matrix; calculate a similarity matrix based on the Query matrix and the Key matrix; obtain a self-attention matrix by performing a Softmax operation on the similarity matrix; and multiply the self-attention matrix with the Value matrix to obtain the high-order disease features.

[0013] In an embodiment of the present application, the threshold learning network comprises: one or more layers of graph neural networks, configured to input a high-order functional connection network to obtain corresponding graph neural output data, and also configured to input a low-order functional connection network to obtain corresponding graph neural output data; a fully connected layer, configured to obtain a set number of thresholds according to the graph neural output data; a threshold contraction module connected to the fully connected layer, configured to contract and screen each threshold by a hyperbolic tangent method to obtain multiple screened thresholds; and a brain functional connection network module connected to the threshold contraction module, configured to construct a corresponding multi-level first brain functional connection network according to a low-order functional connection network and the screened thresholds, and also configured to construct a corresponding multi-level second brain functional connection network according to a high-order functional connection network and the screened thresholds.

[0014] In an embodiment of the present application, the obtaining the early cognitive impairment classification result according to the low-order classification results and the high-order classification results comprises: obtaining the early cognitive impairment classification result according to the low-order classification results and the high-order classification results based on a voting mechanism.

[0015] To achieve the above object and other related objects, the present application provides an early cognitive impairment classification method, which comprises: obtaining a collected functional magnetic resonance image to be classified; and obtaining an early cognitive impairment classification result corresponding to the functional magnetic resonance image according to an early cognitive impairment classification model based on a Transformer and a multi-level functional connection.

[0016] To achieve the above object and other related objects, the present application provides an early cognitive impairment classification terminal, which comprises: one or more memories and one or more processors; the one or more memories are used for storing a computer program; and the one or more processors are connected to the memories and used for running the computer program to execute the early cognitive impairment classification method.

[0017] As described above, the present application is an early cognitive impairment classification system, method and terminal, which has the following beneficial effects: the present application obtains an early cognitive impairment classification result corresponding to the functional magnetic resonance image according to an early cognitive impairment classification model based on a Transformer and a multi-level functional connection; and the present application greatly improves the accuracy and efficiency of early cognitive impairment classification by extracting the most relevant features of BOLD signals and disease diagnosis based on a Transformer to obtain a classification result. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 FIG. 1 shows a structural schematic diagram of an early cognitive impairment classification system according to an embodiment of the present application.

[0019] Figure 2 FIG. 2 shows a structural schematic diagram of an early cognitive impairment classification model according to an embodiment of the present application.

[0020] Figure 3 FIG. 3 shows a structural schematic diagram of an early cognitive impairment classification model according to an embodiment of the present application.

[0021] Figure 4 FIG. 4 shows a structural schematic diagram of a threshold learning network according to an embodiment of the present application.

[0022] Figure 5 FIG. 5 shows a structural schematic diagram of a Transformer unit according to an embodiment of the present application.

[0023] Figure 6A network overall architecture diagram of the early cognitive impairment classification tool in an embodiment of the present application is shown.

[0024] Figure 7 A flow diagram of the early cognitive impairment classification method in an embodiment of the present application is shown.

[0025] Figure 8 A structure diagram of the early cognitive impairment classification terminal in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0026] Other advantages and effects of the present application can be easily understood by those skilled in the art from the above description of the embodiments of the present application. The present application can also be implemented or applied by other different embodiments, and the details in the specification can be modified or changed based on different views and applications without departing from the spirit of the present application. It should be noted that the embodiments below and the features in the embodiments can be combined with each other without conflict.

[0027] It should be noted that in the following description, reference is made to the accompanying drawings, which form a part hereof, and in which are shown by way of illustration several embodiments of the present application. It is to be understood that other embodiments can be utilized and that mechanical, structural, electrical, and operational changes can be made without departing from the spirit and scope of the present application. The following detailed description is not to be taken in a limiting sense, and the scope of embodiments of the present application are defined only by the claims of the issued patent. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. Spatially relative terms, such as "upper", "lower", "left", "right", "below", "below", "bottom", "top", and the like, can be used herein for ease of description to describe one element or feature's relationship to another element(s) or feature(s) as illustrated in the figures.

[0028] Throughout this specification, when it is said that a certain part is "connected" to another part, it includes not only the case of "direct connection" but also the case of "indirect connection" in which other elements are interposed therebetween. In addition, when it is said that a certain part "includes" a certain constituent element, other constituent elements are not excluded unless specifically stated to the contrary, and it means that other constituent elements can also be included.

[0029] The first, second, and third terms mentioned therein are used for the purpose of describing various parts, components, regions, layers, and / or sections, but are not limited thereto. These terms are used only to distinguish a certain part, component, region, layer, or section from another part, component, region, layer, or section. Therefore, the first part, component, region, layer, or section described below can be referred to as the second part, component, region, layer, or section within the scope of the present application.

[0030] Furthermore, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context indicates otherwise. It will be further understood that the terms "comprises", "comprising", "includes" and / or "including" when used herein, specify the presence of stated features, operations, elements, components, items, kinds and / or groups but do not preclude the presence or addition of one or more other features, operations, elements, components, items, kinds, and / or groups thereof. As used herein, the terms "or" and "and / or" are to be interpreted as inclusive, i.e., as meaning one or any combination of the items. Thus, "A, B or C" or "A, B and / or C" means any of the following: A; B; C; A and B; A and C; B and C; A, B and C. An exception to this definition will occur only when a combination of elements, functions, or operations are in some way inherently mutually exclusive.

[0031] The application provides an early cognitive dysfunction classification system, method and terminal, which obtains an early cognitive dysfunction classification result corresponding to the functional magnetic resonance image according to an early cognitive dysfunction classification model based on a Transformer and a multi-level function connection; the application extracts the most relevant features of the BOLD signal and disease diagnosis based on the Transformer to obtain a classification result, thereby greatly improving the accuracy and efficiency of early cognitive dysfunction classification.

[0032] The embodiments of the application will be described in detail below with reference to the accompanying drawings, so that those skilled in the art in the technical field to which the application pertains can easily implement the application. The application can be embodied in various different forms, and is not limited to the embodiments described herein.

[0033] As Figure 1 The structure of an early cognitive dysfunction classification system in an embodiment of the application is shown in the figure.

[0034] The system comprises:

[0035] A data acquisition module 11 is configured to acquire a collected functional magnetic resonance image to be classified.

[0036] A classification module 12 is connected to the data acquisition module 11 and configured to obtain an early cognitive dysfunction classification result corresponding to the functional magnetic resonance image according to an early cognitive dysfunction classification model based on a Transformer and a multi-level function connection.

[0037] Optionally, as Figure 2 The early cognitive dysfunction classification model comprises:

[0038] The preprocessing and partition module 21 is configured to preprocess and partition the functional magnetic resonance image to obtain partition data, wherein the partition data comprises BOLD signals of one or more brain regions.

[0039] The low-order classification module 22 is connected to the preprocessing and partition module 21 and is configured to construct a low-order functional connection network according to the partition data, construct a multi-level first brain functional connection network by using a threshold learning network, and obtain features corresponding to each level to obtain a low-order classification result corresponding to each feature.

[0040] The high-order classification module 23 is connected to the preprocessing and partition module 22 and is configured to construct a low-order functional connection network according to the partition data based on a Transformer, construct a multi-level second brain functional connection network by using a threshold learning network, and obtain features corresponding to each level to obtain a high-order classification result corresponding to each feature.

[0041] The final classification result determination module 24 is connected to the low-order classification channel module 22 and the high-order classification channel module 23 and is configured to obtain an early cognitive dysfunction classification result according to each low-order classification result and each high-order classification result.

[0042] Optionally, the preprocessing and partition module 21 is configured to perform a preprocessing operation on the functional magnetic resonance image, and then obtain BOLD signals of each brain region according to a template-defined brain region distribution based on the preprocessed image.

[0043] Optionally, as shown in Figure 3 The low-order classification channel module 22 comprises:

[0044] The low-order functional connection construction submodule 221 is configured to directly calculate similarity data of BOLD signals of each brain region according to the partition data to construct a low-order functional connection network, and preferably directly calculate Pearson correlation coefficients of BOLD signals of each brain region according to the partition data to construct a low-order functional connection network.

[0045] The first threshold learning submodule 232 is connected to the low-order functional connection construction submodule 231 and is configured to obtain a plurality of thresholds and a first brain functional connection network corresponding to a plurality of levels according to an input low-order functional connection network based on a threshold learning network.

[0046] The low-order classification result acquisition submodule 223 is connected to the first threshold learning submodule 232 and is configured to extract features corresponding to each level according to the first brain functional connection network of each level based on a classification network structure using a graph neural network and a fully connected layer, and obtain a low-order classification result corresponding to each feature.

[0047] Optionally, as shown inFigure 3 As shown, the high-order classification module 23 comprises:

[0048] The high-order functional connection construction sub-module 231 is configured to extract high-order disease features from the partition data based on a Transformer and calculate similarity data of the corresponding high-order disease features to construct a high-order functional connection network.

[0049] The second threshold learning sub-module 232 is connected to the high-order functional connection construction sub-module 231 and is configured to obtain multiple thresholds and a multi-level second brain functional connection network from the input high-order functional connection network based on a threshold learning network.

[0050] The high-order classification result acquisition sub-module 233 is connected to the second threshold learning sub-module 232 and is configured to extract features corresponding to each level from each level of the second brain functional connection network based on a classification network structure adopting a graph neural network and a fully connected layer, and obtain high-order classification results corresponding to each feature.

[0051] Optionally, the threshold learning network adopted by the first threshold learning sub-module 232 and the second threshold learning sub-module 232 has the same structure, as shown in Figure 4 As shown, the threshold learning network comprises:

[0052] One or more layers of graph neural networks GCN are configured to obtain corresponding graph neural output data from the input high-order functional connection network and are also configured to obtain corresponding graph neural output data from the input low-order functional connection network.

[0053] A fully connected layer FL is connected to the last layer of graph neural networks GCN and is configured to obtain a set number k of thresholds th1-th k ;

[0054] A threshold contraction module is connected to the fully connected layer FL and is configured to contract and screen each threshold in a hyperbolic tangent manner to obtain multiple screened thresholds; preferably, all thresholds are contracted by tanh to obtain thresholds between [-1, 1].

[0055] A brain functional connection network module is connected to the threshold contraction module and is configured to construct a corresponding multi-level first brain functional connection network from the low-order functional connection network and the screened thresholds, and is also configured to construct a corresponding multi-level second brain functional connection network from the high-order functional connection network and the screened thresholds. Each threshold corresponds to a level. Since the threshold learning network adopted by the first threshold learning sub-module 232 and the second threshold learning sub-module 232 has the same structure, the number of first brain functional connection networks is the same as that of second brain functional connection networks, i.e., the levels are the same.

[0056] Preferably, a shifter leaky ReLU function is used to construct a corresponding multi-level first brain functional connectivity network according to the low-order functional connectivity network and the screened thresholds; and a corresponding multi-level second brain functional connectivity network is constructed according to the high-order functional connectivity network and the screened thresholds.

[0057] Optionally, the low-order classification result acquisition submodule 223 and the high-order classification result acquisition submodule 233 adopt the same classification network structure, which includes a plurality of graph neural networks and a plurality of fully connected layers; preferably, a 3-layer graph neural network is used, and the plurality of different levels of functional connectivity networks obtained are input into the graph neural network, and after passing through the three-layer graph neural network, the corresponding features of the functional connectivity of each level can be obtained. Each feature can correspond to a disease classification result.

[0058] Optionally, the high-order functional connectivity construction submodule 231 includes:

[0059] a Transformer unit, configured to extract corresponding high-order disease features from the partitioned data;

[0060] a high-order similarity calculation unit connected to the Transformer unit, configured to calculate a Pearson correlation coefficient of the high-order disease features;

[0061] a high-order functional connectivity network construction unit connected to the similarity calculation unit, configured to obtain a high-order functional connectivity network related to disease diagnosis according to the Pearson correlation coefficient.

[0062] Optionally, as shown in Figure 5 the Transformer unit is configured to perform matrix mapping transformation on the partitioned data to obtain a Query matrix, a Key matrix and a Value matrix; a similarity matrix is calculated based on the Query matrix and the Key matrix; a self-attention matrix is obtained by performing a Softmax operation on the similarity matrix; and the high-order disease features are obtained by multiplying the self-attention matrix and the Value matrix.

[0063] Optionally, obtaining the early cognitive impairment classification result according to the low-order classification results and the high-order classification results includes: obtaining the early cognitive impairment classification result according to the low-order classification results and the high-order classification results based on a voting mechanism.

[0064] In order to better illustrate the above-mentioned early cognitive impairment classification system, the present application provides the following specific embodiments.

[0065] Embodiment 1: An early cognitive impairment classification tool based on Transformer and multi-level functional connectivity. As shown in FIG. 1, it is a whole architecture diagram of the network of the early cognitive impairment classification tool. Figure 6

[0066] The method for classification by using the tool includes:

[0067] First, we pre-process the fMRI, and then obtain the BOLD signals of each brain region according to the brain regions defined by the template. After obtaining the BOLD signals of each brain region, we further construct low-order functional connectivity and high-order functional connectivity according to the BOLD signals of the brain regions. On the one hand, we directly calculate the Pearson correlation coefficient of the BOLD signals of each brain region; on the other hand, we use the high-order and disease-related features extracted by the Transformer module to establish a high-order functional connectivity network. After inputting the two functional connectivity networks into the threshold learning network, we obtain the functional connectivity networks of different levels. Finally, we input the multiple functional connectivity networks of different levels obtained by us into the graph neural network, and after passing through three layers of graph neural network, we can obtain the features corresponding to each level of functional connectivity. Each feature can correspond to a disease classification result. Finally, we use a voting mechanism to obtain the final classification result.

[0068] Among them, the way of extracting effective features of brain BOLD signals by using Transformer includes: first, we obtain Query, Key, and Value three matrices through matrix mapping transformation. Then, we use Query and Key two matrices to calculate a similarity matrix, and then obtain a self-attention matrix through Softmax operation. After obtaining the attention matrix, we multiply it with the Value matrix to obtain the features we finally want to obtain. By calculating the Pearson similarity of the features, we finally obtain the high-order functional connectivity network related to disease diagnosis.

[0069] The way of obtaining functional connectivity networks of different levels by using threshold learning network includes: first, we input the functional connectivity as input into the graph neural network. After passing through multiple layers of graph neural network, we obtain multiple threshold values through a fully connected layer. Finally, all threshold values are contracted to [-1, 1] through tanh. Finally, multiple brain functional connections of different levels are obtained by threshold operation on the input FCN by shifter leakyReLU.

[0070] ​In this embodiment, we refer to the method of Ashish Vaswani and Biao Jie. At the same time, we further promote two works, for the first work, we innovatively apply the Transformer invented by them for text processing to the extraction of brain signals. For the BOLD signal of the brain, it can be regarded as a continuous speech signal itself. We can extract the most relevant features for disease diagnosis in the BOLD signal through the self-attention mechanism of the Transformer. For the method of Biao Jie to construct multiple levels of brain functional connections, we improve it to dynamically adjust the threshold of different levels according to the prediction results of the network. Through the above two methods, we solve the difficulty of previous work in extracting effective features from BOLD signals to establish high-order networks and the difficulty of manually setting appropriate thresholds. Through the comparison experiment of the final result and other methods, we find that our method exceeds the traditional method in terms of accuracy, specificity, sensitivity, and F1 value.

[0071] Similar to the principle of the above embodiment, the application provides an early cognitive dysfunction classification method.

[0072] The following provides specific embodiments in combination with the accompanying drawings:

[0073] As Figure 7 A flowchart of an early cognitive dysfunction classification method in an embodiment of the application is shown.

[0074] The method comprises:

[0075] Step S71: acquiring a collected functional magnetic resonance image to be classified.

[0076] Step S72: obtaining an early cognitive dysfunction classification result corresponding to the functional magnetic resonance image according to an early cognitive dysfunction classification model based on a Transformer and a multi-level functional connection.

[0077] Optionally, the early cognitive impairment classification model comprises: a preprocessing and partitioning module, configured to preprocess and partition the functional magnetic resonance image to obtain partitioned data; wherein the partitioned data comprises BOLD signals of one or more brain regions; a low-order classification module connected to the preprocessing and partitioning module, configured to construct a low-order functional connection network according to the partitioned data, and construct a multi-level first brain functional connection network using a threshold learning network, and obtain features corresponding to each level to obtain a low-order classification result corresponding to each feature; a high-order classification module connected to the preprocessing and partitioning module, configured to construct a low-order functional connection network according to the partitioned data based on a Transformer, and construct a multi-level second brain functional connection network using a threshold learning network, and obtain features corresponding to each level to obtain a high-order classification result corresponding to each feature; and a final classification result determination module connected to the low-order classification channel module and the high-order classification channel module, configured to obtain an early cognitive impairment classification result according to each low-order classification result and each high-order classification result.

[0078] Optionally, the low-order classification module comprises: a low-order functional connection construction submodule, configured to calculate similarity data of BOLD signals corresponding to each brain region according to the partitioned data to construct a low-order functional connection network; a first threshold learning submodule connected to the low-order functional connection construction submodule, configured to obtain a plurality of thresholds and obtain a multi-level first brain functional connection network according to an input low-order functional connection network based on a threshold learning network; and a low-order classification result acquisition submodule connected to the first threshold learning submodule, configured to extract features corresponding to each level according to the first brain functional connection network of each level based on a classification network structure using a graph neural network and a fully connected layer, and obtain a low-order classification result corresponding to each feature.

[0079] Optionally, the high-order classification module comprises: a high-order functional connection construction submodule, configured to extract high-order disease features from the partitioned data based on a Transformer, and calculate similarity data of the corresponding high-order disease features to construct a high-order functional connection network; a second threshold learning submodule connected to the high-order functional connection construction submodule, configured to obtain a plurality of thresholds and obtain a multi-level second brain functional connection network according to an input high-order functional connection network based on a threshold learning network; and a high-order classification result acquisition submodule connected to the second threshold learning submodule, configured to extract features corresponding to each level according to the second brain functional connection network of each level based on a classification network structure using a graph neural network and a fully connected layer, and obtain a high-order classification result corresponding to each feature.

[0080] Optionally, the high-order functional connection construction sub-module comprises: a Transformer unit, configured to extract corresponding high-order disease features from the partition data; a high-order similarity calculation unit connected to the Transformer unit, configured to calculate a Pearson correlation coefficient of the high-order disease features; and a high-order functional connection network construction unit connected to the similarity calculation unit, configured to obtain a high-order functional connection network according to the Pearson correlation coefficient.

[0081] Optionally, the Transformer unit is configured to perform matrix mapping transformation on the partition data to obtain a Query matrix, a Key matrix, and a Value matrix; calculate a similarity matrix based on the Query matrix and the Key matrix; perform a Softmax operation on the similarity matrix to obtain a self-attention matrix; and multiply the self-attention matrix with the Value matrix to obtain the high-order disease features.

[0082] Optionally, the threshold learning network comprises: one or more layers of graph neural networks, configured to input the high-order functional connection network to obtain corresponding graph neural output data, and configured to input the low-order functional connection network to obtain corresponding graph neural output data; a fully connected layer, configured to obtain a set number of thresholds based on the graph neural output data; a threshold contraction module connected to the fully connected layer, configured to contract and screen each threshold in a hyperbolic tangent manner to obtain a plurality of screened thresholds; and a brain functional connection network module connected to the threshold contraction module, configured to construct a corresponding multi-level first brain functional connection network based on the low-order functional connection network and the screened thresholds, and configured to construct a corresponding multi-level second brain functional connection network based on the high-order functional connection network and the screened thresholds.

[0083] Optionally, obtaining the early cognitive impairment classification result based on the low-order classification results and the high-order classification results comprises: obtaining the early cognitive impairment classification result based on the low-order classification results and the high-order classification results based on a voting mechanism.

[0084] As Figure 8 An early cognitive impairment classification terminal 10 in an embodiment of the present application is shown in a structural schematic diagram.

[0085] The early cognitive impairment classification terminal 80 comprises a memory 81 and a processor 82. The memory 81 is configured to store a computer program. The processor 82 is configured to run the computer program to implement the early cognitive impairment classification method as described above. Figure 2 The early cognitive impairment classification method.

[0086] Optionally, the number of the memories 81 can be one or more, and the number of the processors 82 can be one or more.Figure 8 Each example is taken as an instance.

[0087] Optionally, the processor 82 in the early cognitive impairment classification terminal 80 will perform the following actions: Figure 1 The steps described involve loading one or more instructions corresponding to the process of an application into memory 81, and then having the processor 82 run the application stored in the first memory 81, thereby achieving the following: Figure 1 The various functions in the classification method of early cognitive impairment.

[0088] Optionally, the memory 81 may include, but is not limited to, high-speed random access memory and non-volatile memory. For example, one or more disk storage devices, flash memory devices, or other non-volatile solid-state storage devices; the processor 82 may include, but is not limited to, a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0089] Optionally, the processor 82 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0090] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed, implements as follows: Figure 1The illustrated early cognitive impairment classification method. The computer readable storage medium can include, but is not limited to, a floppy disk, an optical disk, a CD-ROM (Compact Disc-Read Only Memory), a magneto-optical disk, a ROM (Read Only Memory), a RAM (Random Access Memory), an EPROM (Erasable Programmable Read Only Memory), an EEPROM (Electrically Erasable Programmable Read Only Memory), a magnetic card or an optical card, a flash memory, or other types of media / machine readable media suitable for storing machine executable instructions. The computer readable storage medium can be a product that has not been accessed by a computer device, or a component that has been accessed by a computer device.

[0091] In summary, the early cognitive impairment classification system, method and terminal of the present application obtain the early cognitive impairment classification result corresponding to the functional magnetic resonance image through the early cognitive impairment classification model based on the Transformer and the multi-level function connection; the present application extracts the most relevant features of the BOLD signal and the disease diagnosis based on the Transformer to obtain the classification result, greatly improving the accuracy and efficiency of the early cognitive impairment classification. Therefore, the present application effectively overcomes the various shortcomings in the prior art and has high industrial utilization value.

[0092] The above embodiments only exemplarily illustrate the principles and effects of the present application, and are not intended to limit the present application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical thought disclosed by the present application should be covered by the claims of the present application.

Claims

1. An early cognitive dysfunction classification system, characterized by, The system comprises: a data acquisition module configured to acquire a collected functional magnetic resonance image to be classified; a classification module connected to the data acquisition module and configured to obtain an early cognitive impairment classification result corresponding to the functional magnetic resonance image according to an early cognitive impairment classification model based on a Transformer and multi-level functional connectivity; the early cognitive impairment classification model comprises: a preprocessing and partitioning module configured to preprocess and partition the functional magnetic resonance image to obtain partitioned data; wherein the partitioned data comprises BOLD signals of one or more brain regions; a low-order classification module connected to the preprocessing and partitioning module and configured to construct a low-order functional connectivity network according to the partitioned data, construct a multi-level first brain functional connectivity network using a threshold learning network, and obtain features corresponding to each level to obtain a low-order classification result corresponding to each feature; a high-order classification module connected to the preprocessing and partitioning module and configured to construct a high-order functional connectivity network based on a Transformer according to the partitioned data, construct a multi-level second brain functional connectivity network using a threshold learning network, and obtain features corresponding to each level to obtain a high-order classification result corresponding to each feature; a final classification result determination module connected to the low-order classification channel module and the high-order classification channel module and configured to obtain an early cognitive impairment classification result according to each low-order classification result and each high-order classification result; the high-order classification module comprises: a high-order functional connectivity construction submodule configured to extract high-order disease features from the partitioned data based on a Transformer, calculate similarity data corresponding to the high-order disease features, and construct a high-order functional connectivity network; a second threshold learning submodule connected to the high-order functional connectivity construction submodule and configured to obtain a plurality of thresholds and a multi-level second brain functional connectivity network according to an input high-order functional connectivity network based on a threshold learning network; a high-order classification result acquisition submodule connected to the second threshold learning submodule and configured to extract features corresponding to each level from each level of the second brain functional connectivity network based on a classification network structure using a graph neural network and a fully connected layer, and obtain a high-order classification result corresponding to each feature; the high-order functional connectivity construction submodule comprises: a Transformer unit configured to extract corresponding high-order disease features from the partitioned data; a high-order similarity calculation unit connected to the Transformer unit and configured to calculate a Pearson correlation coefficient of the high-order disease features; a high-order functional connectivity network construction unit connected to the similarity calculation unit and configured to obtain a high-order functional connectivity network according to the Pearson correlation coefficient. The Transformer unit is used for matrix mapping transformation of the partition data to obtain a Query matrix, a Key matrix, and a Value matrix; a similarity matrix is calculated based on the Query matrix and the Key matrix; a self-attention matrix is obtained by performing a Softmax operation on the similarity matrix; and the high-order disease characteristics are obtained by multiplying the self-attention matrix with the Value matrix.

2. The early cognitive impairment classification system of claim 1, wherein, The low-order classification module comprises: a low-order functional connection construction submodule, configured to calculate similarity data of BOLD signals of corresponding brain regions according to the partition data, so as to construct a low-order functional connection network; a first threshold learning submodule connected to the low-order functional connection construction submodule, configured to obtain a plurality of threshold values and obtain a first brain functional connection network of a corresponding multi-level based on a threshold learning network and input low-order functional connection network; a low-order classification result acquisition submodule connected to the first threshold learning submodule, configured to extract features of each level based on a classification network structure adopting a graph neural network and a fully connected layer according to the first brain functional connection network of each level, and obtain low-order classification results corresponding to each feature.

3. The early cognitive impairment classification system according to claim 2 or 1, characterized in that, The threshold learning network comprises: one or more layers of graph neural networks, configured to input the high-order functional connection network to obtain corresponding graph neural output data, and also configured to input the low-order functional connection network to obtain corresponding graph neural output data; a fully connected layer, configured to obtain a set number of threshold values according to the graph neural output data; a threshold contraction module connected to the fully connected layer, configured to contract and screen each threshold value in a hyperbolic tangent manner to obtain a plurality of screened threshold values; a brain functional connection network module connected to the threshold contraction module, configured to construct a first brain functional connection network of a corresponding multi-level according to the low-order functional connection network and the screened threshold values, and also configured to construct a second brain functional connection network of a corresponding multi-level according to the high-order functional connection network and the screened threshold values.

4. The early cognitive impairment classification system of claim 1, wherein, The early cognitive impairment classification result is obtained according to each low-order classification result and each high-order classification result, comprising: obtaining the early cognitive impairment classification result based on a voting mechanism according to each low-order classification result and each high-order classification result.

5. A method of classifying early cognitive dysfunction, characterized by, The method comprises: acquiring a functional magnetic resonance image to be classified; obtaining an early cognitive impairment classification result corresponding to the functional magnetic resonance image according to an early cognitive impairment classification model based on a Transformer and a multi-level functional connection; The early cognitive impairment classification model comprises: preprocessing and partitioning the functional magnetic resonance image to obtain partition data; wherein the partition data comprises BOLD signals of one or more brain regions; constructing a low-order functional connection network according to the partition data and constructing a first brain functional connection network of a multi-level using a threshold learning network, and obtaining features of each level to obtain low-order classification results corresponding to each feature; The high-order functional connection network is constructed based on the partition data by using a Transformer, a threshold learning network is used to construct a multi-level second brain functional connection network, and features of each level are obtained to obtain a high-order classification result corresponding to each feature. Specifically, high-order disease features are extracted from the partition data by using the Transformer, and similarity data corresponding to the high-order disease features is calculated to construct a high-order functional connection network. A Transformer unit is used to extract corresponding high-order disease features from the partition data. A high-order similarity calculation unit is connected to the Transformer unit and is used to calculate a Pearson correlation coefficient of the high-order disease features. A high-order functional connection network construction unit is connected to the similarity calculation unit and is used to obtain a high-order functional connection network according to the Pearson correlation coefficient. The Transformer unit is used to perform matrix mapping transformation on the partition data to obtain a Query matrix, a Key matrix and a Value matrix. A similarity matrix is calculated based on the Query matrix and the Key matrix. A self-attention matrix is obtained by performing a Softmax operation on the similarity matrix. The self-attention matrix is multiplied with the Value matrix to obtain the high-order disease features. A threshold learning network is used to obtain a plurality of thresholds and a multi-level second brain functional connection network according to an input high-order functional connection network. A classification network structure based on a graph neural network and a fully connected layer is used to extract features of each level according to the second brain functional connection network of each level, and a high-order classification result corresponding to each feature is obtained. An early cognitive impairment classification result is obtained according to the low-order classification results and the high-order classification results.

6. An early cognitive impairment classification terminal, characterized by, The method comprises the following steps: one or more memories and one or more processors; the one or more memories are used to store a computer program; the one or more processors are connected to the memories and are used to run the computer program to perform the method of claim 5.

Images