CN110555828A - Brain age prediction method and device based on 3D convolutional neural network - Google Patents
Brain age prediction method and device based on 3D convolutional neural network Download PDFInfo
- Publication number
- CN110555828A CN110555828A CN201910733532.9A CN201910733532A CN110555828A CN 110555828 A CN110555828 A CN 110555828A CN 201910733532 A CN201910733532 A CN 201910733532A CN 110555828 A CN110555828 A CN 110555828A
- Authority
- CN
- China
- Prior art keywords
- data
- brain
- age
- magnetic resonance
- sampling
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Quality & Reliability (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Radiology & Medical Imaging (AREA)
- Genetics & Genomics (AREA)
- Proteomics, Peptides & Aminoacids (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biotechnology (AREA)
- Evolutionary Biology (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The application discloses a brain age prediction method based on a 3D convolutional neural network. The method comprises the following steps: carrying out layered sampling on the brain nuclear magnetic resonance imaging data; inputting the brain nuclear magnetic resonance imaging data after the layered sampling into a 3D convolutional neural network for training and extracting characteristic data through multithreading; constructing a brain age prediction regression model according to the extracted feature data; and outputting a brain age prediction result according to the brain age prediction regression model. The method and the device solve the technical problem that the prediction of the brain age of the child is not accurate enough due to the fact that information is easy to lose in the aspect of feature selection in the traditional machine learning model in the related technology, and achieve the technical effect of improving the accuracy of the prediction of the brain age of the child.
Description
Technical Field
The application relates to the technical field of deep learning, in particular to a brain age prediction method and device based on a 3D convolutional neural network.
Background
magnetic Resonance Imaging (MRI) offers the opportunity for the assessment of brain development with its high spatial and high density resolution, but the maturity of brain development in children is difficult to assess empirically by imaging physicians and must be calculated based on computer quantitative measurements. Healthy brain development is a very complex process in children, adolescents and early adulthood, manifested by heterogeneity in the sequence and pattern of tissue development in different regions of the brain. In general, the White Matter (WM) volume gradually increases with age in children and adolescents, while the Grey Matter (GM) volume decreases with age, with different brain areas having different developmental trends and rates. The potential patterning of these brain microstructures during development provides a basis for the assessment of brain development.
Convolutional Neural Networks (CNN) are a type of feed forward Neural Networks (fed Neural Networks) that include convolution calculation and have a deep structure, and are one of the representative algorithms of deep learning (deep learning), and Convolutional Neural Networks have a feature learning (representation) capability.
The inventor finds that the traditional machine learning model in the related art is easy to cause information loss in the aspect of feature selection, so that the brain age of the child cannot be accurately estimated and predicted.
Aiming at the problem that the traditional machine learning model in the related technology is not accurate enough in the prediction of the brain age of children due to the fact that information is easy to lose in the aspect of feature selection, an effective solution is not provided at present.
Disclosure of Invention
The application mainly aims to provide a brain age prediction method and a brain age prediction device based on a 3D convolutional neural network, so as to solve the problem that the traditional machine learning model in the related art is not accurate enough in the brain age prediction of children due to the fact that information is easy to lose in the aspect of feature selection.
in order to achieve the above object, according to one aspect of the present application, there is provided a brain age prediction method based on a 3D convolutional neural network.
the brain age prediction method based on the 3D convolutional neural network comprises the following steps: carrying out layered sampling on the brain nuclear magnetic resonance imaging data; inputting the brain nuclear magnetic resonance imaging data after the layered sampling into a 3D convolutional neural network for training and extracting characteristic data through multithreading; constructing a brain age prediction regression model according to the extracted feature data; and outputting a brain age prediction result according to the brain age prediction regression model.
Further, the stratified sampling of the brain magnetic resonance imaging data comprises: extracting cranial data in the cerebral nuclear magnetic resonance imaging data; performing data registration on the cranial data; and segmenting the registered brain and cranium data to obtain brain gray matter and brain white matter segmentation data.
further, the hierarchically sampling brain magnetic resonance imaging data comprises: setting the same data volume threshold value for data of different age sections corresponding to the brain nuclear magnetic resonance imaging data; judging whether the data volume of the age data is higher than the data volume threshold value; if the data volume of the age data is higher than the data volume threshold value, numbering the age data and storing the age data; and if the data volume of the age data is lower than the data volume threshold value, performing a sample with a place back on the age data.
Further, if the data amount of the age data is lower than the data amount threshold, performing a sample with put back on the data in the age data further includes: stopping the put-back sampling when the number of times of the put-back sampling reaches a sampling threshold; and numbering and storing the age data obtained after the sample is replaced.
Further, the step of inputting the brain magnetic resonance imaging data after the hierarchical sampling into a 3D convolutional neural network for training and extracting feature data through multiple threads includes: sequentially inputting the brain nuclear magnetic resonance imaging data into a data generator according to the object list of the brain nuclear magnetic resonance imaging data; calling a keras preprocessing interface to preprocess the brain nuclear magnetic resonance imaging data; and matching the preprocessed brain nuclear magnetic resonance imaging data with a corresponding age label.
In order to achieve the above object, according to another aspect of the present application, there is provided a brain age prediction apparatus based on a 3D convolutional neural network.
The brain age prediction device based on the 3D convolutional neural network comprises: the sampling module is used for carrying out layered sampling on the brain nuclear magnetic resonance imaging data; the training module is used for inputting the brain nuclear magnetic resonance imaging data after the layered sampling into a 3D convolutional neural network for training and extracting characteristic data through multithreading; the construction module is used for constructing a brain age prediction regression model according to the extracted feature data; and the output module is used for outputting a brain age prediction result according to the brain age prediction regression model.
Further, still include: the extraction module is used for extracting the cranial data in the cerebral magnetic resonance imaging data; the registration module is used for carrying out data registration on the cranial data; and the segmentation module is used for segmenting the registered brain data to obtain brain gray matter and brain white matter segmentation data.
Further, the sampling module includes: a setting unit configured to set the same data amount threshold for data of different age groups corresponding to the brain nuclear magnetic resonance imaging data; a judging unit configured to judge whether or not a data amount of the age data is higher than the data amount threshold; a first storage unit configured to number and store the age data if the data amount of the age data is higher than the data amount threshold; a sampling unit for performing a put-back sampling on the age data if the data amount of the age data is lower than the data amount threshold.
Further, the sampling module further comprises: a termination unit for stopping the put-back sampling when the number of times of the put-back sampling reaches a sampling threshold; and the second storage unit is used for numbering and storing the age section data obtained after the sample is put back.
Further, the training module further comprises: the input unit is used for sequentially inputting the brain nuclear magnetic resonance imaging data into the data generator according to the object list of the brain nuclear magnetic resonance imaging data; the processing unit is used for calling a keras preprocessing interface to preprocess the brain nuclear magnetic resonance imaging data; and the matching unit is used for matching the preprocessed brain nuclear magnetic resonance imaging data with the corresponding age label.
In the embodiment of the application, a mode of hierarchically sampling brain nuclear magnetic resonance imaging data is adopted, the hierarchically sampled brain nuclear magnetic resonance imaging data is input into a 3D convolutional neural network through multiple threads to be trained, characteristic data is extracted, a brain age prediction regression model is constructed according to the extracted characteristic data, and the purpose of accurately outputting a brain age prediction result according to the brain age prediction regression model is achieved, so that the technical effect of improving the accuracy of children's brain age prediction is achieved, and the technical problem that the children's brain age prediction is not accurate enough due to the fact that information is easily lost in the aspect of feature selection in a traditional machine learning model in the related technology is solved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, serve to provide a further understanding of the application and to enable other features, objects, and advantages of the application to be more apparent. The drawings and their description illustrate the embodiments of the invention and do not limit it. In the drawings:
Fig. 1 is a schematic flow chart of a brain age prediction method based on a 3D convolutional neural network according to a first embodiment of the present application;
FIG. 2 is a schematic flow chart of a brain age prediction method based on a 3D convolutional neural network according to a second embodiment of the present application;
FIG. 3 is a schematic flow chart of a brain age prediction method based on a 3D convolutional neural network according to a third embodiment of the present application;
FIG. 4 is a schematic flow chart of a brain age prediction method based on a 3D convolutional neural network according to a fourth embodiment of the present application;
FIG. 5 is a schematic flow chart of a brain age prediction method based on a 3D convolutional neural network according to a fifth embodiment of the present application;
FIG. 6 is a schematic diagram of a structure of a brain age prediction device based on a 3D convolutional neural network according to a first embodiment of the present application;
FIG. 7 is a schematic diagram of a structure of a brain age prediction device based on a 3D convolutional neural network according to a second embodiment of the present application;
FIG. 8 is a schematic structural diagram of a brain age prediction device based on a 3D convolutional neural network according to a third embodiment of the present application; and
fig. 9 is a schematic structural diagram of a brain age prediction device based on a 3D convolutional neural network according to a fourth embodiment of the present application.
Detailed Description
In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only partial embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
it should be noted that the terms "first," "second," and the like in the description and claims of this application and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It should be understood that the data so used may be interchanged under appropriate circumstances such that embodiments of the application described herein may be used. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.
According to an embodiment of the present application, there is provided a brain age prediction method based on a 3D convolutional neural network, as shown in fig. 1, the method includes steps S101 to S104 as follows:
and step S101, carrying out layered sampling on the brain nuclear magnetic resonance imaging data.
During specific implementation, for training of a child brain age model, the data volume is required to be sufficient, and the data distribution of each age group is required to be as uniform as possible, so that hierarchical sampling is firstly required to be performed on brain nuclear magnetic resonance imaging data, the data distribution of each age group is ensured to be sufficient and uniform as far as possible, the probability of data input models of different age groups is ensured to be equivalent, and the reliability of subsequent output models is improved.
and S102, inputting the brain nuclear magnetic resonance imaging data after the hierarchical sampling into a 3D convolutional neural network for training and extracting characteristic data through multithreading.
In particular, Magnetic Resonance Imaging (MRI) scanning can provide 3D images of specific body parts, which generate highly detailed images from various angles, thereby resulting in a very large number of 3D MRI data pixel points. Machine learning in the related art generally loads all training data into a memory and then transfers the training data to a network when training a model, but when the memory is limited and the data size is too large, the method cannot be used any more. Therefore, the embodiment of the application changes the mode that all data are loaded into the memory at one time, and data are generated in real time through multi-thread processing and immediately input into the model for training.
Preferably, since the 2D convolutional neural network cannot capture context information of the brain image structure, the embodiment of the present application performs feature extraction using a 3D Convolutional Neural Network (CNN) in order to extract spatial features of the brain structure. The 3D CNN excavates spatial local correlation information of the brain MRI image by enhancing a local connection mode of nodes between adjacent layers in the neural network, so that important information of the image is effectively extracted. The inputs to the level m nodes are only a portion of the level m-1 nodes, which have spatially adjacent visual fields. In addition, 3D CNNs also have the advantage of sharing convolution kernels, which is not stressful for high-dimensional data processing. The method preferably uses 3D max firing to compress the input features, extracts main features and reduces the spatial feature dimension of the data.
And step S103, constructing a brain age prediction regression model according to the extracted feature data.
In specific implementation, a brain age prediction regression model is constructed based on the feature data extracted after the 3D convolutional neural network training. Preferably, supervised learning with age labels can enable the 3D convolutional neural network to extract features with significant brain age differentiation as early as possible in the learning process, and the model convergence speed is high. The network adds smoothness loss on the basis of mean square error (mse) loss, so that the model is insensitive to outliers, the robustness of the model is improved, the uncertainty of the model is reduced, and the probability distribution of model prediction is closer to the true value of the mean value.
and step S104, outputting a brain age prediction result according to the brain age prediction regression model.
In specific implementation, the brain age of the child can be predicted after the brain age prediction regression model is obtained, and the brain nuclear magnetic resonance imaging data of the child is input into the prediction regression model, so that the brain age prediction result of the child can be obtained, a doctor can be assisted in brain age evaluation, and the workload of the doctor is reduced.
According to a preferred implementation of the embodiment of the present application, as shown in fig. 2, the step S201 to step S203 before the hierarchical sampling of the brain mri data includes the following steps:
Step S201, extracting cranial data in the cerebral magnetic resonance imaging data.
In specific implementation, before performing hierarchical sampling on the brain mri data, certain preprocessing needs to be performed on the brain mri data. Preferably, since the original brain mri data includes data of structures such as skull, neck, cerebellum, etc., which do not contribute to the brain age prediction, but rather add extra noise to the model, the data of the structures should be removed first, and only the skull data in the brain mri data is extracted as the basic data for the subsequent analysis.
and S202, carrying out data registration on the cranial data.
in specific implementation, because the size, shape and position of the anatomical structure of the brain are different due to physiological differences of different people, the extracted cranial data are preferably strictly aligned, and the data are spatially consistent through registration, so that the reliability and accuracy of the data are ensured.
And S203, segmenting the registered brain data to obtain brain gray matter and brain white matter segmentation data.
in specific implementation, analysis of the volume size of gray matter and white matter is one of the main methods for predicting brain age, the ratio of the volume of Gray Matter (GM) to the volume of White Matter (WM) shows different trends with the age, generally speaking, the volume of brain WM gradually increases with age in children and adolescents, while the volume of brain GM continuously decreases with age, and the development trends and rates of different brain areas are different. Therefore, according to the embodiment of the application, the gray matter segmentation map and the white matter segmentation map in the brain nuclear magnetic resonance imaging are respectively obtained through probability segmentation, and the segmentation data of the gray matter and the white matter of the brain are obtained through calculation. All the above data preprocessing processes can be implemented by spm (statistical Parametric mapping)12 software.
according to a preferred embodiment of the present application, as shown in fig. 3, the hierarchical sampling of the brain mri data includes the following steps S301 to S304:
step S301, setting the same data volume threshold value for data of different age groups corresponding to the brain nuclear magnetic resonance imaging data.
in specific implementation, when hierarchical sampling is performed on brain magnetic resonance imaging data, the same data volume threshold needs to be set for each age group, so as to ensure that the probability of entering the model training of each age group data is equivalent.
step S302, determining whether the data amount of the age data is higher than the data amount threshold.
In specific implementation, the data volume in each age group is compared with a preset data volume threshold value respectively to determine whether the data volume of each age group meets preset requirements, and then the probability that data of different age groups enter model training is ensured to be equivalent.
Step S303, if the data amount of the age data is higher than the data amount threshold, numbering and storing the age data.
in specific implementation, if the data volume in the age data is higher than a preset data volume threshold, which indicates that the data volume of the age reaches a preset requirement, the actual data volume in the age is numbered one by one and stored in a file. For example, if the preset threshold value of the data amount is 100 and the data amount in the age range of 1 to 3 years is actually 120, 120 data in the age range of 1 to 3 years are numbered one by one and stored.
Step S304, if the data volume of the age data is lower than the data volume threshold value, performing a sample with a place back on the age data.
In specific implementation, if the data amount in the age data is lower than a preset data amount threshold, it indicates that the data amount of the age does not meet the preset requirement, and at this time, it is preferable to perform back sampling on the age data to ensure that the probability of the data of the following different ages entering the model training is equivalent.
According to a preferred implementation of the embodiment of the present application, as shown in fig. 4, if the data amount of the age data is lower than the data amount threshold, the method further includes the following steps S401 to S402 after performing the replacement sampling on the data in the age data:
Step S401, when the number of times of the putting back sampling reaches a sampling threshold value, stopping the putting back sampling.
In a specific implementation, when the number of times of the put-back sampling reaches a preset sampling threshold, the put-back sampling is terminated, and preferably, the sampling threshold of the put-back sampling is determined by a difference between the data amount of the age group data subjected to the put-back sampling and the preset data amount threshold. For example, if the data size of the data of the age group of 3 to 5 years is 80 and is lower than the preset data size threshold 100, the data of the age group of 3 to 5 years is subjected to the sample with put back, the sample threshold with put back sample or the sample frequency is 100 to 80 times to 20 times, and when the sample with put back sample frequency reaches 20 times, the sampling is ended.
And step S402, numbering and storing the age group data obtained after the sample is replaced.
In the concrete implementation, when the number of times of putting back samples reaches the sampling threshold, the data size of the age data with the putting back samples reaches the preset requirement, the probability that the data of different ages enter the model training is basically ensured to be equivalent, and at the moment, the age data obtained after the putting back samples are numbered one by one and stored in the file as the processing mode of other age data.
According to a preferred implementation of the embodiment of the present application, as shown in fig. 5, the step of inputting the brain mri data after the hierarchical sampling into a 3D convolutional neural network for training and extracting feature data through multiple threads includes the following steps S501 to S503:
Step S501, the brain nuclear magnetic resonance imaging data are sequentially input into a data generator according to the object list of the brain nuclear magnetic resonance imaging data.
in specific implementation, when the brain mri data after the hierarchical sampling is input into the 3D convolutional neural network for multi-thread processing, the data is preferably sequentially transmitted into the data generator according to an object list of the brain mri data.
And step S502, calling a keras preprocessing interface to preprocess the brain nuclear magnetic resonance imaging data.
In specific implementation, the data is preprocessed by whitening (ZCA) and the like by calling a keras preprocessing interface (keras).
And step S503, matching the preprocessed brain nuclear magnetic resonance imaging data with a corresponding age label.
In specific implementation, the preprocessed brain magnetic resonance imaging data are matched with corresponding age labels, then the data are disordered in sequence, and data flow is continuously provided in the circulating process of model training.
From the above description, it can be seen that the present invention achieves the following technical effects: the method comprises the steps of inputting the brain nuclear magnetic resonance imaging data subjected to hierarchical sampling into a 3D convolutional neural network through multiple threads to train and extract characteristic data, and constructing a brain age prediction regression model according to the extracted characteristic data, so that the purpose of accurately outputting a brain age prediction result according to the brain age prediction regression model is achieved, the technical problem that the traditional machine learning model in the related technology is not accurate in children brain age prediction due to the fact that information is easy to lose in the aspect of feature selection is solved, and the technical effect of improving the accuracy of children brain age prediction is achieved.
according to an embodiment of the present invention, there is also provided a brain age prediction apparatus for implementing the above-mentioned brain age prediction method based on a 3D convolutional neural network, as shown in fig. 6, the apparatus including: the device comprises a sampling module 1, a training module 2, a construction module 3 and an output module 4.
The sampling module 1 in the embodiment of the present application is configured to perform hierarchical sampling on brain magnetic resonance imaging data.
During specific implementation, for training of a child brain age model, the data volume is required to be sufficient, and the data distribution of each age group is required to be as uniform as possible, so that the brain nuclear magnetic resonance imaging data is firstly required to be hierarchically sampled through the sampling module 1, the data distribution of each age group is ensured to be sufficient and uniform as much as possible, the probability weights of data input models of different age groups are ensured to be equivalent, and the reliability of subsequent output models is improved.
The training module 2 in the embodiment of the application is used for inputting the brain nuclear magnetic resonance imaging data after the hierarchical sampling into a 3D convolutional neural network for training and extracting feature data through multithreading.
In particular, Magnetic Resonance Imaging (MRI) scanning can provide 3D images of specific body parts, which generate highly detailed images from various angles, thereby resulting in a very large number of 3D MRI data pixel points. Machine learning in the related art generally loads all training data into a memory and then transfers the training data to a network when training a model, but when the memory is limited and the data size is too large, the method cannot be used any more. Therefore, the embodiment of the application changes the mode that all data are loaded into the memory at one time, but the data are generated in real time through multi-thread processing, and the data are immediately input into the model through the training module 2 for training.
Preferably, the application preferably uses 3D max pooling to compress the input features, extracts main features, and reduces the spatial feature dimension of the data.
the building module 3 in the embodiment of the present application is configured to build a brain age prediction regression model according to the extracted feature data.
in specific implementation, a brain age prediction regression model is constructed through the construction module 3 based on the feature data extracted after the 3D convolutional neural network training. Preferably, supervised learning with age labels can enable the 3D convolutional neural network to extract features with significant brain age differentiation as early as possible in the learning process, and the model convergence speed is high. The network adds smoothness loss on the basis of mean square error (mse) loss, so that the model is insensitive to outliers, the robustness of the model is improved, the uncertainty of the model is reduced, and the probability distribution of model prediction is closer to the true value of the mean value.
the output module 4 in this embodiment of the application is configured to output a brain age prediction result according to the brain age prediction regression model.
In specific implementation, the brain age of the child can be predicted after the brain age prediction regression model is obtained, the brain nuclear magnetic resonance imaging data of the child is input into the prediction regression model, the brain age prediction result of the child can be obtained through the output module 4, and therefore the brain age prediction model can assist a doctor in brain age assessment, and the workload of the doctor is relieved.
According to a preferred implementation of the embodiment of the present application, as shown in fig. 7, the apparatus further includes: an extraction module 5, a registration module 6 and a segmentation module 7.
the extraction module 5 in the embodiment of the present application is configured to extract the cranial data in the brain magnetic resonance imaging data.
In specific implementation, before performing hierarchical sampling on the brain mri data, certain preprocessing needs to be performed on the brain mri data. Preferably, since the original brain mri data includes data of structures such as skull, neck, cerebellum, etc., which do not contribute to the brain age prediction, but rather add extra noise to the model, the data of the structures should be removed first, and the extraction module 5 extracts the skull data in the brain mri data as the basic data for the subsequent analysis.
The registration module 6 in the embodiment of the present application is configured to perform data registration on the cranial data.
In specific implementation, because the size, shape and position of the anatomical structure of the brain are different due to physiological differences of different people, the extracted cranial data are preferably aligned strictly, and the data are aligned spatially by the registration module 6, so that the reliability and accuracy of the data are ensured.
the segmentation module 7 in this embodiment of the application is configured to obtain the gray brain matter and white brain matter segmentation data by segmenting the registered cranial data.
In specific implementation, analysis of the volume size of gray matter and white matter is one of the main methods for predicting brain age, the ratio of the volume of Gray Matter (GM) to the volume of White Matter (WM) shows different trends with the age, generally speaking, the volume of brain WM gradually increases with age in children and adolescents, while the volume of brain GM continuously decreases with age, and the development trends and rates of different brain areas are different. Therefore, in the embodiment of the application, the segmentation module 7 is used for performing probability segmentation to respectively obtain the gray matter segmentation map and the white matter segmentation map in the brain nuclear magnetic resonance imaging, and the segmentation data of the gray matter and the white matter of the brain are obtained through calculation.
according to a preferred implementation of the embodiment of the present application, as shown in fig. 8, the sampling module 1 includes:
The setting unit 11 in the embodiment of the present application is configured to set the same data amount threshold for data of different age groups corresponding to the brain magnetic resonance imaging data.
In specific implementation, when performing hierarchical sampling on the brain mri data, the same data volume threshold needs to be set for each age group through the setting unit 11, so as to ensure that the probability of entering the model training of each age group data is equivalent.
The judging unit 12 in the embodiment of the present application is configured to judge whether the data amount of the age data is higher than the data amount threshold.
in specific implementation, the data volume in each age group is compared with a preset data volume threshold value through the judging unit 12, so as to determine whether the data volume of each age group meets the preset requirement, and further ensure that the probability of the data of different age groups entering the model training is equivalent.
The first storage unit 13 in the embodiment of the present application is configured to number and store the age data if the data amount of the age data is higher than the data amount threshold.
In specific implementation, if the data volume in the age data is higher than a preset data volume threshold, which indicates that the data volume of the age reaches a preset requirement, the actual data volume in the age is numbered one by one and stored in a file. For example, if the preset threshold value of the data amount is 100 and the data amount in the age range of 1 to 3 years is actually 120, 120 data in the age range of 1 to 3 years are numbered one by one and stored in the first storage unit 13.
The sampling unit 14 in the embodiment of the present application is configured to perform a put-back sampling on the age data if the data amount of the age data is lower than the data amount threshold.
In specific implementation, if the data amount in the age data is lower than the preset data amount threshold, which indicates that the data amount of the age data does not meet the preset requirement, at this time, the sampling unit 14 preferably performs the sampling with the replacement on the age data to ensure that the probability of the data of the following different ages entering the model training is equivalent.
According to a preferred implementation of the embodiment of the present application, as shown in fig. 8, the sampling module 1 further includes: a termination unit 15 and a second storage unit 16.
the terminating unit 15 in this embodiment of the present application is configured to stop the putting back sample when the number of times of the putting back sample reaches a sample threshold.
in practical implementation, when the number of times of the put-back sampling reaches a preset sampling threshold, the put-back sampling is terminated by the termination unit 15, and preferably, the sampling threshold of the put-back sampling is determined by a difference between the data amount of the age group data subjected to the put-back sampling and the preset data amount threshold. For example, if the data size of the data of the age group of 3 to 5 years is 80 and is lower than the preset data size threshold 100, the data of the age group of 3 to 5 years is subjected to the sample with put back, the sample threshold with put back sample or the sample frequency is 100 to 80 times to 20 times, and when the sample with put back sample frequency reaches 20 times, the sampling is ended.
The second storage unit 16 in the embodiment of the present application is configured to number and store the age group data obtained after the sample with put back.
in specific implementation, when the number of times of sample putting back reaches the sampling threshold, it is described that the data size of the age data with sample putting back reaches the preset requirement, and basically, it can be ensured that the probability of the data of different ages entering the model training is equivalent, at this time, as with the processing mode of other age data, the age data obtained after sample putting back is numbered one by one through the second storage unit 16 and stored in the file.
According to a preferred implementation of the embodiment of the present application, as shown in fig. 9, the training module 2 further includes: an input unit 21, a processing unit 22 and a matching unit 23.
The input unit 21 in this embodiment is configured to sequentially input the brain mri data into the data generator according to the object list of the brain mri data.
In specific implementation, when the brain mri data after the hierarchical sampling is input into the 3D convolutional neural network for multi-thread processing, the data is preferably sequentially transmitted into the data generator through the input unit 21 according to the object list of the brain mri data.
the processing unit 22 in this embodiment of the application is configured to invoke a keras preprocessing interface to preprocess the brain mri data.
In specific implementation, the processing unit 22 calls a keras preprocessing interface (keras preprocessing) to perform preprocessing such as whitening (ZCA) on the data, so as to remove redundant information in the input data and achieve the purpose of reducing dimensions.
The matching unit 23 in this embodiment of the application is configured to match the preprocessed brain mri data with a corresponding age tag.
in specific implementation, the matching unit 23 matches the preprocessed brain mri data with corresponding age labels, and then data sequence is disturbed, so as to continuously provide data flow in the cyclic process of model training.
It will be apparent to those skilled in the art that the modules or steps of the present invention described above may be implemented by a general purpose computing device, they may be centralized on a single computing device or distributed across a network of multiple computing devices, and they may alternatively be implemented by program code executable by a computing device, such that they may be stored in a storage device and executed by a computing device, or fabricated separately as individual integrated circuit modules, or fabricated as a single integrated circuit module from multiple modules or steps. Thus, the present invention is not limited to any specific combination of hardware and software.
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (10)
1. A brain age prediction method based on a 3D convolutional neural network is characterized by comprising the following steps:
Carrying out layered sampling on the brain nuclear magnetic resonance imaging data;
inputting the brain nuclear magnetic resonance imaging data after the layered sampling into a 3D convolutional neural network for training and extracting characteristic data through multithreading;
Constructing a brain age prediction regression model according to the extracted feature data;
And outputting a brain age prediction result according to the brain age prediction regression model.
2. The brain age prediction method based on the 3D convolutional neural network of claim 1, wherein the hierarchically sampling the brain nuclear magnetic resonance imaging data comprises:
Extracting cranial data in the cerebral nuclear magnetic resonance imaging data;
Performing data registration on the cranial data;
and segmenting the registered brain and cranium data to obtain brain gray matter and brain white matter segmentation data.
3. The brain age prediction method based on the 3D convolutional neural network of claim 1, wherein the hierarchically sampling the brain mri data comprises:
Setting the same data volume threshold value for data of different age sections corresponding to the brain nuclear magnetic resonance imaging data;
Judging whether the data volume of the age data is higher than the data volume threshold value;
If the data volume of the age data is higher than the data volume threshold value, numbering the age data and storing the age data;
And if the data volume of the age data is lower than the data volume threshold value, performing a sample with a place back on the age data.
4. The brain age prediction method based on the 3D convolutional neural network of claim 3, wherein if the data amount of the age data is lower than the data amount threshold, after performing the live-back sampling on the data in the age data, further comprises:
Stopping the put-back sampling when the number of times of the put-back sampling reaches a sampling threshold;
and numbering and storing the age data obtained after the sample is replaced.
5. The brain age prediction method based on the 3D convolutional neural network as claimed in claim 1, wherein the multithreading of inputting the brain magnetic resonance imaging data after the hierarchical sampling into the 3D convolutional neural network for training and extracting feature data comprises:
Sequentially inputting the brain nuclear magnetic resonance imaging data into a data generator according to the object list of the brain nuclear magnetic resonance imaging data;
Calling a keras preprocessing interface to preprocess the brain nuclear magnetic resonance imaging data;
And matching the preprocessed brain nuclear magnetic resonance imaging data with a corresponding age label.
6. A brain age prediction device based on a 3D convolutional neural network, comprising:
the sampling module is used for carrying out layered sampling on the brain nuclear magnetic resonance imaging data;
The training module is used for inputting the brain nuclear magnetic resonance imaging data after the layered sampling into a 3D convolutional neural network for training and extracting characteristic data through multithreading;
The construction module is used for constructing a brain age prediction regression model according to the extracted feature data;
and the output module is used for outputting a brain age prediction result according to the brain age prediction regression model.
7. the 3D convolutional neural network-based brain age prediction device of claim 6, further comprising:
The extraction module is used for extracting the cranial data in the cerebral magnetic resonance imaging data;
The registration module is used for carrying out data registration on the cranial data;
And the segmentation module is used for segmenting the registered brain data to obtain brain gray matter and brain white matter segmentation data.
8. The 3D convolutional neural network-based brain age prediction device of claim 6, wherein the sampling module comprises:
a setting unit configured to set the same data amount threshold for data of different age groups corresponding to the brain nuclear magnetic resonance imaging data;
A judging unit configured to judge whether or not a data amount of the age data is higher than the data amount threshold;
A first storage unit configured to number and store the age data if the data amount of the age data is higher than the data amount threshold;
a sampling unit for performing a put-back sampling on the age data if the data amount of the age data is lower than the data amount threshold.
9. The 3D convolutional neural network-based brain age prediction device of claim 8, wherein the sampling module further comprises:
A termination unit for stopping the put-back sampling when the number of times of the put-back sampling reaches a sampling threshold;
And the second storage unit is used for numbering and storing the age section data obtained after the sample is put back.
10. The 3D convolutional neural network-based brain age prediction device of claim 6, wherein the training module further comprises:
the input unit is used for sequentially inputting the brain nuclear magnetic resonance imaging data into the data generator according to the object list of the brain nuclear magnetic resonance imaging data;
The processing unit is used for calling a keras preprocessing interface to preprocess the brain nuclear magnetic resonance imaging data;
and the matching unit is used for matching the preprocessed brain nuclear magnetic resonance imaging data with the corresponding age label.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910733532.9A CN110555828A (en) | 2019-08-08 | 2019-08-08 | Brain age prediction method and device based on 3D convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910733532.9A CN110555828A (en) | 2019-08-08 | 2019-08-08 | Brain age prediction method and device based on 3D convolutional neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110555828A true CN110555828A (en) | 2019-12-10 |
Family
ID=68737253
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910733532.9A Pending CN110555828A (en) | 2019-08-08 | 2019-08-08 | Brain age prediction method and device based on 3D convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110555828A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110859624A (en) * | 2019-12-11 | 2020-03-06 | 北京航空航天大学 | Brain age deep learning prediction system based on structural magnetic resonance image |
CN110969614A (en) * | 2019-12-11 | 2020-04-07 | 中国科学院自动化研究所 | Brain age prediction method and system based on three-dimensional convolutional neural network |
CN111415361A (en) * | 2020-03-31 | 2020-07-14 | 浙江大学 | Method and device for estimating brain age of fetus and detecting abnormality based on deep learning |
CN111640500A (en) * | 2020-04-30 | 2020-09-08 | 西北大学 | Brain age prediction model and method based on bilinear fusion and attention mechanism |
CN113221927A (en) * | 2020-01-21 | 2021-08-06 | 宏碁股份有限公司 | Model training method and electronic device |
CN113298286A (en) * | 2021-03-31 | 2021-08-24 | 捷佳润科技集团股份有限公司 | Machine learning-based pitaya marketing time prediction method |
CN113545751A (en) * | 2021-07-20 | 2021-10-26 | 遵义医科大学附属医院 | Machine learning-based brain age assessment method for Rolandic epileptic children |
CN114881943A (en) * | 2022-04-22 | 2022-08-09 | 深圳市铱硙医疗科技有限公司 | Artificial intelligence-based brain age prediction method, device, equipment and storage medium |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105046709A (en) * | 2015-07-14 | 2015-11-11 | 华南理工大学 | Nuclear magnetic resonance imaging based brain age analysis method |
-
2019
- 2019-08-08 CN CN201910733532.9A patent/CN110555828A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105046709A (en) * | 2015-07-14 | 2015-11-11 | 华南理工大学 | Nuclear magnetic resonance imaging based brain age analysis method |
Non-Patent Citations (2)
Title |
---|
JAMES H COLE: "Predicting brain age with deep", 《ISI_WEB OF SCIENCE》 * |
刘红英: "关于多目标分层抽样方法及其应用研究", 《HTTP://M.DOC88.COM/MIP-0774738795525.HTML》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110969614A (en) * | 2019-12-11 | 2020-04-07 | 中国科学院自动化研究所 | Brain age prediction method and system based on three-dimensional convolutional neural network |
CN110859624A (en) * | 2019-12-11 | 2020-03-06 | 北京航空航天大学 | Brain age deep learning prediction system based on structural magnetic resonance image |
CN110969614B (en) * | 2019-12-11 | 2024-03-29 | 中国科学院自动化研究所 | Brain age prediction method and system based on three-dimensional convolutional neural network |
CN113221927A (en) * | 2020-01-21 | 2021-08-06 | 宏碁股份有限公司 | Model training method and electronic device |
CN113221927B (en) * | 2020-01-21 | 2024-01-23 | 宏碁股份有限公司 | Model training method and electronic device |
CN111415361A (en) * | 2020-03-31 | 2020-07-14 | 浙江大学 | Method and device for estimating brain age of fetus and detecting abnormality based on deep learning |
CN111415361B (en) * | 2020-03-31 | 2021-01-19 | 浙江大学 | Method and device for estimating brain age of fetus and detecting abnormality based on deep learning |
CN111640500B (en) * | 2020-04-30 | 2023-09-05 | 西北大学 | Brain Age Prediction Model and Method Based on Bilinear Fusion and Attention Mechanism |
CN111640500A (en) * | 2020-04-30 | 2020-09-08 | 西北大学 | Brain age prediction model and method based on bilinear fusion and attention mechanism |
CN113298286A (en) * | 2021-03-31 | 2021-08-24 | 捷佳润科技集团股份有限公司 | Machine learning-based pitaya marketing time prediction method |
CN113545751A (en) * | 2021-07-20 | 2021-10-26 | 遵义医科大学附属医院 | Machine learning-based brain age assessment method for Rolandic epileptic children |
CN114881943B (en) * | 2022-04-22 | 2023-05-16 | 深圳市铱硙医疗科技有限公司 | Brain age prediction method, device, equipment and storage medium based on artificial intelligence |
CN114881943A (en) * | 2022-04-22 | 2022-08-09 | 深圳市铱硙医疗科技有限公司 | Artificial intelligence-based brain age prediction method, device, equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110555828A (en) | Brain age prediction method and device based on 3D convolutional neural network | |
Liu et al. | Attribute-aware face aging with wavelet-based generative adversarial networks | |
WO2020215985A1 (en) | Medical image segmentation method and device, electronic device and storage medium | |
EP4002198A1 (en) | Posture acquisition method and device, and key point coordinate positioning model training method and device | |
CN111754596B (en) | Editing model generation method, device, equipment and medium for editing face image | |
CN110363760B (en) | Computer system for recognizing medical images | |
CN111428071B (en) | Zero-sample cross-modal retrieval method based on multi-modal feature synthesis | |
KR20190028422A (en) | Systems and methods for automatic detection, localization, and semantic segmentation of anatomical objects | |
US10521264B2 (en) | Data processing architecture for improved data flow | |
WO2020024395A1 (en) | Fatigue driving detection method and apparatus, computer device, and storage medium | |
CN112232293A (en) | Image processing model training method, image processing method and related equipment | |
CN111429421A (en) | Model generation method, medical image segmentation method, device, equipment and medium | |
CN110889450B (en) | Super-parameter tuning and model construction method and device | |
WO2021120961A1 (en) | Brain addiction structure map evaluation method and apparatus | |
CN111368911B (en) | Image classification method and device and computer readable storage medium | |
CN113159200B (en) | Object analysis method, device and storage medium | |
CN111259954A (en) | Hyperspectral traditional Chinese medicine tongue coating and tongue quality classification method based on D-Resnet | |
CN111462060A (en) | Method and device for detecting standard section image in fetal ultrasonic image | |
WO2023108873A1 (en) | Brain network and brain addiction connection calculation method and apparatus | |
US20230196578A1 (en) | Brain network and brain addictive connectivity computing method and device | |
CN114896067A (en) | Automatic generation method and device of task request information, computer equipment and medium | |
CN114241240A (en) | Method and device for classifying brain images, electronic equipment and storage medium | |
CN113642360A (en) | Behavior timing method and device, electronic equipment and storage medium | |
CN110599444A (en) | Device, system and non-transitory readable storage medium for predicting fractional flow reserve of a vessel tree | |
CN111275059B (en) | Image processing method and device and computer readable storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191210 |