Brain age prediction method based on ensemble learning and structural magnetic resonance imaging, electronic equipment and storage medium

Abstract

The invention relates to the field of neuroscience and ensemble learning, in particular to a brain age prediction method based on ensemble learning and structural magnetic resonance imaging, which comprises the following steps: collecting sMRI data and corresponding age information of a healthy adult subject; preprocessing the sMRI data, and randomly dividing the preprocessed sMRI data into a training set, a verification set and a test set; constructing a self-adaptive integrated learning model, and training the self-adaptive integrated learning model; inputting sMRI data to be predicted into the trained adaptive integrated learning model, and outputting a final brain age prediction result; and evaluating the final brain age prediction result through a mean absolute error, a root-mean-square error and a decision coefficient. Compared with the prior art, the method has the advantages that the brain age prediction precision and the model generalization ability can be improved, and the like.

Description

Technical Field

[0001] The present invention relates to the fields of neuroscience and integrated learning, and in particular to a brain age prediction method based on integrated learning and structural magnetic resonance imaging, an electronic device and a storage medium. Background Art

[0002] As people age, the human brain gradually enters the aging stage, and its structure, morphology and function will change accordingly, accompanied by the decline of cognitive ability, memory and calculation ability. Brain age is an important indicator for assessing brain health and can be used as a biomarker for judging human brain aging. The brain age of a normal person is close to the actual physiological age. When there is a neurodegenerative disease in the human brain, the brain will age faster, which makes the brain age greater than the actual physiological age. Therefore, accurately predicting brain age is of great significance for assessing the health of the human brain.

[0003] In recent years, the rapid development of medical imaging technology and machine learning has made it possible to analyze neuroimaging data in new ways, and it is expected to extract key features that characterize brain age. Traditional machine learning methods such as support vector regression, correlation vector regression, and Gaussian process regression have been used to predict brain age, but they all have problems such as low accuracy and cumbersome modeling process. Deep learning methods have shown certain advantages in brain age prediction and significantly improved the prediction accuracy, but there is a problem of low model generalization, which is difficult to meet clinical needs. Therefore, the present invention provides a brain age prediction method based on ensemble learning and structural magnetic resonance imaging. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a brain age prediction method based on ensemble learning and structural magnetic resonance imaging, an electronic device and a storage medium.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] In one aspect, the present invention provides a method for predicting brain age based on ensemble learning and structural magnetic resonance imaging, comprising the following steps:

[0007] Step 1: Collect sMRI data and corresponding age information of healthy adult subjects;

[0008] Step 2, preprocessing the sMRI data, and randomly dividing the preprocessed sMRI data into a training set, a validation set, and a test set according to a set ratio;

[0009] Step 3: construct an adaptive ensemble learning model, wherein the adaptive ensemble learning model includes multiple base learners and multiple secondary learners; input the training set into the multiple base learners for training, use the validation set to train the secondary learner, and use the test set to evaluate the adaptive ensemble learning model;

[0010] Step 4, inputting the sMRI data to be predicted into the trained adaptive ensemble learning model, and outputting the final brain age prediction result;

[0011] Step 5: Evaluate the final brain age prediction result by using mean absolute error, root mean square error and determination coefficient.

[0012] Furthermore, the preprocessing in step 2 includes: noise reduction, segmentation, registration and spatial normalization.

[0013] Furthermore, the setting ratio in step 2 is 8:1:1.

[0014] Furthermore, the multiple base learners in step 3 use 3D-DenseNet, 3D-ResNeXt and 3D-Inception v4 neural networks to generate three base learners.

[0015] Furthermore, the training set is input into the multiple base learners for training, and the brain age prediction results of the first stage are outputted, and the average value of the brain age prediction results of the first stage is calculated, and the validation set is divided into groups of different age ranges based on the average value; wherein each group is 5 years old as a layer, and there is overlap in the age ranges of each group.

[0016] Furthermore, the specific division method of the grouping is as follows:

[0017] Group 1: less than 25 years old;

[0018] Group 2: 20 to 30 years old;

[0019] Group 3: 25 to 35 years old;

[0020] Group 4: 30 to 40 years old;

[0021] Group 5: 35 to 45 years old;

[0022] Group 6: 40 to 50 years old;

[0023] Group 7: 45 to 55 years old;

[0024] Group 8: 50 to 60 years old;

[0025] Group 9: 55 to 65 years old;

[0026] Group 10: 60 to 70 years old;

[0027] Group 11: 65 to 75 years old;

[0028] Group 12: 70 to 80 years old;

[0029] Group 13: 75 to 85 years old;

[0030] Group 14: over 80 years old.

[0031] Furthermore, according to the multiple groups of the validation set, a corresponding number of linear regression models are selected as secondary learners, and the secondary learners are trained with the data of the groups.

[0032] Furthermore, the brain age prediction result of the first stage is used as the input of the secondary learner, the brain age prediction result output by the second stage learner is compared with the actual age of healthy adult subjects, and the parameters of the secondary learner are adjusted.

[0033] In a second aspect, the present invention provides an electronic device comprising a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes any one of the methods described above.

[0034] In a third aspect, the present invention provides a storage medium having a program stored thereon, wherein the program, when executed, implements any of the methods described above.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. The present invention performs adaptive ensemble learning on the results output by three base learners. This method not only benefits from the fact that the ensemble learning results are better than the prediction results of a single model, but also takes into account the different sensitivities of different base learners to data of different age groups, thereby achieving good deep ensemble learning effects and improving prediction accuracy. In the test set, the average absolute error reaches 2.75 years and the determination coefficient reaches 0.966.

[0037] 2. The present invention uses a multi-center dataset to train an adaptive ensemble learning model, which improves the prediction accuracy and generalization ability of brain age results. Since the multi-center dataset contains samples from multiple sources, it can better reflect individual differences from various sources, which helps to improve the accuracy and generalization ability of the model. In addition, this multi-center dataset can also effectively avoid the bias of a single source, so that the model can be better generalized to a wider range of data.

[0038] 3. The age ranges of the various groups in the validation set of the present invention overlap, which expands the range of brain age of the data used by the integrated model. Compared with the grouping method without cross-grouping, this allows data at the group boundary to select a more appropriate adaptive integrated learning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of the brain age prediction method based on ensemble learning and structural magnetic resonance imaging of the present invention;

[0040] Figure 2 A flowchart of the present invention for uniformly preprocessing raw sMRI data;

[0041] Figure 3 This is the algorithm flow for predicting brain age using the adaptive ensemble learning model of the present invention. DETAILED DESCRIPTION

[0042] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0043] Example

[0044] This embodiment provides a brain age prediction method based on ensemble learning and structural magnetic resonance imaging (sMRI), which first collects sMRI data of healthy adults from multiple centers and corresponding age information, and then preprocesses the sMRI data, including noise reduction, segmentation, registration and spatial standardization, and then uses the preprocessed sMRI data to train 3D-DenseNet, 3D-ResNeXt and 3D-Inception v4 neural networks, respectively, to generate multiple base learners, and then adaptively ensemble learning is performed on the output results of the multiple base learners generated above to obtain a prediction result: the brain age of a healthy adult. The effect of the prediction result is evaluated by indicators such as mean absolute error, root mean square error and determination coefficient. This method uses sMRI data from multiple centers to train an adaptive ensemble learning model (also referred to as a brain age prediction model), and uses an adaptive ensemble learning algorithm based on the Stacking strategy to combine the outputs of the three prediction models, effectively improving the accuracy and generalization ability of the brain age prediction model.

[0045] The specific implementation steps are as follows: Figure 1 As shown, including the following:

[0046] Step S1, collecting sMRI data of healthy adult subjects.

[0047] In this embodiment, the data are obtained by collecting the following datasets: Southwest University Brain Imaging Dataset (SALD), China Connectome Project (CHCP), IXI dataset, NKI dataset, DLBS dataset, ICBM dataset, CoRR dataset and ADNI dataset. Among them, there are 3479 sMRI data of healthy adults, with an age range of 18 to 90 years old.

[0048] Step S2, preprocessing the sMRI data, and randomly dividing the preprocessed sMRI data into a training set, a validation set, and a test set according to a ratio of 8:1:1.

[0049] In this embodiment, preprocessing the sMRI data refers to preprocessing the sMRI data using the CAT12 computational anatomy toolbox, such as Figure 2 As shown, the image is denoised, segmented, registered and spatially normalized to obtain a 3D sMRI image of size 113×137×113.

[0050] Step S3, constructing an adaptive ensemble learning model, the adaptive ensemble learning model includes three base learners and fourteen secondary learners. The training set is input into the multiple base learners for training, the validation set is used to train the secondary learners, and the test set is used to evaluate the adaptive ensemble learning model.

[0051] Specifically, the data of the training set are used to train three base learners generated by 3D-DenseNet, 3D-ResNeXt and 3D-Inception v4 neural networks, and the output is the brain age prediction result of the first stage.

[0052] The brain age prediction results of the first stage output by the three base learners are applied to the secondary learner.

[0053] According to the brain age prediction results of the first stage, the validation set is grouped. Specifically, the average value of the brain age prediction results of the first stage is calculated, and the validation set data is divided into groups of different age ranges based on this average value, where each group is 5 years old and the age ranges of each group overlap.

[0054] Specifically, these groups are divided into 14 groups, which are:

[0055] Group 1: less than 25 years old;

[0056] Group 2: 20 to 30 years old;

[0057] Group 3: 25 to 35 years old;

[0058] Group 4: 30 to 40 years old;

[0059] Group 5: 35 to 45 years old;

[0060] Group 6: 40 to 50 years old;

[0061] Group 7: 45 to 55 years old;

[0062] Group 8: 50 to 60 years old;

[0063] Group 9: 55 to 65 years old;

[0064] Group 10: 60 to 70 years old;

[0065] Group 11: 65 to 75 years old;

[0066] Group 12: 70 to 80 years old;

[0067] Group 13: 75 to 85 years old;

[0068] Group 14: over 80 years old.

[0069] In this embodiment, for multiple groups of the validation set, fourteen corresponding linear regression models are selected as secondary learners, and the secondary learners are trained using the validation sets of the above groups. The specific process is: on the validation set, a linear regression model is used as a secondary learner, and each group of data is used to train the secondary learner, that is, for each group of data, the brain age outputs of the three base learners are used as the input of the secondary learner, the output results of the secondary learner are compared with the actual age of the subject, and the parameters of the linear regression model are adjusted according to the comparison results, so as to achieve the purpose of optimizing the prediction effect.

[0070] Step S4, inputting the sMRI data of the test set into the trained adaptive ensemble learning model, specifically, obtaining the first-stage brain age prediction result through the base learner, inputting the first-stage brain age prediction result into the secondary learner, and outputting the final brain age prediction result.

[0071] Specifically, Figure 3 As shown, in this embodiment, when the test set data is input, the average values ​​of the brain age predicted by the three base learners are first calculated, and the corresponding secondary learner is selected according to the grouping range of the average value, that is, if the average value is less than 22.5 years old, the secondary learner with a grouping age range of less than 25 years old is used, if the average value is between 22.5 and 27.5 years old, the secondary learner with a grouping age range of 20 to 30 years old is used... and so on, until the average value is greater than or equal to 82.5 years old, the secondary learner with a grouping age range of more than 80 years old is used. The brain age output by the secondary learner is the final output result of this embodiment.

[0072] Step S5: Evaluate the final brain age prediction result by using mean absolute error, root mean square error and determination coefficient.

[0073] Specifically, during the test, the accuracy of brain age prediction was obtained by the mean absolute error and root mean square error evaluation indicators, and the degree of fit of brain age prediction in the test set was evaluated by the coefficient of determination.

[0074] In this embodiment, the prediction results of the test set are taken as an example. The brain age prediction is performed on 347 pre-processed 3D structural magnetic resonance imaging data of the test set, and the mean absolute error, root mean square error and determination coefficient are used to evaluate the brain age prediction results. The formula is as follows:

[0075]

[0076]

[0077]

[0078] Among them, MAE represents the mean absolute error of the brain age prediction results, RMSE represents the root mean square error of the brain age prediction results, and R 2 represents the coefficient of determination of the brain age prediction result, N represents the number of predicted subjects, x i represents the input image, f(x i ) represents the result of the model predicting brain age, y i represents the actual age of the i-th subject, Represents the average value of actual age. By predicting and evaluating the 347 imaging data of the test set of this embodiment, the final result is: the mean absolute error reaches 2.75 years, the root mean square error reaches 3.65 years, and the determination coefficient reaches 0.966. This result shows that the brain age prediction method based on ensemble learning and structural magnetic resonance imaging provided by this embodiment has a relatively precise accuracy.

[0079] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0080] In this embodiment, an electronic device is also provided, including one or more processors, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the brain age prediction method based on integrated learning and structural magnetic resonance imaging as described above.

[0081] This embodiment also provides a storage medium, including: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc., which can store program codes, and a program is stored thereon. When the program is executed, the brain age prediction method based on integrated learning and structural magnetic resonance imaging as described above is implemented.

[0082] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A brain age prediction method based on ensemble learning and structural magnetic resonance imaging, It is characterized in that The following steps are involved: Step 1: Collect sMRI data and corresponding age information of healthy adult subjects; Step 2, preprocessing the sMRI data, and randomly dividing the preprocessed sMRI data into a training set, a validation set, and a test set according to a set ratio; Step 3: construct an adaptive ensemble learning model, wherein the adaptive ensemble learning model includes multiple base learners and multiple secondary learners; input the training set into the multiple base learners for training, use the validation set to train the secondary learner, and use the test set to evaluate the adaptive ensemble learning model; Step 4, inputting the sMRI data to be predicted into the trained adaptive ensemble learning model, and outputting the final brain age prediction result; Step 5: Evaluate the final brain age prediction result by using mean absolute error, root mean square error and determination coefficient.

2. A method for predicting brain age based on ensemble learning and structural magnetic resonance imaging according to claim 1, It is characterized in that The preprocessing in step 2 includes: noise reduction, segmentation, registration and spatial normalization.

3. The method for predicting brain age based on ensemble learning and structural magnetic resonance imaging according to claim 1, It is characterized in that The setting ratio in step 2 is 8:1:

1.

4. The method for predicting brain age based on ensemble learning and structural magnetic resonance imaging according to claim 1, It is characterized in that The multiple base learners in step 3 use 3D-DenseNet, 3D-ResNeXt and 3D-Inception v4 neural networks to generate three base learners.

5. The method for predicting brain age based on ensemble learning and structural magnetic resonance imaging according to claim 1, It is characterized in that The training set is input into the multiple base learners for training, and the brain age prediction results of the first stage are outputted, and the average value of the brain age prediction results of the first stage is calculated. The validation set is divided into groups with different age ranges based on the average value; wherein each group is 5 years old as a layer, and there is overlap in the age ranges of each group.

6. A method for predicting brain age based on ensemble learning and structural magnetic resonance imaging according to claim 5, It is characterized in that The specific division method of the grouping is as follows: Group 1: less than 25 years old; Group 2: 20 to 30 years old; Group 3: 25 to 35 years old; Group 4: 30 to 40 years old; Group 5: 35 to 45 years old; Group 6: 40 to 50 years old; Group 7: 45 to 55 years old; Group 8: 50 to 60 years old; Group 9: 55 to 65 years old; Group 10: 60 to 70 years old; Group 11: 65 to 75 years old; Group 12: 70 to 80 years old; Group 13: 75 to 85 years old; Group 14: over 80 years old.

7. The method for predicting brain age based on ensemble learning and structural magnetic resonance imaging according to claim 6, It is characterized in that According to the multiple groups of the validation set, a corresponding number of linear regression models are selected as secondary learners, and the secondary learners are trained with the data of the groups.

8. The method for predicting brain age based on ensemble learning and structural magnetic resonance imaging according to claim 7, It is characterized in that The brain age prediction result of the first stage is used as the input of the secondary learner, the brain age prediction result output by the second stage learner is compared with the actual age of healthy adult subjects, and the parameters of the secondary learner are adjusted.

9. An electronic device, It is characterized in that It includes a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 8.

10. A storage medium having a program stored thereon, It is characterized in that When the program is executed, the method according to any one of claims 1 to 8 is implemented.

Images