A Method for Extracting Magnetic Resonance Image Features for Predicting the Progression of Alzheimer's Disease

Abstract

A method for extracting magnetic resonance image features for predicting the progression of Alzheimer's disease, comprising: determining magnetic resonance images of selected brain characteristic regions, quantitatively extracting corresponding features for training and screening, and predicting the progression of Alzheimer's disease by combining corresponding clinical features and gene features of Alzheimer's disease. It ensures the reliability of data correlation and makes full use of magnetic resonance image feature data, overcoming the limitation of the existing Alzheimer's prediction model with less judgment data.

Description

Technical Field

[0001] The present invention relates to the field of computers, and particularly to a method for extracting magnetic resonance image features for predicting the progression of Alzheimer's disease. Background Art

[0002] Alzheimer's disease is a chronic and fatal neurodegenerative disease, and its main clinical manifestations include cognitive dysfunction and memory dysfunction, etc. It is very necessary to develop a machine learning model that can identify potential Alzheimer's patients at an early stage by fusing various data such as clinical, genetic, and image data, so as to achieve early detection and early treatment.

[0003] Existing methods for predicting the progression of Alzheimer's disease have different ways of using data, especially for the processing of MRI. MRI uses the magnetic resonance phenomenon to obtain electromagnetic signals from the human brain and reconstructs brain information. There are huge differences in the morphological manifestations of MRI between Alzheimer's patients and normal people. Therefore, MRI contains a large amount of valuable information. However, when existing Alzheimer's prediction models select MRI features, most of them use software to quantify and then simply select a few features from them through correlation for training. The limitations of doing so are as follows: the data volume must be large enough to ensure that the correlation is reliable, and the performance on small datasets is very poor; directly discarding the unselected features will lead to insufficient utilization of MRI.

[0004] Therefore, there is an urgent need for a method for extracting magnetic resonance image features for predicting the progression of Alzheimer's disease that can make full use of MRI data and have good performance on small datasets. Summary of the Invention

[0005] In order to solve the above problems, the present invention provides a method for extracting magnetic resonance image features for predicting the progression of Alzheimer's disease, which can not only ensure the full utilization of MRI data but also have good performance on small datasets. At the same time, a corresponding system for extracting magnetic resonance image features, a system for magnetic resonance image features for predicting the progression of Alzheimer's disease, and a method for predicting the progression of Alzheimer's disease using magnetic resonance image features are provided.

[0006] One of the purposes of the present invention is to provide a system for extracting magnetic resonance image features, and the specific technical solution is as follows:

[0007] A method for extracting magnetic resonance image features for predicting the progression of Alzheimer's disease uses a magnetic resonance image original feature screening module to quantify, determine, and select features of magnetic resonance images. A magnetic resonance image feature statistical integration module is used to integrate and statistically analyze features. The specific steps are as follows:

[0008] S1: Software quantitative nuclear magnetic resonance imaging;

[0009] S2: Determine candidate regions and candidate features, determine the brain regions strongly correlated with the progression of Alzheimer's disease as candidate regions, and the candidate features are selected from the candidate regions;

[0010] S3: Score the candidate regions and select the feature with the highest score in each region;

[0011] S4: Perform secondary screening with a threshold to screen out the candidate features with scores higher than the threshold in the candidate regions;

[0012] S5: Construction and calculation of statistical features, calculate the statistics of each original feature on different brain regions, and the statistics are mathematical expressions representing the data distribution;

[0013] S6: Screen out the statistical feature with the highest score.

[0014] Further, step S1 specifically includes the following steps:

[0015] 1) Use FreeSurfer to label the regions on the cortical surface and the subcortical brain structures;

[0016] 2) Use ANTs for brain volume extraction, segmentation, and registration;

[0017] 3) Use Mindboggle software to automatically perform shape analysis on the labels and features extracted from FreeSurfer and ANTs.

[0018] Further, the Mindboggle software in step 3) will finally calculate the volumes of all labeled regions, the volumes of all labeled cortical regions, and various shape features including surface area, thickness, and mean curvature for each cortical surface label.

[0019] Further, the specific method of step S2 includes:

[0020] a: Select 8 brain regions strongly correlated with the progression of Alzheimer's disease as candidate regions;

[0021] b: Send each feature of the candidate regions together with clinical features and gene features into a support vector machine model with the same parameters for training to obtain candidate features.

[0022] Further, the candidate regions in step S2 include the prefrontal cingulate, inferior parietal lobule, inferior temporal lobe, lateral occipital lobe, lingual muscle, middle temporal lobe, superior marginal plate, and insula.

[0023] Further, 50 - 100 features are selected from each of the candidate regions as candidate features.

[0024] Further, in step S3, the change amount of the model evaluation index is used as the candidate region score, and the scoring steps are as follows:

[0025] (1) Divide the clinical and genetic data provided by the ADNI dataset into a training set and a validation set according to a ratio of 3:1, and then train a support vector machine model with the relatively best performance by using grid search and cross-validation;

[0026] (2) Add each candidate nuclear magnetic resonance feature to the training set and the validation set, retrain without changing the model parameters, and use the difference between the evaluation index of the trained model and the evaluation index of the support vector machine model described in (1) as the score of this feature;

[0027] (3) Only retain the feature with the highest score in each region.

[0028] Further, the support vector machine model includes a classification model and a regression model; the evaluation index of the classification model is the area under the receiver operating characteristic curve, and the evaluation index of the regression model is the Pearson correlation coefficient.

[0029] Further, in step S4, quadratic threshold screening is added, and the threshold is greater than 1%.

[0030] Further, step S5 includes:

[0031] a) Calculate the statistical features of all brain regions;

[0032] b) Select the statistical features with the best performance on the validation set from the above statistical features by non-repetitive random sampling as the finally integrated statistical features.

[0033] Further, the selected statistical features include the maximum value, the minimum value, the sum, the range, and the standard deviation.

[0034] Further, the specific steps of step S6 are as follows:

[0035] (a) Select k statistical features from step S5, and send them together with the original nuclear magnetic resonance features, clinical features, and genetic features in S2 into a support vector machine model with the same parameters as in S2 for training, and record the evaluation index of the model;

[0036] (b) Repeat step (a) 3000 times;

[0037] (c) Select the k statistical features that maximize the model evaluation index as the finally selected integrated features;

[0038] Among them, the value range of k is 3 - 8.

[0039] The second object of the present invention is to provide a nuclear magnetic resonance image feature extraction system for predicting the progression of Alzheimer's disease. The specific technical solution is as follows:

[0040] A nuclear magnetic resonance image feature extraction system for predicting the progression of Alzheimer's disease includes a screening module and a feature statistical integration module. The screening module quantifies, determines, and selects features from nuclear magnetic resonance images, determines candidate regions, and selects candidate features. The candidate regions are brain regions strongly related to the progression of Alzheimer's disease, and the candidate features are selected from the candidate regions and obtained through training with a support vector machine model. The integration module is used for the construction and calculation of statistical features, calculates the statistics of each original feature in different brain regions, where the statistics are mathematical expressions representing the data distribution, screens out the statistical features with the highest scores, and integrates them to obtain macroscopic features.

[0041] Further, the statistics include maximum value, minimum value, sum, range, and standard deviation.

[0042] Further, the macroscopic features include nuclear magnetic resonance image features, clinical features, and gene features related to Alzheimer's disease.

[0043] Further, the candidate regions are 8 brain regions strongly related to the progression of Alzheimer's disease, and the candidate features are selected from the brain regions and sent together with clinical features and gene features into a support vector machine model with the same parameters for training.

[0044] Further, the brain regions are the prefrontal cingulate, inferior parietal lobule, inferior temporal lobe, lateral occipital lobe, genioglossus, middle temporal lobe, superior marginal plate, and insula.

[0045] Further, 50 features are selected from each of the brain regions as candidate features.

[0046] Further, the screening module can perform secondary threshold screening, and the threshold is greater than 1%.

[0047] Further, the screening module screens out 5 original features of the classification model and 7 original features of the regression model.

[0048] The third object of the present invention is to provide a nuclear magnetic resonance image feature system for predicting the progression of Alzheimer's disease. The specific technical solution is as follows:

[0049] A nuclear magnetic resonance image feature system for predicting the progression of Alzheimer's disease includes a nuclear magnetic resonance image feature extraction system, a nuclear magnetic resonance image feature comparison module, and a result output module. The nuclear magnetic resonance image feature comparison module is used to compare and analyze the newly input nuclear magnetic resonance image features with the feature model. The result output module outputs the comparison and analysis results of the nuclear magnetic resonance image feature comparison module.

[0050] Furthermore, the nuclear magnetic resonance image feature extraction system includes a screening module and a feature statistical integration module. The screening module quantifies, determines, and selects features from nuclear magnetic resonance images, determines candidate regions, and selects candidate features. The candidate regions are brain regions strongly correlated with the progression of Alzheimer's disease, and the candidate features are selected from the candidate regions and obtained through training with a support vector machine model. The integration module is used for the construction and calculation of statistical features, calculates the statistics of each original feature in different brain regions, where the statistics are mathematical expressions representing the data distribution, screens out the statistical features with the highest scores, and integrates them to obtain macroscopic features.

[0051] Furthermore, the statistics include maximum value, minimum value, sum, range, and standard deviation.

[0052] Furthermore, the macroscopic features include nuclear magnetic resonance image features, clinical features, and gene features related to Alzheimer's disease.

[0053] Furthermore, the candidate regions are 8 brain regions strongly correlated with the progression of Alzheimer's disease, and the candidate features are selected from these brain regions and sent together with clinical features and gene features into a support vector machine model with the same parameters for training.

[0054] Furthermore, the brain regions are the prefrontal cingulate, inferior parietal lobule, inferior temporal lobe, lateral occipital lobe, genioglossus, middle temporal lobe, superior marginal plate, and insula.

[0055] Furthermore, 50 features are selected from each of the brain regions as candidate features.

[0056] Furthermore, the screening module can perform secondary threshold screening, and the threshold is greater than 1%.

[0057] Furthermore, the screening module screens out 5 original features for the classification model and 7 original features for the regression model.

[0058] The fourth object of the present invention is to provide a method for predicting the progression of Alzheimer's disease using nuclear magnetic resonance image features. The specific technical solution is as follows:

[0059] A method for predicting the progression of Alzheimer's disease using nuclear magnetic resonance image features includes the following steps:

[0060] Ⅰ: Establish the nuclear magnetic resonance image feature system related to Alzheimer's disease described in the above solution;

[0061] Ⅱ: Quantify the brain nuclear magnetic resonance image of the person to be predicted and input it into the nuclear magnetic resonance image feature system described in step Ⅰ for feature matching.

[0062] Furthermore, the nuclear magnetic resonance (NMR) image feature system in step I includes an NMR image feature extraction subsystem, an NMR image feature comparison module, and a result output module. It is characterized in that the NMR image feature extraction subsystem is used to extract the original features of the NMR image and establish a feature model; the NMR image feature comparison module is used to compare and analyze the newly input NMR image features with the feature model; and the result output module outputs the comparison and analysis results of the NMR image feature comparison module.

[0063] Furthermore, the NMR image feature extraction subsystem includes a screening module for the original features of the NMR image and a statistical integration module for NMR features. The screening module determines candidate regions and selects candidate features with the help of prior knowledge; the integration module constructs various statistics to represent the macroscopic features of NMR.

[0064] Furthermore, the screening module can perform secondary threshold screening, and the threshold is greater than 1%.

[0065] Furthermore, the candidate regions are 8 brain regions strongly related to the progression of Alzheimer's disease, and the brain regions are the prefrontal cingulate, inferior parietal lobule, inferior temporal lobe, lateral occipital lobe, genioglossus, middle temporal lobe, superior marginal plate, and insula.

[0066] Furthermore, the candidate features come from each of the brain regions, and 50 features are selected as candidate features for each brain region.

[0067] Furthermore, the macroscopic features include NMR image features, clinical features, and gene features related to Alzheimer's disease.

[0068] Furthermore, the NMR image original feature screening module quantifies, determines, and selects the features of the NMR image; the NMR image feature statistical integration module is used to integrate statistical features.

[0069] Furthermore, the statistical features include the maximum value, minimum value, sum, range, and standard deviation.

[0070] The advantages of the present invention are as follows: By screening the MRI data with prior knowledge, performing secondary threshold screening, and integrating the MRI data by constructing statistical features, the present invention can maximize the extraction of the effective information contained in the MRI data, solve the problems of overfitting and insufficient feature expressiveness caused by simply using correlation to screen features, and improve the prediction accuracy of the progression of Alzheimer's disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a flowchart of the method for extracting NMR image features for predicting the progression of Alzheimer's disease according to the present invention

[0072] Figure 2 Schematic diagram of the nuclear magnetic resonance image feature system for predicting the progression of Alzheimer's disease in the present invention Detailed implementation manners

[0073] The present invention will be further described in detail below through examples and drawings. It should be understood that the specific examples described herein are only used to explain the present invention and are not used to limit the present invention. Those skilled in the art should understand that without departing from the structural idea and scope of use of the present invention, details and forms of the technical solution of the present invention can be modified or replaced, but these modifications and replacements all fall within the protection scope of the present invention.

[0074] As a disease lacking specific detection indicators, it is often difficult for clinicians to accurately identify Alzheimer's disease at an early stage. However, individuals in different stages will show differences in various morphological manifestations of nuclear magnetic resonance images. It is necessary to use the information contained in nuclear magnetic resonance images to predict individuals at an early stage so that doctors can treat potential patients at an early stage and reduce the risk of their conversion into Alzheimer's disease patients.

[0075] Example 1

[0076] Combined with Figure 1 , a method for extracting nuclear magnetic resonance image features for predicting the progression of Alzheimer's disease includes five processes, namely determining candidate regions and candidate features with the help of prior knowledge, selecting the feature with the highest score in each region, performing secondary threshold screening to reduce the influence of randomness, calculating the statistical characteristics of each feature on all regions as integrated features, and randomly sampling without repetition to select k statistical features. The candidate regions are respectively the prefrontal cingulate, inferior parietal lobule, inferior temporal lobe, lateral occipital lobe, genioglossus muscle, middle temporal lobe, superior marginal plate and insula; the basic model for feature scoring is a support vector machine model with a linear kernel; the threshold value is 1%; the statistics for integration are the maximum value, minimum value, sum, range and standard deviation; the value range of the finally selected number k of statistical features can be between 3 and 8, and we select it to be 5.

[0077] Combined with Figure 1 , a method for extracting nuclear magnetic resonance image features for predicting the progression of Alzheimer's disease includes the following steps:

[0078] Step 1: Quantify the original MRI features from the pictures using software;

[0079] In this embodiment, FreeSurfer, Advanced Normalization Tools and Mindboggle are used, and other software with similar functions can also be used for quantification.

[0080] Step 2: Determine the candidate regions and candidate features;

[0081] Specifically, from the 2,150 features quantified by the software, 400 features are selected as candidate features, including 50 features for each of the 8 brain regions: the prefrontal cingulate, inferior parietal lobule, inferior temporal lobe, lateral occipital lobe, genioglossus muscle, middle temporal lobe, superior marginal plate, and insula.

[0082] Step 3: Select the feature with the highest score in each region;

[0083] The specific steps for scoring are as follows: (1) Divide the clinical and genetic data provided by the ADNI dataset into a training set and a validation set at a ratio of 3:1. Then, use grid search and cross-validation to train a support vector machine model with relatively optimal performance. In this embodiment, the goal of the classification model is to predict whether the MMSE score of an individual will significantly decrease two years later (individuals with a significant decrease are highly likely to develop Alzheimer's disease); the goal of the regression model is to predict the MMSE score of an individual two years later. The parameters of both the classification and regression models are: linear kernel, C = 10, gamma = 0.1. The evaluation index of the classification model is the area under the receiver operating characteristic curve (AUROC), and the evaluation index of the regression model is the Pearson correlation coefficient. (2) Add each candidate MRI feature to the training set and the validation set, and retrain without changing the model parameters. Use the difference between the evaluation index of the trained model and the evaluation index of the model in (1) as the score of this feature. (3) Only retain the feature with the highest score in these 8 regions.

[0084] Step 4: Conduct a secondary threshold screening;

[0085] The threshold is set at 1%, and only features with scores higher than this score are retained. In this embodiment, the threshold screening filters out 3 features in the classification model and 1 feature in the classification model. Among them, the classification model finally retains 5 original features, and the regression model retains 7 original features.

[0086] Step 5: Calculate the statistical properties of each feature across all regions;

[0087] First, the 2150 original features altogether include 43 brain regions, and each brain region contains 50 original features. Then, several statistical measures of these 50 features on the 43 brain regions are calculated. The statistical measure can be any mathematical expression that can represent the data distribution. In this embodiment, the maximum value, minimum value, sum, range, and standard deviation of these 50 features on all brain regions are calculated, totaling 250 statistical measures. The principle of introducing statistical measures to improve the utilization rate of MRI data and the model accuracy is as follows: (1) Each statistical measure synthesizes the information of each brain region, which can make up for the decrease in the utilization rate of MRI data caused by directly discarding most brain regions in Steps 1 to 3. (2) Each statistical feature may be strongly correlated with the development of Alzheimer's disease. For example, the sum of the surface areas can reflect the atrophy of the entire brain, and it has been medically proven that the brains of Alzheimer's patients atrophy significantly compared to the normal population. This method can also be used to find those indicators that have a great impact on the development of Alzheimer's disease.

[0088] Step 6: Randomly select k statistical features without repetition.

[0089] (1) Select k features from the 250 statistical features and train a support vector machine model with the same parameters as the previous model together with the previous clinical, genetic, and original MRI features, and record the evaluation indicators of the model. In this embodiment, k is set to 5.

[0090] (2) Repeat operation (1) 3000 times.

[0091] (3) Select the 5 statistical features that maximize the model evaluation indicators as the finally selected integrated features.

[0092] After Step 5, the feature extraction of magnetic resonance images for predicting the progression of Alzheimer's disease is completed. The core of this method is to select a single-digit number of quantization features for magnetic resonance images to reflect the original information, and select a single-digit number of statistical features to improve the data utilization rate and discover potential indicators affecting the progression of Alzheimer's disease.

[0093] Embodiment 2

[0094] A method for extracting magnetic resonance image features for predicting the progression of Alzheimer's disease includes three links, namely software processing and quantization, original feature screening and secondary threshold screening, and construction and screening of integrated features.

[0095] Among them, the software processing and quantization is to perform labeling and quantization processing on the MRI using three professional medical image software, namely FreeSurfer, Advanced NormalizationTools (ANTs), and Mindboggle.

[0096] The original feature screening described above determines the most valuable candidate regions based on prior knowledge to narrow the screening scope. The threshold secondary screening sets a threshold for re-filtering to reduce the impact of randomness and prevent the introduction of noise.

[0097] The principle of the construction and screening of the integrated features is to increase the information utilization rate of MRI by calculating the global statistics of the MRI data. The statistics can be any expression that can represent the characteristics of the MRI, such as the maximum value, range, sum, average value, standard deviation, etc.

[0098] A method for extracting magnetic resonance image features for predicting the progression of Alzheimer's disease, comprising the following steps:

[0099] S1: Quantify the magnetic resonance image using professional software;

[0100] S2: Determine the candidate regions and candidate features;

[0101] S3: Score the candidate regions and select the features with the highest scores for each region;

[0102] S4: Perform secondary screening using a threshold;

[0103] S5: Construct and calculate the statistical features;

[0104] S6: Screen the statistical features with expressiveness.

[0105] Furthermore, for step S1, FreeSurfer, Advanced Normalization Tools (ANTs), and Mindboggle are required to preprocess the image. Specifically: (1) Use FreeSurfer to label the regions on the cortical surface and the subcortical brain structures; (2) Use ANTs for brain volume extraction, segmentation, and registration; (3) Use Mindboggle software to automatically perform shape analysis on the labels and features extracted from FreeSurfer and ANTs. Among them, Mindboggle will finally calculate (1) the volume of all labeled regions; (2) the volume of all labeled cortical regions; (3) various shape features of each cortical surface label, including surface area, thickness, average curvature, etc.

[0106] Further, in step S2, the method of introducing prior knowledge is adopted to help screen out expressive original features. According to relevant research, 8 brain regions that are strongly correlated with the progression of Alzheimer's disease in morphological manifestations are selected as candidate regions. Each feature of these 8 regions, together with clinical and genetic features, is fed into a support vector machine model with the same parameters for training. In step S3, the change amount of the model evaluation index is used as their score, and the feature with the highest score in each region is selected as the candidate feature in that region. To reduce the influence of randomness, in step S4, an empirical threshold n is used to perform a secondary screening on the features of the 8 regions, and only the features with a score higher than n are retained.

[0107] Further, fewer than 8 out of thousands of original features are retained, and most of the MRI information is actually lost. To improve the utilization rate of MRI data, in step S5, statistical integration of MRI features is performed. The core idea is: (1) Calculate statistical features such as the maximum value, average value, etc. of each type of feature in all brain regions; (2) Select a few statistical features that perform best on the validation set from the above statistical features through non-repetitive random sampling as the final integrated features.

[0108] Embodiment 3

[0109] A nuclear magnetic resonance image feature system for predicting the progression of Alzheimer's disease, as Figure 2 shown, includes:

[0110] A nuclear magnetic resonance image feature extraction subsystem: including a screening module and a statistical integration module for extracting original nuclear magnetic resonance image features and establishing a feature model.

[0111] A screening module for original nuclear magnetic resonance image features: quantifying the nuclear magnetic resonance image, determining and selecting features. With the help of prior knowledge, candidate regions are determined and candidate features are selected. The candidate regions are 8 brain regions that are strongly correlated with the progression of Alzheimer's disease, specifically the prefrontal cingulate, inferior parietal lobule, inferior temporal lobe, lateral occipital lobe, lingual muscle, middle temporal lobe, superior marginal plate, and insula; 50 features are selected as candidate features for each brain region, and a secondary threshold screening can be performed, with the threshold greater than 1%;

[0112] A nuclear magnetic resonance feature statistical integration module: constructing various statistics to represent the macroscopic features of nuclear magnetic resonance, including nuclear magnetic resonance image features, clinical features, and genetic features related to Alzheimer's disease; integrating statistical features, including the maximum value, minimum value, sum, range, and standard deviation.

[0113] A nuclear magnetic resonance image feature comparison module for comparing and analyzing newly input nuclear magnetic resonance image features with the feature model;

[0114] The result output module outputs the comparative analysis result of the nuclear magnetic resonance image feature comparison module.

[0115] Example 4

[0116] A method for predicting the progression of Alzheimer's disease using nuclear magnetic resonance image features includes the following steps:

[0117] Ⅰ: Establish a nuclear magnetic resonance image feature system related to Alzheimer's disease;

[0118] Ⅱ: Quantify the brain nuclear magnetic resonance image of the person to be predicted and input it into the nuclear magnetic resonance image feature system described in step Ⅰ for feature matching.

[0119] The system in Ⅰ is as Figure 2 shown and includes: a nuclear magnetic resonance image feature extraction subsystem, a nuclear magnetic resonance image feature comparison module, and a result output module.

[0120] The nuclear magnetic resonance image feature extraction subsystem includes a screening module and a statistical integration module for extracting the original features of nuclear magnetic resonance images and establishing a feature model.

[0121] The screening module for the original features of nuclear magnetic resonance images: Quantify the nuclear magnetic resonance images, determine and select features. With the help of prior knowledge, determine the candidate regions and select candidate features. The candidate regions are 8 brain regions strongly related to the progression of Alzheimer's disease, specifically the prefrontal cingulate, inferior parietal lobule, inferior temporal lobe, lateral occipital lobe, genioglossus, middle temporal lobe, superior marginal plate, and insula; 50 features are selected as candidate features for each brain region, and secondary threshold screening can be performed, with the threshold greater than 1%.

[0122] The nuclear magnetic resonance feature statistical integration module: Construct various statistics to represent the macroscopic features of nuclear magnetic resonance, including nuclear magnetic resonance image features, clinical features, and gene features related to Alzheimer's disease; integrate statistical features, including the maximum value, minimum value, sum, range, and standard deviation.

[0123] The nuclear magnetic resonance image feature comparison module is used to compare and analyze the newly input nuclear magnetic resonance image features with the feature model;

[0124] The result output module outputs the comparative analysis result of the nuclear magnetic resonance image feature comparison module.

[0125] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for extracting magnetic resonance image features for predicting the progression of Alzheimer's disease, characterized in that, The original feature screening module for nuclear magnetic resonance (NMR) images is used to quantify, determine, and select features of NMR images. The statistical integration module for NMR image features is used to integrate statistical features. The specific steps are as follows: S1: Quantify the NMR images using software; S2: Determine the candidate regions and candidate features. Determine the brain regions strongly related to the progression of Alzheimer's disease as candidate regions. The candidate features are selected from the candidate regions. In step S2, the candidate regions are the prefrontal cingulate, inferior parietal lobule, inferior temporal lobe, lateral occipital lobe, genioglossus, middle temporal lobe, superior marginal plate, and insula; S3: Score the candidate regions and select the feature with the highest score for each region. Input the features of each brain region and the clinical / gene features into a linear kernel SVM model, score using the change in the model evaluation index, and retain the feature with the highest score for each brain region; S4: Perform a secondary screening using a threshold to screen out the candidate features with scores higher than the threshold in the candidate regions; S5: Construct and calculate statistical features. Calculate the statistics of each original feature on different brain regions. The statistic is a mathematical expression of the distribution of expression data; S6: Screen out the statistical feature with the highest score. The specific steps are as follows: (a) Select k statistical features from step S5, and send them together with the previous original NMR features, clinical features, and gene features into a support vector machine model with the same parameters as the previous model for training, and record the evaluation index of the model; (b) Repeat step (a) 3000 times; (c) Select the k statistical features that maximize the model evaluation index as the finally selected integrated features; where the value range of k is 3 - 8; The specific quantification method in step S1 is as follows: 1) Use FreeSurfer to label the regions on the cortical surface and the subcortical brain structures; 2) Use ANTs for brain volume extraction, segmentation, and registration; 3) Use Mindboggle software to automatically perform shape analysis on the labels and features extracted from FreeSurfer and ANTs. Mindboggle will finally calculate the volume of all labeled regions, the volume of all labeled cortical regions, and the shape features of each cortical surface label, including surface area, thickness, mean curvature, etc.

2. The method according to claim 1, characterized in that, The specific method in step S2 includes: a: Select 8 brain regions strongly related to the progression of Alzheimer's disease as candidate regions; b: Send each feature of the candidate regions together with the clinical features and gene features into a support vector machine model with the same parameters for training to obtain candidate features.

3. The method according to claim 1, characterized in that In step S3, the change in the model evaluation index is used as the score for the candidate regions. The scoring steps are as follows: (1) Divide the clinical and gene data provided by the ADNI dataset into a training set and a validation set in a ratio of 3:1, and then use grid search and cross - validation to train a support vector machine model with relatively optimal performance; (2) Add each candidate NMR feature to the training set and the validation set, retrain without changing the parameters of the model, and use the difference between the evaluation index of the trained model and the evaluation index of the support vector machine model in (1) as the score of this feature; (3) Only retain the features with the highest scores in each region.

4. The method according to claim 1, characterized in that, The secondary threshold screening is added in the step S4, and the threshold is 1%.

5. The method according to claim 1, characterized in that, The threshold screens out 5 original features of the classification model and 7 original features of the regression model.

6. The method according to claim 1, characterized in that The step S5 includes: a) Calculate the statistical features of all brain regions; b) Select the statistical features with the best performance on the validation set from the above statistical features by non-repetitive random sampling as the finally integrated statistical features.

7. A method for predicting the progression of Alzheimer's disease using nuclear magnetic resonance image features, characterized in that It includes the following steps: I: Establish a nuclear magnetic resonance image feature system by using a method for extracting nuclear magnetic resonance image features for predicting the progression of Alzheimer's disease according to claim 1; II: Quantify the nuclear magnetic resonance image of the person to be predicted and input it into the nuclear magnetic resonance image feature system described in step I for feature matching.

Images