An Intuitive Fuzzy Encoder for Feature Selection in Alzheimer's Disease Images

Abstract

This invention provides a feature selection method for Alzheimer's disease images based on an intuitionistic fuzzy encoder, belonging to the field of intelligent medical information processing technology. It solves the technical problem of high-dimensional feature redundancy and poor classification performance in Alzheimer's disease analysis. The technical solution includes the following steps: S10, collecting and preprocessing a sample set of Alzheimer's disease data from subjects; S20, calculating the membership and non-membership degrees of each sample and constructing a score matrix based on an intuitionistic fuzzy set; S30, pre-training an intuitionistic fuzzy autoencoder model for global samples; S40, constructing a sparse feature encoding network, transferring learning parameters, and sparsifying the weight matrix to obtain important features. The beneficial effects of this invention are: it effectively handles the feature redundancy problem in pathological data, reducing data dimensionality to decrease computational requirements and improving classification accuracy, thus providing technical support for actual clinical diagnosis.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical information processing technology, specifically to a method for selecting features from Alzheimer's disease images based on an intuitionistic fuzzy encoder. Technical Background

[0002] Alzheimer's disease is a slowly progressing neurodegenerative disease characterized by progressively worsening cognitive impairment, involving declines in memory, language, thinking, and judgment. Patients may initially experience mild memory loss; as the disease progresses, they gradually experience a weakening of their daily activities and personality changes, eventually potentially losing all ability to care for themselves.

[0003] In recent years, scholars have developed feature selection methods such as mutual information method, variance selection method and moth flame optimization algorithm to help understand data, reduce irrelevant and redundant features, and reduce data dimensionality to reduce computational requirements. However, the above feature selection methods have problems such as sensitivity to noisy data, inability to handle linear data, and easy getting trapped in local optima.

[0004] The emergence of deep learning technology in recent years has brought new solutions to the field of smart healthcare. Autoencoders can automatically learn and recognize medical images with complex features, possessing powerful feature extraction and pattern recognition capabilities, and show great promise in the field of medical image processing. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an image feature selection method for Alzheimer's disease based on an intuitionistic fuzzy encoder. In the data processing stage, intuitionistic fuzzy membership is applied to mine the relationships between data points, assigning different weights to each data point for preprocessing. Simultaneously, a one-to-one selection layer is introduced to reduce irrelevant and redundant features, lowering the data dimensionality to reduce computational requirements and improving the efficiency of Alzheimer's brain image feature selection. This method has significant application value for Alzheimer's brain image feature selection.

[0006] The inventive concept of this invention is as follows: First, the Alzheimer's disease test sample set is divided; then, the membership degree and non-membership degree of each sample are calculated to construct a score matrix based on intuitionistic fuzzy sets; next, an intuitionistic fuzzy encoder model for global samples is pre-trained; finally, a sparse feature encoding network is constructed, the encoder parameters from the pre-training process are loaded, and the weight matrix is ​​sparsified to obtain important features.

[0007] This invention is achieved through the following measures: a method for selecting brain image features in Alzheimer's disease based on an intuitionistic fuzzy autoencoder, comprising the following steps:

[0008] S10: Collect samples of brain images of subjects with Alzheimer's disease, construct a decision information system, and divide the data into a training sample set and a test sample set with a ratio of 9:1;

[0009] S20: Calculate the membership and non-membership of each sample and construct a score matrix based on intuitionistic fuzzy sets;

[0010] S30: Pre-train an intuitionistic fuzzy autoencoder model for a global sample using the training sample set. During training, forward and backward propagation are performed, and the mean squared error loss function is used to update the network parameters.

[0011] S40: Construct a sparse feature encoding network, load the encoder parameters from the pre-training stage, and sparsify the weight matrix to obtain important features.

[0012] Further, step S10 includes the following steps:

[0013] S11: Collect raw MRI images, apply brain imaging data processing and analysis to denoise the data, assign labels to the samples based on clinical indications, and establish a raw decision information system. ,in This represents a sample set of brain image data related to Alzheimer's disease. Indicates the first One sample, ; This indicates the number of features in brain imaging data related to Alzheimer's disease. Indicates the first One characteristic, ; Class labels representing brain image data related to Alzheimer's disease. Indicates the first The class labels of each sample, where , indicating that the sample does not / has Alzheimer's disease.

[0014] S12: Divide the above Alzheimer's brain image data into a training sample set and a test sample set, with a ratio of 9:1.

[0015] Further, step S20 includes the following steps:

[0016] S21: Construct an intuitionistic fuzzy set by bidirectionally determining the fuzzy relationship between samples and categories in an Alzheimer's brain image dataset using membership and non-membership values. ,in This represents the membership value of a sample. This represents the non-membership value of the sample;

[0017] S22: Calculate the symptomatic class center of the sample. and asymptomatic samples :

[0018]

[0019] in, This refers to the number of symptomatic samples. It is the number of asymptomatic samples;

[0020] S23: Calculate the symptomatic class radius of the sample. and asymptomatic radius :

[0021]

[0022] S24: Construct samples based on the symptomatic and asymptomatic classes in the dataset. The membership function is mapped to the [0,1] space to represent the sample. The degree of importance is defined as follows:

[0023]

[0024] in, Kernel mapping represents mapping data to a high-dimensional feature space. This is the bias number;

[0025] S25: Calculation Construct non-membership functions for potential heterogeneous samples. Its definition is as follows:

[0026]

[0027] Indicates sample The proportion of heterogeneous data points in its neighborhood is expressed as follows:

[0028]

[0029] in, Adjustable parameters for creating a neighborhood;

[0030] S26: Compare and classify the calculated membership function values ​​and non-membership function values. The quality function for each sample is represented by the sample name. The fuzzy similarity relation corresponding to the class centers is represented as follows:

[0031]

[0032] The Alzheimer's disease dataset is denoted as , For the sample size, Given the eigenvalues, we obtain its quality matrix, which is represented as follows:

[0033] .

[0034] Further, step S30 includes the following steps:

[0035] S31: Initialize encoder and decoder parameters According to the mass matrix in step S26 Initialize dataset The initial feature mapping matrix is ​​obtained. ;

[0036] S32: Retrieve the current batch from the training dataset. Using an automatic encoder Batch Mapping to a low-dimensional space yields the latent representation of the encoder network. ;

[0037] S33: Receive the latent representation using the decoder. Mapping it back to the original data space yields the reconstructed original data. , where the decoder The definition is as follows:

[0038]

[0039] in, and These represent the weight matrices of the first hidden layer and the output layer, respectively. and It is a bias term. The activation function for the hidden layer;

[0040] S34: Calculate the initial feature map matrix using the mean squared error loss function. and the data reconstructed by the decoder The differences between them are expressed as follows:

[0041]

[0042] in, yes Norm, representing the sum of the squared differences of all elements in a matrix;

[0043] S35: Update the parameters using gradient descent based on the calculated loss value. Returns the optimized optimal parameter values. .

[0044] Further, step S40 includes the following steps:

[0045] S41: Construct a sparse feature encoding network, containing a one-to-one selection layer and an encoder. The one-to-one selection layer assigns an initial weight to each feature. The feature used for filtering is represented as follows:

[0046]

[0047] in Weight This reflects the importance of features, identifying and selecting the most relevant features after training;

[0048] S42: Utilize transfer learning to inherit the optimal encoder parameters learned in step S35. The objective function for updating the weight parameters during training is expressed as follows:

[0049]

[0050] in Represents a pre-trained encoder network The generated latent representation, Indicates encoder Regularization term, Represents a one-to-one selection layer The regularization term is used to sparsify the weight matrix to obtain important features.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] 1. This invention discloses a method for selecting brain image features in Alzheimer's disease based on an intuitionistic fuzzy autoencoder. It introduces intuitionistic fuzzy membership degrees to mine the relationships between data and assign different weights to the data for preprocessing. This method improves the accuracy and reliability of disease classification.

[0053] 2. This invention employs a method for pre-training an intuitionistic fuzzy autoencoder model on a global sample, enabling the model to learn more general feature representations and exhibiting better adaptability to data distributions of unseen samples. This method improves the reconstruction and classification performance of the transfer learning model.

[0054] 3. This invention constructs a sparse feature selection network method. While ensuring reconstruction results, it introduces a one-to-one selection layer to reduce irrelevant and redundant features, thereby reducing data dimensionality and computational requirements. This method improves the efficiency of feature selection in Alzheimer's disease brain images. Attached Figure Description

[0055] The technical solution and its beneficial effects of the present invention will become apparent from the following detailed description of specific embodiments in conjunction with the accompanying drawings.

[0056] Figure 1 This is a flowchart of a brain image feature selection method for Alzheimer's disease based on an intuitionistic fuzzy autoencoder, as described in an embodiment of the present invention.

[0057] Figure 2 This is a block diagram of a brain image feature selection method for Alzheimer's disease based on an intuitionistic fuzzy autoencoder, according to an embodiment of the present invention.

[0058] Figure 3 is a pre-trained global sample model diagram of a brain image feature selection method for Alzheimer's disease based on an intuitionistic fuzzy autoencoder in an embodiment of the present invention.

[0059] Figure 4 is a sparse feature encoding network model for selecting important features in an Alzheimer's disease brain image feature selection method based on an intuitionistic fuzzy autoencoder in an embodiment of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Example 1

[0062] This embodiment provides a method for feature selection of Alzheimer's disease images based on an intuitionistic fuzzy encoder, such as... Figures 1-4 As shown, it includes the following steps:

[0063] S10: Collect samples of brain images of subjects with Alzheimer's disease, construct a decision information system, and divide the data into a training sample set and a test sample set with a ratio of 9:1;

[0064] S20: Calculate the membership and non-membership of each sample and construct a score matrix based on intuitionistic fuzzy sets;

[0065] S30: Pre-train an intuitionistic fuzzy autoencoder model for a global sample using the training sample set. During training, forward and backward propagation are performed, and the mean squared error loss function is used to update the network parameters.

[0066] S40: Construct a sparse feature encoding network, load the encoder parameters from the pre-training stage, and sparsify the weight matrix to obtain important features.

[0067] Specifically, step S10 includes the following steps:

[0068] S11: Collect raw MRI images, apply brain imaging data processing and analysis to denoise the data, assign labels to the samples based on clinical indications, and establish a raw decision information system. ,in This represents a sample set of brain image data related to Alzheimer's disease. Indicates the first One sample, ; This indicates the number of features in brain imaging data related to Alzheimer's disease. Indicates the first One characteristic, ; Class labels representing brain image data related to Alzheimer's disease. Indicates the first The class labels of each sample, where The expression indicates that the sample does not have / has Alzheimer's disease. The original dataset is transformed into a decision information system as shown in Table 1 below:

[0069] Table 1

[0070]

[0071] S12: The above Alzheimer's brain image data is divided into a training sample set and a test sample set in a ratio of 9:1. The sample set database, the number of training samples, and the number of test samples are shown in Table 2 below:

[0072] Table 2

[0073]

[0074] Specifically, step S20 includes the following steps:

[0075] S21: Construct an intuitionistic fuzzy set by bidirectionally determining the fuzzy relationship between dataset samples and categories using membership and non-membership values. ,in This represents the membership value of a sample. This represents the non-membership value of the sample;

[0076] S22: Calculate the symptomatic class center of the sample. and asymptomatic samples :

[0077]

[0078] in, This refers to the number of symptomatic samples. It is the number of asymptomatic samples;

[0079] S23: Calculate the symptomatic class radius of the sample. and asymptomatic radius :

[0080]

[0081] S24: Construct samples based on the symptomatic and asymptomatic classes in the dataset. The membership function is mapped to the [0,1] space to represent the sample. The degree of importance is defined as follows:

[0082]

[0083] in, Kernel mapping represents mapping data to a high-dimensional feature space. This is the bias number;

[0084] S25: Calculation Construct non-membership functions for potential heterogeneous samples. Its definition is as follows:

[0085]

[0086] Indicates sample The proportion of heterogeneous data points in its neighborhood is expressed as follows:

[0087]

[0088] in, Adjustable parameters for creating a neighborhood;

[0089] S26: Compare and classify the calculated membership function values ​​and non-membership function values. The quality function for each sample is represented by the sample name. The fuzzy similarity relation corresponding to the class centers is represented as follows:

[0090]

[0091] Table 3 shows the quality function for each sample, from which the fuzzy similarity relationships between samples can be derived:

[0092] Table 3

[0093]

[0094] The Alzheimer's disease dataset is denoted as , For the sample size, Given the eigenvalues, we obtain its quality matrix, which is represented as follows:

[0095] .

[0096] Specifically, step S30 includes the following steps:

[0097] S31: Initialize encoder and decoder parameters According to the mass matrix in step S26 Initialize dataset The initial feature mapping matrix is ​​obtained. The calculation results are as follows:

[0098]

[0099] S32: Retrieve the current batch from the training dataset. Using an automatic encoder Batch Mapping to a low-dimensional space yields the latent representation of the encoder network. ;

[0100] S33: Receive the latent representation using the decoder. Mapping it back to the original data space yields the reconstructed original data. , where the decoder The definition is as follows:

[0101]

[0102] in, and These represent the weight matrices of the first hidden layer and the output layer, respectively. and It is a bias term. The activation function for the hidden layer;

[0103] S34: Calculate the initial feature map matrix using the mean squared error loss function. and the data reconstructed by the decoder The differences between them are expressed as follows:

[0104]

[0105] in, yes Norm, representing the sum of the squared differences of all elements in a matrix;

[0106] S35: Update the parameters using gradient descent based on the calculated loss value. Returns the optimized optimal parameter values. .

[0107] Specifically, step S40 includes the following steps:

[0108] S41: Construct a sparse feature encoding network, including a selection layer and an encoder. The selection layer assigns an initial weight to each feature. The feature used for filtering is represented as follows:

[0109]

[0110] in Weight This reflects the importance of features, identifying and selecting the most relevant features after training;

[0111] S42: Utilize transfer learning to inherit the optimal encoder parameters learned in step S35. The objective function for updating the weight parameters during training is expressed as follows:

[0112]

[0113] in Represents a pre-trained encoder network The generated latent representation, Indicates encoder Regularization term, Represents a one-to-one selection layer The regularization term is used to sparsify the weight matrix to obtain important features. In this example, the batch size is set to 20, the global encoder iterations are 400, and the feature selection encoder iterations are 600. The feature selection method for Alzheimer's brain images based on intuitionistic fuzzy autoencoders ultimately selects a subset of important features. .

[0114] Example 2

[0115] Referring to Example 1, this example uses the parameters and results calculated in Example 1 to compare with traditional methods, demonstrating the superiority of this example. In the specific comparison, we employed different models and compared different indicators. The final results show that this example is preferred over traditional methods.

[0116] 1. Traditional Model

[0117] CNN Model: When used for Alzheimer's feature selection tasks, the CNN (Convolutional Neural Network) model first preprocesses the brain image data, including denoising, standardization, and segmentation. Then, the CNN processes the image layer by layer through multiple convolutional and pooling layers, automatically extracting disease-related local features, such as changes in cortical thickness and hippocampal atrophy. Disadvantages: CNN networks are deep and computationally complex. When processing large-scale image data, the computational and storage requirements are relatively high, resulting in long training times and high hardware resource consumption.

[0118] DNN Model: The DNN (Deep Neural Network) model standardizes and denoise Alzheimer's brain images through preprocessing steps before feeding them into multiple fully connected layers. The DNN learns high-level abstract features of the input data through multi-layer nonlinear transformations. The output of each layer is processed by an activation function, progressively extracting complex features associated with the Alzheimer's brain images. Disadvantages: DNNs typically lack spatial structure perception and cannot effectively capture local structural features in images like other neural networks.

[0119] 2. Comparison Indicators

[0120] This example demonstrates a feature selection task on brain images associated with Alzheimer's disease, using classification accuracy (ACC) as the evaluation metric. We employ three different classification models for evaluation: Support Vector Machine (SVM), Logistic Regression (LR), and Multilayer Perceptron (MLP).

[0121] 3. Classifier Model

[0122] Support Vector Machines (SVMs): The core idea is to find a hyperplane in the feature space that separates samples of different classes as much as possible while maximizing the margin between classes. SVMs perform well in high-dimensional spaces and are particularly suitable for datasets with dimensions greater than the number of samples.

[0123] Logistic regression is suitable for binary classification problems. It can quickly find the optimal separating hyperplane, thus achieving high classification accuracy. Logistic regression is robust, thus avoiding the risk of overfitting.

[0124] Multilayer perceptron (MLP): By mimicking the workings of a biological neural system, it uses multiple fully connected layers for feature learning and classification decisions. Each layer of the MLP has a non-linear activation function, enabling it to handle data with complex non-linear relationships, surpassing traditional linear classifiers.

[0125] 4. Comparison Results

[0126] As can be seen from the charts, this embodiment performs excellently across all evaluation metrics, significantly outperforming other models. In the three classifiers—Support Vector Machine (SVM), Logistic Regression (LR), and Multilayer Perceptron (MLP)—the feature subset classification accuracy after feature selection in this embodiment reaches 77.03% in SVM, significantly higher than DNN's 62.82% and CNN's 66.67%, indicating that this method performs well in feature selection on high-dimensional feature datasets. The accuracy of this method in Logistic Regression (LR) reaches 81.08%, significantly higher than DNN's 65.38% and CNN's 67.95%, indicating that this embodiment has more robust training performance and high robustness. The classification accuracy of this method in Multilayer Perceptron (MLP) reaches 83.78%, significantly higher than DNN's 65.38% and CNN's 67.95%, indicating that this embodiment can learn the non-linear relationships between data and handle more complex data. The feature selection model in this embodiment can select important features to accurately classify Alzheimer's disease in different classifiers, demonstrating its superior performance. Specific comparisons are shown in Table 4:

[0127] Table 4

[0128]

[0129] The above description is merely an exemplary embodiment of the present invention and does not limit the scope of patent protection of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An Alzheimer's disease image feature selection method based on an intuitionistic fuzzy encoder, characterized in that, Includes the following steps: S10: Collect samples of brain images of subjects with Alzheimer's disease, construct a decision information system, and divide the data into a training sample set and a test sample set with a ratio of 9:1; S20: Calculate the membership and non-membership of each sample, and construct a quality matrix based on intuitionistic fuzzy sets ; S30: Pre-train an intuitionistic fuzzy autoencoder model for a global sample using the training sample set. During training, forward and backward propagation are performed, and the mean squared error loss function is used to update the network parameters. Step S30 includes the following steps: S31: initialize parameters of the encoder and the decoder , initialize a dataset according to a quality matrix , obtain an initial feature mapping matrix ;​ S32: take out a current batch from the training data set , use the auto-encoder to map the batch to a low-dimensional space to obtain a latent representation of the encoder network ; S33: receiving the latent representation using a decoder to the original data space, resulting in reconstructed original data where the decoder is defined as follows: ； wherein, and W1and W0denote the weight matrices of the first hidden layer and the output layer, respectively, and are bias terms, is an activation function of the hidden layer; S34: Calculate initial feature mapping matrix using mean square error loss function and the data reconstructed by the decoder The difference between the two is represented as follows: ； wherein is norm, representing the sum of the squared differences of all elements in the matrix; S35: updating the parameters using gradient descent method according to the calculated loss value , returning the optimized optimal parameter value ; S40: constructing a sparse feature encoding network, loading the encoder parameters during pre-training, and performing sparse processing on the weight matrix to obtain important features; Step S40 includes the following steps: S41: Construct a sparse feature encoding network, including a one-to-one selection layer and an encoder, and the one-to-one selection layer assigns an initial weight to each feature For screening features, it is represented as follows: ； wherein , the weights reflect the importance of the features, the most relevant features being identified and selected after training; S42: Utilize transfer learning to inherit the optimal encoder parameters learned in step S35. The objective function for updating the weight parameters during training is expressed as follows: ； in Represents a pre-trained encoder network The generated latent representation, Indicates encoder Regularization term, Indicates the selection layer The regularization term is used to sparsify the weight matrix to obtain important features.

2. The method for feature selection of Alzheimer's disease images based on an intuitionistic fuzzy encoder according to claim 1, characterized in that, Step S10 includes the following steps: S11: Collect raw MRI images, apply brain imaging data processing and analysis to denoise the data, assign labels to the samples based on clinical diagnosis, and establish a raw decision information system. ,in This represents a sample set of brain image data related to Alzheimer's disease. Indicates the first One sample, ; This indicates the number of features in brain imaging data related to Alzheimer's disease. Indicates the first One characteristic, ; Class labels representing brain image data related to Alzheimer's disease. Indicates the first The class labels of each sample, where This indicates that the sample does not / has Alzheimer's disease; S12: Divide the above Alzheimer's brain image data into a training sample set and a test sample set, with a ratio of 9:

1.

3. The method for feature selection of Alzheimer's disease images based on an intuitionistic fuzzy encoder according to claim 2, characterized in that, Step S20 includes the following steps: S21: Construct an intuitionistic fuzzy set by bidirectionally determining the fuzzy relationship between samples and categories in an Alzheimer's brain image dataset using membership and non-membership values. ,in This represents the membership value of a sample. This represents the non-membership value of the sample; S22: Calculate the symptomatic class center of the sample. and asymptomatic samples : ； in, This refers to the number of symptomatic samples. It is the number of asymptomatic samples; S23: Calculate the symptomatic class radius of the sample. and asymptomatic radius : ； S24: Construct samples based on the symptomatic and asymptomatic classes in the dataset. The membership function is mapped to the [0,1] space to represent the sample. The degree of importance is defined as follows: ； in, Kernel mapping represents mapping data to a high-dimensional feature space. This is the bias number; S25: Calculation Construct non-membership functions for potential heterogeneous samples. Its definition is as follows: ； Indicates sample The proportion of heterogeneous data points in its neighborhood is expressed as follows: ； in, Adjustable parameters for creating a neighborhood; S26: Compare and classify the calculated membership function values ​​and non-membership function values. The quality function for each sample is represented by the sample name. The fuzzy similarity relation corresponding to the class centers is represented as follows: ； The Alzheimer's disease dataset is denoted as , For the sample size, Given the eigenvalues, we obtain its quality matrix, which is represented as follows: 。

Images