Medical image feature extraction method and system

Abstract

The invention discloses a medical image feature extraction method and system, and belongs to the technical field of image processing. The method comprises the steps of firstly collecting and preprocessing medical image data to construct a data set, constructing an adaptive feature extraction model based on a U-Net network, obtaining an optimal model through an Adam algorithm in combination with a cross entropy loss function, and inputting the data set into the model to obtain initial image features; and then calculating image space and feature space uncertainty of the preliminary image features to determine comprehensive uncertainty, obtaining feature weights based on a machine learning model, performing weighted calculation to obtain comprehensive evaluation indexes, and finally selecting first q% of features as optimal image features according to index sorting. According to the method, the model is optimized to improve the feature extraction accuracy, the uncertainty perception mechanism can effectively screen the features, the redundancy and irrelevant feature influence are reduced, the comprehensive evaluation indexes comprehensively evaluate the features, the medical image analysis efficiency and diagnosis accuracy are improved, and a more reliable feature set is provided for medical image processing.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and more particularly to a method and system for extracting medical image features. Background Art

[0002] Against the backdrop of the booming development of smart healthcare, massive amounts of medical image data are constantly being generated, such as X-rays, CT scans, MRI images, etc. Accurate medical image feature extraction can extract valuable information from these complex and massive data. Medical image feature extraction methods have become a core and key link in the field of medical image analysis.

[0003] Traditional medical image feature extraction methods often rely on the experience and knowledge of doctors or experts. However, this method is subjective and inconsistent. Different doctors or experts may extract different feature sets, which affects the effectiveness and repeatability of the features. In addition, in medical images, the number of features may be very large, but not all features are relevant to the diagnosis or analysis tasks. Existing technologies often have difficulty effectively distinguishing which features are important and which are redundant or irrelevant. This results in a large number of irrelevant or redundant features being included in the feature extraction process, which reduces the effectiveness of the features and leads to low efficiency in disease diagnosis.

[0004] Therefore, how to construct an efficient, accurate and reliable medical image feature extraction method and system is a problem that technical personnel in this field need to solve urgently. Summary of the invention

[0005] In view of this, the present invention provides a method and system for extracting medical image features, which effectively improve the accuracy and efficiency of medical image feature extraction and provide strong support for medical image analysis.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] In one aspect, the present invention provides a method for extracting medical image features, comprising the following steps:

[0008] Collecting medical image data, and preprocessing the medical image data to construct a data set;

[0009] Build and train an adaptive feature extraction model to obtain the optimal feature extraction model;

[0010] Inputting the data set into the adaptive feature extraction model to obtain preliminary image features;

[0011] The preliminary image features are screened based on an uncertainty perception mechanism, and preferred image features are determined based on the importance and uncertainty of the preliminary image features.

[0012] Preferably, an adaptive feature extraction model is constructed based on a U-Net network, and the adaptive feature extraction model is optimized by an Adam algorithm and based on a cross entropy loss function to obtain an optimal feature extraction model.

[0013] Preferably, the adaptive feature extraction model is optimized by the Adam algorithm based on the cross entropy loss function to obtain the optimal feature extraction model, including:

[0014] Taking the minimization of the loss of the cross entropy function as the optimization goal, the Adam algorithm optimizes the model parameters according to the optimization goal;

[0015] BatchNorm technology is used to optimize the data distribution during the training process of the adaptive feature extraction model.

[0016] Preferably, the preliminary image features are screened based on an uncertainty perception mechanism, and the preferred image features are determined in combination with the importance and uncertainty of the preliminary image features, including:

[0017] Calculate the image space uncertainty and feature space uncertainty of each preliminary image feature;

[0018] Determining a comprehensive uncertainty of a preliminary image feature based on the image space uncertainty and the feature space uncertainty;

[0019] The importance of preliminary image features is obtained based on the machine learning model, and weighted calculations are performed with the corresponding comprehensive uncertainties to obtain comprehensive evaluation indicators.

[0020] The preliminary image features are ranked according to the comprehensive evaluation index, and the top q% of preliminary image features are selected as the preferred image features.

[0021] Preferably, for the nth preliminary image feature, the image space uncertainty CV n The calculation formula is as follows:

[0022]

[0023] |N(x,y)| represents the number of pixels in the neighborhood N(x,y) of pixel (x,y), μ(x,y) is the mean of all pixel values ​​in the neighborhood N(x,y), and I(i,j) is the pixel value of the image at position (i,j).

[0024] Preferably, for the m feature points of the nth preliminary image feature, the feature space uncertainty FPD n The calculation formula is as follows:

[0025]

[0026] In the formula, d(p l ,p k ) is the distance between feature points, and σ is a parameter that controls the distance sensitivity.

[0027] Preferably, the calculation formula of the comprehensive uncertainty A of the nth preliminary image feature is as follows:

[0028] A=α·CV n +β·FPD n ;

[0029] Where α and β are the image space uncertainty weight and feature space uncertainty weight, respectively.

[0030] On the other hand, the present invention also provides a medical image feature extraction system for implementing the above-mentioned medical image feature extraction method, comprising:

[0031] A data preprocessing module, used to collect medical image data, preprocess the medical image data, and construct a data set;

[0032] A model building module is used to build and train an adaptive feature extraction model to obtain an optimal feature extraction model;

[0033] A preliminary feature extraction module, used for inputting the data set into the adaptive feature extraction model to obtain preliminary image features;

[0034] The feature screening module is used to screen the preliminary image features based on the uncertainty perception mechanism, and determine the preferred image features in combination with the importance and uncertainty of the preliminary image features.

[0035] Preferably, the model building module includes:

[0036] A parameter optimization unit, used to minimize the loss of the cross entropy function as an optimization goal, and the Adam algorithm optimizes the model parameters according to the optimization goal;

[0037] The data distribution optimization unit is used to optimize the data distribution in the training process of the adaptive feature extraction model by using BatchNorm technology.

[0038] Preferably, the feature screening module includes:

[0039] an uncertainty calculation unit, for calculating the image space uncertainty and feature space uncertainty of each preliminary image feature;

[0040] a comprehensive uncertainty calculation unit, used for determining a comprehensive uncertainty of a preliminary image feature according to the image space uncertainty and the feature space uncertainty;

[0041] A comprehensive evaluation index acquisition unit is used to acquire the importance of preliminary image features based on a machine learning model, and perform weighted calculations with the corresponding comprehensive uncertainties to obtain a comprehensive evaluation index;

[0042] The preferred feature image determination unit is used to sort the preliminary image features according to the comprehensive evaluation index and select the first q% of the preliminary image features as the preferred image features.

[0043] It can be seen from the above technical solutions that compared with the prior art, the present invention discloses a method and system for extracting medical image features, which constructs a data set by collecting and preprocessing medical image data, constructs an adaptive feature extraction model using a U-Net network, optimizes model parameters with the goal of minimizing cross entropy loss, and uses BatchNorm to optimize data distribution to obtain the best model, thereby extracting preliminary image features. Subsequently, based on the uncertainty perception mechanism, combined with the uncertainty of image space and the uncertainty of feature space, the weights of preliminary image features are calculated through a machine learning model, and the preferred image features are comprehensively evaluated and screened out. The present invention improves the accuracy of feature extraction by optimizing the model, and the uncertainty perception mechanism can effectively screen features, reduce the influence of redundant and irrelevant features, and comprehensively evaluate features with comprehensive evaluation indicators, which helps to improve the efficiency and diagnostic accuracy of medical image analysis and provide a more reliable feature set for medical image processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0045] Figure 1 A flow chart of the method provided by the present invention;

[0046] Figure 2 An adaptive feature extraction model based on a U-Net network provided in an embodiment of the present invention;

[0047] Figure 3 This is a system architecture diagram provided by the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] On the one hand, reference Figure 1 The embodiment of the present invention provides a method for extracting medical image features, comprising the following steps:

[0050] S1. Collect medical image data, preprocess the medical image data, and construct a data set.

[0051] Collect data from a variety of medical imaging devices, including but not limited to CT scanners, MRI instruments, X-ray machines, etc., to ensure that images with different imaging principles and resolutions are covered to obtain comprehensive patient information. Develop corresponding interface programs and conversion tools based on the image formats and data characteristics of different devices so that the collected data can be processed uniformly. For example, for the DICOM format of CT images and the specific format of MRI images, perform format parsing and data extraction, and convert them into a common format suitable for subsequent preprocessing and analysis, such as the common image matrix format.

[0052] Noise removal: Use a variety of filtering algorithms, such as Gaussian filtering and median filtering, and select appropriate filter parameters according to the type of image noise. For images with more Gaussian noise, use a Gaussian filter to smooth the image by setting an appropriate kernel size (such as 3×3 or 5×5) to reduce the interference of noise on image details; for salt and pepper noise, use a median filter and select an appropriate neighborhood size (such as 3×3) to replace the noise pixel value.

[0053] Grayscale normalization: Calculate the grayscale histogram of the image, determine the grayscale range of the image, and normalize it to a specific grayscale interval, such as [0,255]. By counting the number of pixels at each grayscale level in the image, find the minimum and maximum grayscale values, and then map the grayscale value of each pixel to the target interval according to the normalization formula, the contrast and brightness consistency of the image are enhanced, which is convenient for subsequent feature extraction.

[0054] Image enhancement: Use techniques such as histogram equalization and contrast stretching. Histogram equalization redistributes the grayscale histogram of the image to make the grayscale distribution of the image more uniform and improve the contrast of the image; contrast stretching selects an appropriate stretching coefficient based on the grayscale range of the image to expand the grayscale dynamic range of the image and highlight the details in the image.

[0055] Image cropping and scaling: Automatically crop images according to the region of interest (ROI), remove irrelevant background information, and reduce data volume and computational burden. For example, in lung CT images, the lung area is automatically identified and cropped, retaining only the lung and its surrounding related areas. At the same time, according to the subsequent model input requirements, the image is scaled, such as scaling images of different resolutions to standard sizes such as 256×256 or 512×512 to ensure data consistency.

[0056] Data labeling and classification: Invite professional medical imaging doctors to label some images, including lesion areas, organ contours, abnormal structures, etc. After labeling, the images are classified according to the labeling categories and subdivided into multiple categories according to the specific disease types, providing accurate label information for subsequent model training.

[0057] S2. Build and train an adaptive feature extraction model to obtain the optimal feature extraction model.

[0058] Build an adaptive feature extraction model based on the U-Net network, refer to Figure 2 The network is an encoder-decoder structure, where the encoder is used to extract contextual information of the image and the decoder is used to restore image information.

[0059] U-Net uses the same-level convolution between the up and down sampling layers. The corresponding up and down sampling layers are spliced ​​in the form of jump connections instead of pixel-by-pixel splicing layer by layer. This jump connection structure helps to transfer information from the shallow layer to the deep layer of the network, thereby obtaining better segmentation results.

[0060] The U-Net network structure is shaped like the Latin letter U and mainly includes two paths, namely the contraction path and the expansion path.

[0061] The contraction path contains two repeated 3×3 convolutions, followed by a convolution (RELU), a max pooling (maxpooling, stride 2), which is used for downsampling. Each downsampling reduces the feature channel by half. The expansion path contains an upsampling (2×2 upconvolution), followed by a feature map of the corresponding contraction path, and then two 3×3 convolutions, each followed by a RELU. Because each convolution will lose the edge of the image, the size of the image will change, which is not conducive to the subsequent information propagation, so it is cropped, followed by a 1×1 convolution. This network structure has no fully connected layers. The deeper layers are used to solve the problem of pixel classification, and the shallower high-resolution layers are used to solve the problem of pixel positioning.

[0062] The adaptive feature extraction model is optimized by the Adam algorithm based on the cross entropy loss function to obtain the best feature extraction model, including:

[0063] Taking the minimization of the loss of the cross entropy function as the optimization goal, the Adam algorithm optimizes the model parameters according to the optimization goal.

[0064] BatchNorm technology is used to optimize data distribution during the training process of the adaptive feature extraction model.

[0065] S3. Input the dataset into the adaptive feature extraction model to obtain preliminary image features.

[0066] S4. Screen the preliminary image features based on the uncertainty perception mechanism, and determine the preferred image features based on the importance and uncertainty of the preliminary image features, including:

[0067] S41. Calculate the image space uncertainty and feature space uncertainty of each preliminary image feature.

[0068] For the nth preliminary image feature, the image space uncertainty CV n The calculation formula is as follows:

[0069]

[0070] |N(x,y)| represents the number of pixels in the neighborhood N(x,y) of pixel (x,y), μ(x,y) is the mean of all pixel values ​​in the neighborhood N(x,y), and I(i,j) is the pixel value of the image at position (i,j).

[0071] For the m feature points of the nth preliminary image feature, the feature space uncertainty FPD n The calculation formula is as follows:

[0072]

[0073] In the formula, d(p l ,p k ) is the distance between feature points, and σ is a parameter that controls the distance sensitivity.

[0074] S42. Determine the combined uncertainty of preliminary image features based on the image space uncertainty and the feature space uncertainty.

[0075] The calculation formula of the comprehensive uncertainty A of the nth preliminary image feature is as follows:

[0076] A=α·CV n +β·FPD n ;

[0077] Where α and β are the image space uncertainty weight and feature space uncertainty weight, respectively, and α+β=1.

[0078] S43. Based on the machine learning model, the importance of preliminary image features is obtained, and weighted calculations are performed with the corresponding comprehensive uncertainties to obtain comprehensive evaluation indicators.

[0079] In this embodiment, a machine learning model (such as a random forest, a support vector machine, etc.) is used for training, and the importance of features is evaluated based on the output of the model (such as accuracy, F1 score, etc.) or the internal mechanism of the model (such as the branch structure of a decision tree, the weight coefficient of a support vector machine, etc.).

[0080] A weight coefficient is set, uncertainty and importance are weighted and summed to obtain a comprehensive evaluation index. The choice of weight can be adjusted according to the specific situation of the problem.

[0081] S44. Sort the preliminary image features according to the comprehensive evaluation index, and select the first q% of the preliminary image features as the preferred image features.

[0082] On the other hand, Figure 3 As shown, the present invention also proposes a medical image feature extraction system, which is used to implement the above-mentioned medical image feature extraction method, comprising:

[0083] A data preprocessing module is used to collect medical image data, preprocess the medical image data, and construct a data set;

[0084] A model building module is used to build and train an adaptive feature extraction model to obtain an optimal feature extraction model;

[0085] A preliminary feature extraction module is used to input the data set into the adaptive feature extraction model to obtain preliminary image features;

[0086] The feature screening module is used to screen the preliminary image features based on the uncertainty perception mechanism, and determine the preferred image features based on the importance and uncertainty of the preliminary image features.

[0087] Preferably, the model building module includes:

[0088] The parameter optimization unit is used to minimize the loss of the cross entropy function as the optimization goal. The Adam algorithm optimizes the model parameters according to the optimization goal;

[0089] The data distribution optimization unit is used to optimize the data distribution in the adaptive feature extraction model training process using BatchNorm technology.

[0090] Preferably, the feature screening module includes:

[0091] an uncertainty calculation unit, for calculating the image space uncertainty and feature space uncertainty of each preliminary image feature;

[0092] A comprehensive uncertainty calculation unit, used for determining the comprehensive uncertainty of the preliminary image features according to the image space uncertainty and the feature space uncertainty;

[0093] A comprehensive evaluation index acquisition unit is used to acquire the importance of preliminary image features based on a machine learning model, and perform weighted calculations with the corresponding comprehensive uncertainties to obtain a comprehensive evaluation index;

[0094] The preferred feature image determination unit is used to sort the preliminary image features according to the comprehensive evaluation index and select the first q% of the preliminary image features as the preferred image features.

[0095] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0096] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for extracting medical image features, characterized in that: The following steps are involved: Collecting medical image data, and preprocessing the medical image data to construct a data set; Build and train an adaptive feature extraction model to obtain the optimal feature extraction model; Inputting the data set into the adaptive feature extraction model to obtain preliminary image features; The preliminary image features are screened based on an uncertainty perception mechanism, and preferred image features are determined based on the importance and uncertainty of the preliminary image features.

2. A method for extracting medical image features according to claim 1, characterized in that: An adaptive feature extraction model is constructed based on the U-Net network, and the adaptive feature extraction model is optimized by the Adam algorithm based on the cross entropy loss function to obtain the optimal feature extraction model.

3. A method for extracting medical image features according to claim 2, characterized in that: The adaptive feature extraction model is optimized by the Adam algorithm based on the cross entropy loss function to obtain the best feature extraction model, including: Taking the minimization of the loss of the cross entropy function as the optimization goal, the Adam algorithm optimizes the model parameters according to the optimization goal; BatchNorm technology is used to optimize the data distribution during the training process of the adaptive feature extraction model.

4. A method for extracting medical image features according to claim 1, characterized in that: The preliminary image features are screened based on the uncertainty perception mechanism, and the preferred image features are determined in combination with the importance and uncertainty of the preliminary image features, including: Calculate the image space uncertainty and feature space uncertainty of each preliminary image feature; Determining a comprehensive uncertainty of a preliminary image feature based on the image space uncertainty and the feature space uncertainty; The importance of preliminary image features is obtained based on the machine learning model, and weighted calculations are performed with the corresponding comprehensive uncertainties to obtain comprehensive evaluation indicators. The preliminary image features are ranked according to the comprehensive evaluation index, and the top q% of preliminary image features are selected as the preferred image features.

5. A method for extracting medical image features according to claim 4, characterized in that: For the nth preliminary image feature, the image space uncertainty CV n The calculation formula is as follows: |N(x,y)| represents the number of pixels in the neighborhood N(x,y) of pixel (x,y), μ(x,y) is the mean of all pixel values ​​in the neighborhood N(x,y), and I(i,j) is the pixel value of the image at position (i,j).

6. A method for extracting medical image features according to claim 5, characterized in that: For the m feature points of the nth preliminary image feature, the feature space uncertainty FPD n The calculation formula is as follows: In the formula, d(p l ,p k ) is the distance between feature points, and σ is a parameter that controls the distance sensitivity.

7. A method for extracting medical image features according to claim 6, characterized in that: The calculation formula of the comprehensive uncertainty A of the nth preliminary image feature is as follows: A=α·CV n +β·FPD n ; Where α and β are the image space uncertainty weight and feature space uncertainty weight, respectively.

8. A medical image feature extraction system, characterized in that: include: A data preprocessing module, used to collect medical image data, preprocess the medical image data, and construct a data set; A model building module is used to build and train an adaptive feature extraction model to obtain an optimal feature extraction model; A preliminary feature extraction module, used for inputting the data set into the adaptive feature extraction model to obtain preliminary image features; The feature screening module is used to screen the preliminary image features based on the uncertainty perception mechanism, and determine the preferred image features in combination with the importance and uncertainty of the preliminary image features.

9. A medical image feature extraction system according to claim 8, characterized in that: The model building module includes: A parameter optimization unit, used to minimize the loss of the cross entropy function as an optimization goal, and the Adam algorithm optimizes the model parameters according to the optimization goal; The data distribution optimization unit is used to optimize the data distribution in the training process of the adaptive feature extraction model by using BatchNorm technology.

10. A medical image feature extraction system according to claim 8, characterized in that: The feature screening module includes: an uncertainty calculation unit, for calculating the image space uncertainty and feature space uncertainty of each preliminary image feature; a comprehensive uncertainty calculation unit, used for determining a comprehensive uncertainty of a preliminary image feature according to the image space uncertainty and the feature space uncertainty; A comprehensive evaluation index acquisition unit is used to acquire the importance of preliminary image features based on a machine learning model, and perform weighted calculations with the corresponding comprehensive uncertainties to obtain a comprehensive evaluation index; The preferred feature image determination unit is used to sort the preliminary image features according to the comprehensive evaluation index and select the first q% of the preliminary image features as the preferred image features.

Images