Three-dimensional magnetic resonance image super-resolution reconstruction method, electronic equipment and storage medium

Abstract

The invention discloses a three-dimensional magnetic resonance image super-resolution reconstruction method, electronic equipment and a storage medium. The method comprises the following steps: data acquisition: acquiring a low-resolution 3D MRI image as input data; image preprocessing: preprocessing an input low-resolution image; feature extraction: extracting features from the low-resolution image; a super-resolution reconstruction method based on a deep learning method is selected; network training: a deep learning method is selected, and network training needs to be carried out; super-resolution reconstruction: mapping a low-resolution image to a high-resolution space by using a super-resolution reconstruction method of a deep learning method; and post-processing: carrying out post-processing on the reconstructed high-resolution image. Complex correlation and texture features between different structures and organizations can be adaptively learned through end-to-end learning by using a deep learning method, and compared with the prior art, the traditional method often depends on some simplified hypotheses and possibly has poor performance for complex image contents.

Description

Technical Field

[0001] The invention relates to the field of image processing, and in particular to an electronic device and a storage medium for a three-dimensional magnetic resonance image super-resolution reconstruction method. Background Art

[0002] Magnetic resonance refers to the phenomenon of spin magnetic resonance, which has a broad meaning and includes nuclear magnetic resonance, electron paramagnetic resonance or electron spin resonance. In addition, the magnetic resonance that people often talk about in daily life refers to magnetic resonance imaging, which is a type of imaging device used for medical examinations that uses the nuclear magnetic resonance phenomenon.

[0003] Traditional methods are often based on some assumptions, such as image smoothness, local image correlation, etc. However, in magnetic resonance images, the correlation and texture features between different structures and tissues may be complex and diverse, and cannot be covered by simple assumptions, thus limiting the performance of traditional methods. The interpolation methods and statistical models in traditional methods usually involve a large number of parameters, such as the interpolation type and interpolation step size in the interpolation algorithm. The parameter selection has a great influence on the final reconstruction result, but the selection of specific parameters is usually empirical, and it is difficult to determine the optimal value. Summary of the invention

[0004] The object of the present invention is to provide an electronic device and a storage medium for a three-dimensional magnetic resonance image super-resolution reconstruction method to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A three-dimensional magnetic resonance image super-resolution reconstruction method comprises the following steps:

[0007] Step S1, data acquisition: obtaining a low-resolution 3D MRI image as input data, and obtaining a corresponding high-resolution image as reference data according to requirements;

[0008] Step S2, image preprocessing: preprocessing the input low-resolution image, including denoising, grayscale standardization, and artifact removal to reduce noise and improve image quality;

[0009] Step S3, feature extraction: using a local statistical feature extraction method to extract features from the low-resolution image, the extracted features specifically include texture, shape, structure and other features;

[0010] Step S4, method selection: selecting a super-resolution reconstruction method based on a deep learning method;

[0011] Step S5, network training: After the deep learning method is selected, network training is also required, using the existing low-resolution images and the corresponding high-resolution images for training to learn the mapping relationship from low resolution to high resolution;

[0012] Step S6, super-resolution reconstruction: using a super-resolution reconstruction method of a deep learning method to map the low-resolution image to a high-resolution space and generate a reconstructed high-resolution image based on the output of the deep learning model;

[0013] Step S7, post-processing: performing post-processing on the reconstructed high-resolution image, the post-processing specifically includes artifact removal, detail enhancement, edge enhancement, etc., to further improve the image quality;

[0014] Furthermore, in step S1, data acquisition also includes the following specific steps:

[0015] Step S1-1, before performing an MRI scan, the scanning area and scanning direction need to be determined, and the doctor or technician needs to discuss with the patient to determine the scanning range and the required scanning sequence;

[0016] Step S1-2, the patient is asked to lie on the MRI scanning bed and needs to stay as still as possible. The doctor or technician needs to help the patient maintain the correct posture and use markers or positioning devices to ensure the required scanning position and angle;

[0017] Step S1-3, the MRI system uses coils to transmit and receive signals, and the doctor or technician needs to select the appropriate coil and correctly place it on the specific area of ​​the patient's body according to the needs of the scanning area;

[0018] Step S1-4, according to the doctor or technician's instructions, the operator will enter the control interface of the MRI system and set the parameters required for scanning, including scanning sequence, repetition time, echo time, slice thickness and inter-slice interval, which determine the characteristics and quality of the scanned image;

[0019] Step S1-5, when the scanning parameters are set, the patient is pushed into the magnetic field of the MRI device. During the scanning process, the device generates a magnetic field and radio waves, and obtains image data by stimulating the water molecules in the patient's body and receiving signals. The scanning time may last from several minutes to tens of minutes depending on the type of scanning and the required image details;

[0020] Step S1-6, after completing the scan, the MRI system saves the acquired image data in a computer, and the data can be transmitted over a network or stored in a medical information system for doctors and other medical professionals to interpret and diagnose the images;

[0021] Furthermore, in step S2, the image preprocessing further includes the following specific steps:

[0022] Step S2-1, if there is noise in the MRI image, a denoising algorithm can be applied to reduce the influence of the noise. The denoising method used is Gaussian filtering, which can reduce random noise in the image and improve the effect of subsequent processing;

[0023] Step S2-2, in the 3D MRI data, there may be inconsistency in image brightness and contrast, and image calibration using a histogram matching calibration method can make the image have consistent brightness and contrast;

[0024] Step S2-3, sometimes artifacts may appear in the MRI image, such as frosting artifacts, bone artifacts, etc. By using an artifact removal algorithm, the influence of the artifacts can be reduced and the image quality can be improved;

[0025] Step S2-4: if it is necessary to compare and analyze MRI images of multiple different patients or scanning devices, spatial normalization is usually required. This step involves normalizing the images in position, scale and direction to ensure spatial consistency between different images;

[0026] Step S2-5, in some cases, to better reveal structures and details in the image, contrast-limited adaptive histogram equalization may be applied;

[0027] Step S2-6, if it is necessary to compare or register multiple MRI images of different time points or different patients, image registration is a necessary step, which involves spatially or morphologically transforming the images so that they are aligned in the same coordinate system;

[0028] Furthermore, in step S3, feature extraction also includes the following specific steps:

[0029] Step S3-1, first, it is necessary to select a suitable region or a suitable structure through manual labeling and automatic segmentation algorithm;

[0030] Step S3-2, for automatic feature extraction, it is necessary to perform regional segmentation on the image to separate different tissue structures or anatomical parts. The segmentation methods used include threshold segmentation, region growing, edge detection, morphology-based segmentation, etc.;

[0031] Step S3-3, select a suitable feature extraction algorithm according to the actual problem and research purpose.

[0032] Step S3-4, representing and encoding the extracted features, using vectors and tensors to represent the features;

[0033] Step S3-5, selecting and reducing the dimension of the extracted features to reduce the dimension and redundant information of the features, and the feature selection and dimension reduction methods used include variance threshold selection, correlation analysis, principal component analysis, and linear discriminant analysis;

[0034] Furthermore, in step S3-3, the feature extraction method includes:

[0035] Statistical characteristics: including mean, standard deviation, maximum, minimum, etc.;

[0036] Texture features: including gray-level co-occurrence matrix, gray-level value histogram, gray-level co-occurrence matrix, gray-level co-occurrence matrix, gray-level co-occurrence matrix and gray-level co-occurrence matrix, etc.;

[0037] Shape characteristics: including volume, surface area, shape index, convexity, etc.;

[0038] Nearest neighbor algorithm: including nearest neighbor, k nearest neighbor, etc.;

[0039] Principal component analysis: used to reduce dimension and extract main features;

[0040] Deep learning methods: including convolutional neural networks for feature extraction and learning.

[0041] Furthermore, in step S7, the reconstructed image is evaluated, and the evaluation indicators such as peak signal-to-noise ratio and structural similarity index are used to perform adjustment and optimization to ensure that the reconstruction result meets expectations.

[0042] An electronic device includes a computer, wherein the computer is provided with a central processing unit, a graphics processing unit and a memory, the memory stores an operating program, the operating program runs on the central processing unit, and the operating program includes a deep learning module, a data set module, a network architecture module, a training data processing module, a loss function module, an optimizer module, a learning rate scheduling module, a preprocessing module and a post-processing module.

[0043] A storage medium including computer hard drives, solid-state drives, USB drives, and network storage.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The present invention uses deep learning methods to adaptively learn complex correlations and texture features between different structures and tissues through end-to-end learning. In contrast, traditional methods often rely on some simplified assumptions and may perform poorly for complex image content;

[0046] 2. The present invention uses deep learning methods to automatically learn the features and representations of images through large-scale data training and automatic optimization processes, without the need to manually select and adjust a large number of parameters, thus reducing the need for manual intervention and artificial assumptions;

[0047] 3. The present invention uses deep learning methods to achieve effective recovery of high-frequency details by learning the mapping relationship between a large amount of low-resolution and corresponding high-resolution data, which makes the reconstructed image richer in texture and details and improves the image quality. Traditional methods are usually based on interpolation or statistical models, which limits the ability to recover missing high-frequency details. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention is a schematic diagram of the electronic equipment and storage medium flow of a three-dimensional magnetic resonance image super-resolution reconstruction method. DETAILED DESCRIPTION

[0049] 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.

[0050] See also Figure 1 , the present invention provides a technical solution:

[0051] A three-dimensional magnetic resonance image super-resolution reconstruction method comprises the following steps:

[0052] Step S1, data acquisition: obtaining a low-resolution 3D MRI image as input data, and obtaining a corresponding high-resolution image as reference data according to requirements;

[0053] Step S2, image preprocessing: preprocessing the input low-resolution image, including denoising, grayscale standardization, and artifact removal to reduce noise and improve image quality;

[0054] Step S3, feature extraction: using a local statistical feature extraction method to extract features from the low-resolution image, the extracted features specifically include texture, shape, structure and other features;

[0055] Step S4, method selection: selecting a super-resolution reconstruction method based on a deep learning method;

[0056] Step S5, network training: After the deep learning method is selected, network training is also required, using the existing low-resolution images and the corresponding high-resolution images for training to learn the mapping relationship from low resolution to high resolution;

[0057] Step S6, super-resolution reconstruction: using a super-resolution reconstruction method of a deep learning method to map the low-resolution image to a high-resolution space and generate a reconstructed high-resolution image based on the output of the deep learning model;

[0058] Step S7, post-processing: performing post-processing on the reconstructed high-resolution image, the post-processing specifically includes artifact removal, detail enhancement, edge enhancement, etc., to further improve the image quality;

[0059] In the present invention, in step S1, data acquisition further includes the following specific steps:

[0060] Step S1-1, before performing an MRI scan, the scanning area and scanning direction need to be determined, and the doctor or technician needs to discuss with the patient to determine the scanning range and the required scanning sequence;

[0061] Step S1-2, the patient is asked to lie on the MRI scanning bed and needs to stay as still as possible. The doctor or technician needs to help the patient maintain the correct posture and use markers or positioning devices to ensure the required scanning position and angle;

[0062] Step S1-3, the MRI system uses coils to transmit and receive signals, and the doctor or technician needs to select the appropriate coil and correctly place it on the specific area of ​​the patient's body according to the needs of the scanning area;

[0063] Step S1-4, according to the doctor or technician's instructions, the operator will enter the control interface of the MRI system and set the parameters required for scanning, including scanning sequence, repetition time, echo time, slice thickness and inter-slice interval, which determine the characteristics and quality of the scanned image;

[0064] Step S1-5, when the scanning parameters are set, the patient is pushed into the magnetic field of the MRI device. During the scanning process, the device generates a magnetic field and radio waves, and obtains image data by stimulating the water molecules in the patient's body and receiving signals. The scanning time may last from several minutes to tens of minutes depending on the type of scanning and the required image details;

[0065] Step S1-6, after completing the scan, the MRI system saves the acquired image data in a computer, and the data can be transmitted over a network or stored in a medical information system for doctors and other medical professionals to interpret and diagnose the images;

[0066] In the present invention, in step S2, the image preprocessing further includes the following specific steps:

[0067] Step S2-1, if there is noise in the MRI image, a denoising algorithm can be applied to reduce the influence of the noise. The denoising method used is Gaussian filtering, which can reduce random noise in the image and improve the effect of subsequent processing;

[0068] Step S2-2, in the 3D MRI data, there may be inconsistency in image brightness and contrast, and image calibration using a histogram matching calibration method can make the image have consistent brightness and contrast;

[0069] Step S2-3, sometimes artifacts may appear in the MRI image, such as frosting artifacts, bone artifacts, etc. By using an artifact removal algorithm, the influence of the artifacts can be reduced and the image quality can be improved;

[0070] Step S2-4: if it is necessary to compare and analyze MRI images of multiple different patients or scanning devices, spatial normalization is usually required. This step involves normalizing the images in position, scale and direction to ensure spatial consistency between different images;

[0071] Step S2-5, in some cases, to better reveal structures and details in the image, contrast-limited adaptive histogram equalization may be applied;

[0072] Step S2-6, if it is necessary to compare or register multiple MRI images of different time points or different patients, image registration is a necessary step, which involves spatially or morphologically transforming the images so that they are aligned in the same coordinate system;

[0073] In the present invention, in step S3, feature extraction further includes the following specific steps:

[0074] Step S3-1, first, it is necessary to select a suitable region or a suitable structure through manual labeling and automatic segmentation algorithm;

[0075] Step S3-2, for automatic feature extraction, it is necessary to perform regional segmentation on the image to separate different tissue structures or anatomical parts. The segmentation methods used include threshold segmentation, region growing, edge detection, morphology-based segmentation, etc.;

[0076] Step S3-3, select a suitable feature extraction algorithm according to the actual problem and research purpose.

[0077] Step S3-4, representing and encoding the extracted features, using vectors and tensors to represent the features;

[0078] Step S3-5, selecting and reducing the dimension of the extracted features to reduce the dimension and redundant information of the features, and the feature selection and dimension reduction methods used include variance threshold selection, correlation analysis, principal component analysis, and linear discriminant analysis;

[0079] In the present invention, in step S3-3, the feature extraction method includes:

[0080] Statistical characteristics: including mean, standard deviation, maximum, minimum, etc.;

[0081] Texture features: including gray-level co-occurrence matrix, gray-level value histogram, gray-level co-occurrence matrix, gray-level co-occurrence matrix, gray-level co-occurrence matrix and gray-level co-occurrence matrix, etc.;

[0082] Shape characteristics: including volume, surface area, shape index, convexity, etc.;

[0083] Nearest neighbor algorithm: including nearest neighbor, k nearest neighbor, etc.;

[0084] Principal component analysis: used to reduce dimension and extract main features;

[0085] Deep learning methods: including convolutional neural networks for feature extraction and learning;

[0086] In the present invention, in step S7, the reconstructed image is evaluated, and the evaluation indicators such as peak signal-to-noise ratio and structural similarity index are used to perform adjustment and optimization to ensure that the reconstruction result meets expectations.

[0087] An electronic device includes a computer, wherein the computer is provided with a central processing unit, a graphics processing unit and a memory, wherein the memory stores an operating program, and the operating program runs on the central processing unit. The operating program includes a deep learning module, a data set module, a network architecture module, a training data processing module, a loss function module, an optimizer module, a learning rate scheduling module, a preprocessing module and a post-processing module. The computer is used to execute deep learning algorithms and process large amounts of data, and cooperates with the central processing unit and the graphics processing unit to accelerate the training and reasoning process.

[0088] A storage medium includes a computer hard disk, a solid-state hard disk, a USB drive and a network storage. The computer hard disk and the solid-state hard disk are commonly used storage media with large capacity and fast data reading and writing speed. The USB drive is portable and easy to connect to different computers. This storage medium can be used for temporary storage or transmission of data. In addition, the data is stored in a dedicated network storage. This storage medium has large capacity and reliability and is suitable for long-term storage and data backup.

[0089] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0090] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional magnetic resonance image super-resolution reconstruction method, characterized in that: The following steps are involved: Step S1, data acquisition: obtaining a low-resolution 3D MRI image as input data, and obtaining a corresponding high-resolution image as reference data according to requirements; Step S2, image preprocessing: preprocessing the input low-resolution image, including denoising, grayscale standardization, and artifact removal to reduce noise and improve image quality; Step S3, feature extraction: using a local statistical feature extraction method to extract features from the low-resolution image, the extracted features specifically include texture, shape, structure and other features; Step S4, method selection: selecting a super-resolution reconstruction method based on a deep learning method; Step S5, network training: After the deep learning method is selected, network training is also required, using the existing low-resolution images and the corresponding high-resolution images for training to learn the mapping relationship from low resolution to high resolution; Step S6, super-resolution reconstruction: using a super-resolution reconstruction method of a deep learning method to map the low-resolution image to a high-resolution space and generate a reconstructed high-resolution image based on the output of the deep learning model; Step S7, post-processing: performing post-processing on the reconstructed high-resolution image. The post-processing specifically includes artifact removal, detail enhancement, edge enhancement, etc., to further improve the image quality.

2. A three-dimensional magnetic resonance imaging super-resolution reconstruction method according to claim 1, characterized in that: In step S1, data acquisition also includes the following specific steps: Step S1-1, before performing an MRI scan, the scanning area and scanning direction need to be determined, and the doctor or technician needs to discuss with the patient to determine the scanning range and the required scanning sequence; Step S1-2, the patient is asked to lie on the MRI scanning bed and needs to stay as still as possible. The doctor or technician needs to help the patient maintain the correct posture and use markers or positioning devices to ensure the required scanning position and angle; Step S1-3, the MRI system uses coils to transmit and receive signals, and the doctor or technician needs to select the appropriate coil and correctly place it on the specific area of ​​the patient's body according to the needs of the scanning area; Step S1-4, according to the doctor or technician's instructions, the operator will enter the control interface of the MRI system and set the parameters required for scanning, including scanning sequence, repetition time, echo time, slice thickness and inter-slice interval, which determine the characteristics and quality of the scanned image; Step S1-5, when the scanning parameters are set, the patient is pushed into the magnetic field of the MRI device. During the scanning process, the device generates a magnetic field and radio waves, and obtains image data by stimulating the water molecules in the patient's body and receiving signals. The scanning time may last from several minutes to tens of minutes depending on the type of scanning and the required image details; Step S1-6, after completing the scan, the MRI system saves the acquired image data in the computer, and the data can be transmitted over the network or stored in the medical information system for doctors and other medical professionals to interpret and diagnose the images.

3. The method for super-resolution reconstruction of three-dimensional magnetic resonance images according to claim 1, characterized in that: In step S2, the image preprocessing further includes the following specific steps: Step S2-1, if there is noise in the MRI image, a denoising algorithm can be applied to reduce the influence of the noise. The denoising method used is Gaussian filtering, which can reduce random noise in the image and improve the effect of subsequent processing; Step S2-2, in the 3D MRI data, there may be inconsistency in image brightness and contrast, and image calibration using a histogram matching calibration method can make the image have consistent brightness and contrast; Step S2-3, sometimes artifacts may appear in the MRI image, such as frosting artifacts, bone artifacts, etc. By using an artifact removal algorithm, the influence of the artifacts can be reduced and the image quality can be improved; Step S2-4: if it is necessary to compare and analyze MRI images of multiple different patients or scanning devices, spatial normalization is usually required. This step involves normalizing the images in position, scale and direction to ensure spatial consistency between different images; Step S2-5, in some cases, to better reveal structures and details in the image, contrast-limited adaptive histogram equalization may be applied; Step S2-6: If it is necessary to compare or register multiple MRI images of different time points or different patients, image registration is a necessary step. This process involves spatially or morphologically transforming the images so that they are aligned in the same coordinate system.

4. The method for super-resolution reconstruction of three-dimensional magnetic resonance images according to claim 1, characterized in that: In step S3, feature extraction also includes the following specific steps: Step S3-1, first, it is necessary to select a suitable region or a suitable structure through manual labeling and automatic segmentation algorithm; Step S3-2, for automatic feature extraction, it is necessary to perform regional segmentation on the image to separate different tissue structures or anatomical parts. The segmentation methods used include threshold segmentation, region growing, edge detection, morphology-based segmentation, etc.; Step S3-3, select a suitable feature extraction algorithm according to the actual problem and research purpose. Step S3-4, representing and encoding the extracted features, using vectors and tensors to represent the features; Step S3-5, selecting and reducing the dimension of the extracted features to reduce the dimension and redundant information of the features. The feature selection and dimension reduction methods used include variance threshold selection, correlation analysis, principal component analysis, and linear discriminant analysis.

5. A three-dimensional magnetic resonance imaging super-resolution reconstruction method according to claim 4, characterized in that: In step S3-3, the feature extraction method includes: Statistical characteristics: including mean, standard deviation, maximum, minimum, etc.; Texture features: including gray-level co-occurrence matrix, gray-level value histogram, gray-level co-occurrence matrix, gray-level co-occurrence matrix, gray-level co-occurrence matrix and gray-level co-occurrence matrix, etc.; Shape characteristics: including volume, surface area, shape index, convexity, etc.; Nearest neighbor algorithm: including nearest neighbor, k nearest neighbor, etc.; Principal component analysis: used to reduce dimension and extract main features; Deep learning methods: including convolutional neural networks for feature extraction and learning.

6. The method for super-resolution reconstruction of three-dimensional magnetic resonance images according to claim 1, characterized in that: In step S7, the reconstructed image is evaluated, and the evaluation indicators are peak signal-to-noise ratio and structural similarity index, and the image is adjusted and optimized to ensure that the reconstruction result meets expectations.

7. An electronic device, characterized in that: The invention comprises a computer, wherein the computer is provided with a central processing unit, a graphics processing unit and a memory, wherein the memory stores an operating program, and the operating program runs on the central processing unit. The operating program comprises a deep learning module, a data set module, a network architecture module, a training data processing module, a loss function module, an optimizer module, a learning rate scheduling module, a preprocessing module and a post-processing module.

8. A storage medium, characterized in that: This includes computer hard drives, solid-state drives, USB drives, and network storage.

Images