Method and device for correcting geometric distortion of MRI images

Abstract

The present invention discloses a method and device for correcting geometric distortion of MRI images, which relates to the field of medical image processing, and includes the steps of: acquiring echo planar imaging diffusion weighted imaging and fast spin echo diffusion weighted imaging from a magnetic resonance imaging device; performing preprocessing on the echo planar imaging diffusion weighted imaging and the fast spin echo diffusion weighted imaging to obtain preprocessed images. Main technical solutions and effects: Since geometric distortion will directly affect the accurate presentation of anatomical structures and lesion areas in MRI images, the present invention can provide more real and accurate image data by effectively correcting the geometric distortion in the echo planar imaging diffusion weighted imaging images, thereby improving the diagnostic accuracy of doctors, helping doctors better judge the location, size and shape of lesions, and further formulating more reasonable treatment plans.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and particularly to a method and device for correcting geometric distortion of MRI images. Background Art

[0002] Magnetic Resonance Imaging (MRI) technology is one of the widely used imaging technologies in modern medical imaging. MRI technology generates detailed images of internal tissues through strong magnetic fields and radiofrequency pulses, and has become an indispensable tool for diagnosing various diseases (especially diseases of the brain, spine, tumors, etc.). Since MRI images can provide high-resolution tissue details and do not use harmful X-rays, they are widely used in clinical diagnosis.

[0003] However, MRI images, especially Echo-Planar Imaging Diffusion-Weighted Imaging (EPI-DWI) images, often encounter geometric distortion problems. This distortion is caused by various factors, including magnetic field inhomogeneity, patient movement, limitations of the imaging device itself, etc. Geometric distortion will cause distortion of the shape, position, and size of anatomical structures in the image, thereby affecting the doctor's accurate judgment of the location, size, and shape of the lesion and reducing the diagnostic accuracy;

[0004] Currently, methods based on image processing algorithms, such as model-based correction algorithms, usually assume that the distortion of the image has a certain regularity. However, in actual situations, the geometric distortion of Echo-Planar Imaging Diffusion-Weighted Imaging images is affected by a variety of factors, and these algorithms are difficult to accurately model and correct all types of distortion. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to solve the technical problem that in the prior art, MRI images, especially Echo-Planar Imaging Diffusion-Weighted Imaging images, often encounter geometric distortion problems, and to provide a method, device, equipment, and medium for correcting geometric distortion of MRI images.

[0006] One or more embodiments of this specification simultaneously relate to a device for correcting geometric distortion of MRI images, an electronic device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

[0007] Technical Solution:

[0008] In a first aspect, the present application proposes a method for correcting geometric distortion of MRI images, including the steps of:

[0009] Obtain Echo-Planar Imaging Diffusion-Weighted Imaging and Fast Spin-Echo Diffusion-Weighted Imaging from a magnetic resonance imaging device;

[0010] Preprocess the echo planar imaging diffusion weighted imaging (EPI-DWI) and fast spin echo diffusion weighted imaging (FSE-DWI) to obtain preprocessed images, where the preprocessing includes a training preprocessing method and an inference preprocessing method;

[0011] Input the preprocessed images into a trained distortion parameter prediction model to obtain distortion parameters;

[0012] Calibrate the initially acquired EPI-DWI using the distortion parameters to obtain distortion-free EPI-DWI.

[0013] Preferably, obtaining the EPI-DWI and FSE-DWI from a magnetic resonance imaging (MRI) device includes:

[0014] Performing diffusion weighted imaging using an echo planar imaging sequence by configuring the MRI device to obtain EPI-DWI;

[0015] Performing diffusion weighted imaging using a fast spin echo sequence by configuring the MRI device to obtain FSE-DWI.

[0016] Preferably, the training preprocessing method includes image value range normalization, random transformation of image contrast, random rotation of the image, random flipping of the image, cutting out the effective region of the image, resampling the image resolution, and randomly cutting out fixed-size blocks of the image;

[0017] The inference preprocessing method includes image value range normalization, cutting out the effective region of the image, and resampling the image resolution.

[0018] Preferably, before inputting the preprocessed images into a trained distortion parameter prediction model to obtain distortion parameters, training the distortion parameter prediction model includes:

[0019] Obtain historical MRI data, including EPI-DWI and FSE-DWI data;

[0020] Screen and preprocess the historical MRI data;

[0021] Divide the sorted data into a training set, a validation set, and a test set, where the training set is used for model training, the validation set is used to adjust the model parameters and evaluate the model performance during training, and the test set is used to finally evaluate the model's generalization ability and correction effect;

[0022] During training, regularly evaluate the model using the validation set data and adjust the model's hyperparameters according to the evaluation results;

[0023] When the performance of the model on the validation set reaches the preset threshold or no longer improves, stop the training and save the model parameters at this time.

[0024] Preferably, before obtaining the distortion parameters by inputting the preprocessed image into the trained distortion parameter prediction model, training the distortion parameter prediction model includes:

[0025] The loss function for model training, which includes three loss functions: structural similarity index, mutual information, and normalized cross - correlation. After weighting, the final loss function formula is:

[0026] ;

[0027] W1, W2, and W3 are the weight coefficients of the structural similarity index, mutual information, and normalized cross - correlation respectively. SSIM is the structural similarity index loss function, MI is the mutual information loss function, and NCC is the normalized cross - correlation loss function;

[0028] The optimizer for model training adopts a combination of adaptive momentum estimation with weight decay optimizer (AdamW) and look - ahead optimization (Lookahead).

[0029] Preferably, inputting the preprocessed image into the trained distortion parameter prediction model to obtain the distortion parameters includes:

[0030] Input the planar echo - planar imaging diffusion - weighted imaging and fast spin - echo diffusion - weighted imaging for preprocessing to obtain the preprocessed image;

[0031] Perform a stitching operation on the preprocessed image obtained by preprocessing the planar echo - planar imaging diffusion - weighted imaging and the preprocessed image obtained by preprocessing the fast spin - echo diffusion - weighted imaging;

[0032] Successively obtain the encoded features at three scales through three sliding window transformation modules (Swin Transformer Block) and pooling layers;

[0033] Pass the encoded features at the three scales through a decoding module composed of a residual convolution, a rectified linear unit activation layer (Relu activation layer), and an upsampling layer to obtain the decoded features;

[0034] Stitch the 3 decoded features, then perform decoding through a convolutional layer and upsample 4 times to output the distortion parameters.

[0035] Preferably, for performance testing of the trained model, it also includes an inference stage, which includes:

[0036] It is necessary to annotate the initial planar echo - planar imaging diffusion - weighted imaging;

[0037] The trained model is tested using the echo planar imaging diffusion weighted imaging after annotation to obtain the corrected echo planar imaging diffusion weighted imaging;

[0038] Calculate the error index between the corrected echo planar imaging diffusion weighted imaging and the true echo planar imaging diffusion weighted imaging.

[0039] Preferably, the initially acquired echo planar imaging diffusion weighted imaging is calibrated by distortion parameters to obtain distortion-free echo planar imaging diffusion weighted imaging, including:

[0040] Perform abnormal parameter correction and global Gaussian smoothing on the distortion parameters;

[0041] Re-sample the echo planar imaging diffusion weighted imaging image three-dimensionally according to the smoothed distortion parameters to achieve geometric distortion correction.

[0042] Preferably, the training preprocessing method and the inference preprocessing method are respectively for the training stage and the inference stage of the model.

[0043] In a second aspect, an embodiment of the present invention provides a device for correcting geometric distortion of MRI images, including:

[0044] An acquisition module, configured to acquire echo planar imaging diffusion weighted imaging and fast spin echo diffusion weighted imaging from a magnetic resonance imaging device;

[0045] A data preprocessing module, configured to preprocess the echo planar imaging diffusion weighted imaging and the fast spin echo diffusion weighted imaging to obtain preprocessed images, where the preprocessing includes a training preprocessing method and an inference preprocessing method;

[0046] A distortion parameter prediction module, configured to input the preprocessed images into a trained distortion parameter prediction model to obtain distortion parameters;

[0047] A geometric distortion calibration module, configured to calibrate the initially acquired echo planar imaging diffusion weighted imaging by the distortion parameters to obtain distortion-free echo planar imaging diffusion weighted imaging.

[0048] In a third aspect, an embodiment of the present invention provides an electronic device, including a processor and a memory. Among them, the memory is used to store one or more computer programs; when the one or more computer programs stored in the memory are executed by the processor, the electronic device can implement the method of any possible design in the first aspect above.

[0049] In a fourth aspect, the present invention provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method in any one of the above embodiments is implemented.

[0050] In a fifth aspect, an embodiment of the present invention further provides a computer program product. When the computer program product runs on an electronic device, the electronic device is enabled to execute the method of any possible design in any of the above aspects.

[0051] Beneficial effects: The present invention adopts deep learning technology and can automatically learn the complex features and patterns of MRI images. In particular, it shows strong adaptability in the geometric distortion correction of echo-planar imaging diffusion-weighted imaging (EPI-DWI) images. By using turbo spin-echo diffusion-weighted imaging (TSE-DWI) images as reference images and combining advanced deep neural network structures (such as convolution, residual convolution, sliding window transformation modules, etc.), the present invention can accurately identify and correct various types of geometric distortions, significantly improving the accuracy and consistency of the correction results.

[0052] Compared with traditional methods based on hardware adjustment, the present invention does not rely on expensive hardware device adjustment but realizes image correction through software algorithms, having significant cost advantages. This method can not only reduce the maintenance cost of MRI devices but also reduce the debugging requirements of clinical operators for device hardware, lowering the technical threshold.

[0053] The present invention uses deep learning technology for image distortion correction, having strong generality and adaptability. Whether it is different models of MRI devices or individual differences among different patients, a sufficiently trained model can effectively adapt to provide consistent and reliable correction effects, enabling the present invention to be widely applied in multiple clinical scenarios and having strong promotion value.

[0054] The image processing method based on deep learning can realize an automated correction process, reducing manual intervention and greatly improving the processing efficiency. The trained model can complete the correction of newly acquired images in a short time, saving doctors' diagnosis time. In clinical applications, especially in emergency and high-load diagnosis and treatment scenarios, this can significantly improve work efficiency and reduce the risk of misdiagnosis and missed diagnosis.

[0055] Since geometric distortion will directly affect the accurate presentation of anatomical structures and lesion areas in MRI images, by effectively correcting the geometric distortion in echo-planar imaging diffusion-weighted imaging images, the present invention can provide more real and accurate image data, thereby improving doctors' diagnostic accuracy, helping doctors better judge the location, size, and shape of lesions, and further formulating more reasonable treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the method framework provided by the present invention;

[0057] Figure 2Provide the overall flowchart for the present invention;

[0058] Figure 3 Provide the two-dimensional schematic diagram of the principle of the method for cutting out the effective image area for the present invention;

[0059] Figure 4 Provide the schematic diagram of the network structure of the distortion parameter prediction model for the present invention;

[0060] Figure 5 Provide the schematic diagram of the training process of the distortion parameter prediction model for the present invention;

[0061] Figure 6 Is the schematic diagram of the structure of the device provided by an embodiment of the present application;

[0062] Figure 7 Is the structural block diagram of an electronic device provided by an embodiment of the present application. Specific embodiments

[0063] To make the technical solution of the present invention clearer, the following further describes the present invention in detail with specific embodiments in conjunction with the accompanying drawings.

[0064] Embodiment 1

[0065] To make the purpose, technical solution and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein shall have the ordinary meaning understood by those of ordinary skill in the art to which the present invention belongs. The words such as "including" used herein mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0066] Regarding the problems existing in the prior art, such as Figure 1 As shown, a method for correcting geometric distortion of MRI images includes the steps:

[0067] Step S101: Obtain echo planar imaging diffusion - weighted imaging (EPI - DWI) and fast spin - echo diffusion - weighted imaging (FSE - DWI) from a magnetic resonance imaging device; EPI - DWI and FSE - DWI are images obtained by two common diffusion - weighted imaging methods; Echo planar imaging (EPI) often shows significant distortion during DWI scanning, so it needs to be corrected. Obtaining the data of these two types of imaging mainly aims to reduce the distortion effect and improve the image quality by taking advantage of their characteristics under different imaging modalities.

[0068] Step S102: Pre - process the EPI - DWI and FSE - DWI to obtain pre - processed images, where the pre - processing includes a training pre - processing method and an inference pre - processing method; The training pre - processing method is pre - processing through a machine - learning model, which may include data standardization, denoising, image alignment, etc. The purpose is to convert the input data into a format that can be better used for prediction. The purpose of pre - processing is to provide clearer and more consistent image data for subsequent steps. Especially in the process involving a machine - learning model, the quality of the input data has a great impact on the effect of the model.

[0069] Step S103: Input the pre - processed images into a trained distortion parameter prediction model to obtain distortion parameters; Input the pre - processed images into a trained distortion parameter prediction model, which is based on a deep - learning convolutional neural network (CNN). The distortion parameter refers to the geometric distortion in the image caused by the device or imaging method (such as EPI scanning). The predicted distortion parameters will be provided to the next step for precise image correction.

[0070] Step S104: Calibrate the initially obtained EPI - DWI with the distortion parameters to obtain distortion - free EPI - DWI. Use the distortion parameters to correct the original EPI - DWI, thereby obtaining a distortion - free image. The key to this step is to use image - processing algorithms (such as affine transformation, non - linear transformation, etc.) to correct the geometric distortion in the image according to the predicted distortion parameters, ensuring that the final image is more accurate and clear.

[0071] In some preferred embodiments, obtaining the EPI - DWI and FSE - DWI from a magnetic resonance imaging device includes:

[0072] Perform diffusion - weighted imaging through the echo planar imaging sequence by configuring the magnetic resonance imaging device to obtain EPI - DWI;

[0073] Perform diffusion - weighted imaging through the fast spin - echo sequence by configuring the magnetic resonance imaging device to obtain FSE - DWI.

[0074] Specifically, the Echo-Planar Imaging (EPI) sequence is a fast magnetic resonance imaging technique, especially suitable for Diffusion-Weighted Imaging (DWI). Its basic principle is to generate high-spatial-resolution images by rapidly acquiring signals from multiple slices. When performing diffusion-weighted imaging, EPI can quickly capture the diffusion characteristics of water molecules in tissues;

[0075] The Turbo Spin Echo (TSE) sequence is another sequence method for diffusion-weighted imaging. Compared with the echo-planar imaging sequence, turbo spin echo imaging can obtain higher image quality through a longer acquisition time and optimized sequence design, especially having advantages in solving distortions and artifacts that may occur in the echo-planar imaging sequence method.

[0076] In some preferred embodiments, the training preprocessing method includes image value range normalization, random transformation of image contrast, random rotation of images, random flipping of images, cutting out the effective region of images, resampling of image resolution, and randomly cutting out fixed-size blocks of images;

[0077] The inference preprocessing method includes image value range normalization, cutting out the effective region of images, and resampling of image resolution.

[0078] Specifically, the training preprocessing method includes the following ways:

[0079] Image value range normalization maps the numerical range of the image to between 0 and 1. This normalization operation can reduce the numerical differences between different images, so that subsequent model training is not affected by the original image value range;

[0080] Random transformation of image contrast increases the diversity of images and enhances the generalization ability of the model. By linearly stretching or scaling the pixel values of the image, the contrast of the image is adjusted. This transformation helps to simulate scenarios under different image quality conditions;

[0081] Random rotation of images enhances the robustness of the model, enabling the model to process images in different directions. The image is rotated three-dimensionally at a random angle to simulate rotational transformations in different directions and enhance the transformation invariance of the model.

[0082] Random flipping of images further enhances the diversity of training data. The image is randomly flipped left-right, front-back, and up-down. In this way, the symmetry training of the image can be increased, and the generalization ability of the model can be improved.

[0083] The irrelevant regions of the image (such as the background) are cut out from the valid region of the image, focusing on the valid part of the image. The common foreground region (i.e., the important region included in both) in the fast spin echo diffusion weighted imaging image and the echo planar imaging diffusion weighted imaging image is found, and then cropped according to the minimum bounding rectangle of the valid region, which can reduce the interference of background noise on subsequent processing.

[0084] The image resolution is resampled to a standard spatial resolution. Through trilinear interpolation, the image is resampled and its resolution is adjusted to 1mm*1mm*1mm. This operation ensures that all input images have a consistent resolution, facilitating subsequent model training and inference.

[0085] Combined Figure 3 , randomly cut out fixed-size image blocks from the image to enhance data diversity and avoid the model relying too much on the overall structure of the image. If the image block exceeds the boundary of the original image, it is filled with zero values. The purpose of this is to generate more local region data and avoid over-reliance on the global features of the image.

[0086] The inference preprocessing method includes the following ways:

[0087] The specific method of image value range normalization is the same as that in the training preprocessing. By normalizing the pixel values of the image to the [0,1] interval, the range of image data during the inference process is standardized.

[0088] The operation of cutting out the valid region of the image is the same as that in the training preprocessing. The image is cropped according to the common valid region, and only the meaningful part of the image is retained for subsequent processing.

[0089] For the images in the inference process, the same resolution standard (1mm*1mm*1mm) as in the training stage is also used to ensure that the model can obtain a consistent data format during inference.

[0090] In some preferred embodiments, before obtaining the distortion parameters by inputting the preprocessed image into the trained distortion parameter prediction model, the training of the distortion parameter prediction model includes:

[0091] Obtain historical magnetic resonance imaging data, including echo planar imaging diffusion weighted imaging and fast spin echo diffusion weighted imaging data;

[0092] Screen and preprocess the historical magnetic resonance imaging data;

[0093] Divide the organized data into a training set, a validation set, and a test set. Among them, the training set is used for model training, the validation set is used to adjust the model's parameters and evaluate the model's performance during training, and the test set is used to finally evaluate the model's generalization ability and correction effect;

[0094] During the training process, regularly use the validation set data to evaluate the model, and adjust the model's hyperparameters according to the evaluation results;

[0095] When the model's performance on the validation set reaches the preset threshold or no longer improves, stop training and save the model parameters at this time.

[0096] Specifically, obtain a large amount of existing magnetic resonance imaging data, which includes echo planar imaging diffusion-weighted imaging and fast spin-echo diffusion-weighted imaging. Screen and preprocess the data, and select the samples that meet the requirements from the collected MRI images. Images with poor quality, excessive distortion, or other non-standard images need to be excluded to ensure the quality of the training data;

[0097] Training set: Allocate most of the images to the training set, and the model will use these data for learning. The training set contains images and their labels (for example, images with distortion and corrected images), and the model will adjust its parameters on these data.

[0098] Validation set: Extract a part of the data from the original data as the validation set. The validation set does not participate in the model training, but will be regularly used to evaluate the model's performance during training. The functions of the validation set are:

[0099] Evaluate the real-time performance of the model;

[0100] Adjust the model's hyperparameters (such as learning rate, batch size, etc.);

[0101] Detect whether the model is overfitting;

[0102] Test set: The test set is part of the data used to finally evaluate the model's generalization ability. The test set does not participate in training and validation, and its main purpose is to evaluate the model's performance on unknown data;

[0103] During the training process, regularly use the validation set data to evaluate the model's performance. According to the model's performance on the validation set, evaluate indicators such as the model's accuracy, recall rate, and loss function. Hyperparameter adjustment is based on the evaluation results of the validation set to adjust the model's hyperparameters. Common hyperparameters include:

[0104] Learning rate: Affects the speed of model weight update. By adjusting the learning rate, ensure that the model converges at an appropriate speed;

[0105] Batch size: Controls the number of samples used in each training. A smaller batch size usually brings more randomness and may help the model jump out of local optima;

[0106] Network structure: Adjusts the depth, width of the network, or the size of the convolutional kernel, etc.;

[0107] After each training epoch, evaluate the performance on the validation set. If the performance of the model on the validation set reaches the preset performance standard (e.g., accuracy exceeds 95%), the training process will be terminated. If the performance of the model on the validation set no longer improves and there is no improvement after several consecutive iterations, the "early stopping" strategy can be used to stop the training to prevent overfitting. When the training stops, save the model weights and parameters at this time. The saved model can be used for subsequent inference, testing, or deployed to the actual environment for prediction.

[0108] In some preferred embodiments, in combination with Figure 5 , before inputting the preprocessed image into the trained distortion parameter prediction model to obtain the distortion parameter, training the distortion parameter prediction model includes:

[0109] The loss function for model training, which includes three loss functions: structural similarity index, mutual information, and normalized cross-correlation. After weighting, the final loss function formula is:

[0110] ;

[0111] W1, W2, and W3 are the weight coefficients of the structural similarity index, mutual information, and normalized cross-correlation respectively. SSIM is the structural similarity index loss function, MI is the mutual information loss function, and NCC is the normalized cross-correlation loss function;

[0112] The optimizer for model training adopts a combination of adaptive momentum estimation with weight decay optimizer and look-ahead optimization.

[0113] Specifically, during the training process, the loss function includes the Structural Similarity Index (SSIM), Mutual Information (MI), and Normalized Cross-Correlation (NCC) for image processing tasks, and the specific explanations are as follows: The Structural Similarity Index is a metric for measuring the similarity between two images. It takes into account aspects such as the brightness, contrast, and structure of the images. In image reconstruction tasks, the Structural Similarity Index can help evaluate the structural similarity between the restored image and the original image, and is particularly sensitive to the texture and structural details of the image. Mutual Information measures the amount of information shared between two images. It is a classical method for evaluating the similarity between images, especially suitable for image registration and alignment, and is used to evaluate the overlap or information sharing degree between the restored image and the target image. Normalized Cross-Correlation is a way to measure the similarity between two images. It obtains their similarity by calculating the normalized cross-correlation of the two images. In image alignment and registration tasks, Normalized Cross-Correlation is a commonly used similarity metric, which can help ensure that the model tries to maintain the match with the target image when restoring the image. The three loss functions of the Structural Similarity Index, Mutual Information, and Normalized Cross-Correlation are weighted to obtain a final loss function. Among them, W1, W2, and W3 are the weights of these three loss functions, which determine the importance of each loss function in the final loss calculation. Specifically: W1, W2, W3: The adjustment of these weight values will affect the contribution of each loss function during the training process. For example, if you want the model to pay more attention to the restoration of the image structure, you can increase the weight of the Structural Similarity Index; if you want to strengthen the image registration effect, you can increase the weight of the Mutual Information loss function or Normalized Cross-Correlation.

[0114] The Adaptive Moment Estimation optimizer with Weight Decay is a variant of the Adam optimizer, specifically designed to improve the problem of weight decay (L2 regularization). In the Adam optimizer, weight decay usually affects the gradient calculation when updating parameters, while the Adaptive Moment Estimation optimizer with Weight Decay separates weight decay from the gradient update step, which helps to improve the stability and generalization ability of training. Look-ahead optimization is an improved optimization strategy aimed at improving the performance of the optimizer, especially when using adaptive optimizers. Look-ahead optimization maintains the "historical" solutions of multiple optimizers and periodically adjusts the current solution to prevent getting stuck in local optima, which can improve the stability of the model during training and accelerate convergence.

[0115] In some preferred embodiments, in combination with Figure 4 , the preprocessed image is input into the trained distortion parameter prediction model to obtain distortion parameters, including:

[0116] The input planar echo imaging diffusion - weighted imaging and fast spin - echo diffusion - weighted imaging are pre - processed to obtain pre - processed images;

[0117] The pre - processed image obtained by pre - processing the planar echo imaging diffusion - weighted imaging and the pre - processed image obtained by pre - processing the fast spin - echo diffusion - weighted imaging are subjected to a stitching operation;

[0118] Three scales of encoded features are obtained successively through three sliding window transformation modules and pooling layers;

[0119] The three scales of encoded features are respectively passed through a decoding module composed of a residual convolution, a rectified linear unit activation layer, and an up - sampling layer to obtain decoded features;

[0120] The three decoded features are stitched together and then decoded through a convolutional layer and up - sampled by a factor of 4 to output distortion parameters.

[0121] Specifically, the network accepts two types of magnetic resonance imaging data, namely planar echo imaging diffusion - weighted imaging and fast spin - echo diffusion - weighted imaging;

[0122] Each input image (planar echo imaging diffusion - weighted imaging and fast spin - echo diffusion - weighted imaging) is pre - processed, which usually includes operations such as image normalization, noise removal, and size adjustment. The pre - processed images will be input into the network for subsequent processing;

[0123] The two pre - processed images (planar echo imaging diffusion - weighted imaging and fast spin - echo diffusion - weighted imaging) are stitched together. The stitching can be horizontal or vertical. The purpose is to combine two images from different sources into a complete image for input into the network to utilize the complementary information provided by different imaging methods;

[0124] Sliding window transformation module:

[0125] The stitched image is input into the sliding window transformation module. The sliding window transformation module is a structure based on the self - attention mechanism, which is specifically designed to handle long - range dependencies in images and can better capture global and local information in images;

[0126] The model uses three sliding window transformation modules, and each sliding window transformation module extracts features of different scales. In this way, the network can extract multi - scale information of the image;

[0127] Pooling layer: The output feature maps after each sliding window transformation module will pass through the pooling layer. The pooling operation helps to reduce the spatial size and the size of the feature maps, thereby reducing the computational complexity and the risk of overfitting;

[0128] The decoding module consists of a residual convolution, a rectified linear unit activation layer, and an upsampling layer. The residual convolution helps alleviate the vanishing gradient problem in deep networks. The rectified linear unit activation function increases non-linearity, enabling the model to capture more complex patterns.

[0129] The upsampling layer is used to restore the feature map to a higher spatial resolution, allowing the model to better recover the details of the original image.

[0130] The three decoded feature maps are concatenated. Through this concatenation, the network can combine the encoded information from different scales to obtain a richer feature representation.

[0131] The concatenated feature map will pass through a convolutional layer for further decoding. The convolution operation helps extract the final required features.

[0132] The decoded feature map is upsampled 4 times through the upsampling layer to restore to the required image size.

[0133] Finally, the model outputs distortion parameters, which are used to correct the distortion of the input image and help restore the original, distortion-free image.

[0134] In some preferred embodiments, when performing a performance test on the trained model, it also includes an inference phase, which includes:

[0135] It is necessary to annotate the initial echo planar imaging diffusion-weighted imaging.

[0136] Use the annotated echo planar imaging diffusion-weighted imaging to test the trained model to obtain the corrected echo planar imaging diffusion-weighted imaging.

[0137] Calculate the error metric between the corrected echo planar imaging diffusion-weighted imaging and the true echo planar imaging diffusion-weighted imaging.

[0138] Specifically, the annotation step refers to manually annotating the initial echo planar imaging diffusion-weighted imaging image or automatically annotating it through a certain method. The goal of annotation is to provide a "target" or "label" for each image, that is, to indicate the true state or reference image of the image. These annotations will be used as the ground truth or target image to be compared with the output of the model during the training process. In this process, the images can be manually annotated based on expert knowledge or existing datasets, or a pre-trained model can be used to assist in annotation. The task of the model is to use the annotation information of the image to correct the distorted part based on the features learned from the training data and generate an output that is as close as possible to the true distortion-free image. After correction, the generated image should have higher quality and less distortion.

[0139] Use the trained deep learning model to test the labeled echo-planar imaging diffusion-weighted imaging (EPI-DWI) images. The purpose of this step is to verify whether the model can accurately correct the distortion in the new images, and calculate the error between the corrected EPI-DWI and the true EPI-DWI images, aiming to quantify the quality of the correction effect.

[0140] In some preferred embodiments, calibrating the initially acquired EPI-DWI by distortion parameters to obtain distortion-free EPI-DWI includes:

[0141] Performing abnormal parameter correction and global Gaussian smoothing on the distortion parameters;

[0142] Re-sampling the EPI-DWI image in three dimensions according to the smoothed distortion parameters to achieve geometric distortion correction.

[0143] Specifically, distortion parameter correction usually involves adjusting these parameters to correct the geometric distortion of the image. Some algorithms (such as median filtering or local constraint correction, etc.) are used to smooth or correct the outliers in the parameters. Gaussian smoothing is a commonly used image smoothing technique, mainly used for denoising and reducing the influence of local noise in the image. Here, Gaussian smoothing is applied to the distortion parameters to reduce the influence of local instability and noise on the distortion correction result. By applying a Gaussian kernel to smooth the distortion parameters, the spatial distribution of the distortion parameters can be made more smooth and consistent. The role of the Gaussian filter is to suppress the changes in the local area of the distortion parameters and make the smoothed parameters more suitable for geometric transformation. The next step is to use these smoothed distortion parameters for three-dimensional re-sampling of the image. Using a three-dimensional spatial transformation algorithm, the input EPI-DWI image is spatially adjusted. Through the re-sampling algorithm, each pixel in the image can be re-positioned so that each point in the image is mapped to its correct position according to the corrected distortion parameters.

[0144] In some preferred embodiments, the training preprocessing method and the inference preprocessing method are respectively for the training stage and the inference stage of the model.

[0145] Specifically, the training preprocessing method is used to prepare data in the training stage to make it adapt to the needs of the model and improve the training efficiency and generalization ability of the model. The inference preprocessing method is used to process the input data in the inference stage to ensure that the model can accurately predict the input data.

[0146] In some preferred embodiments, in combination with Figure 6 , a device for correcting geometric distortion of MRI images is also proposed, including:

[0147] An acquisition module 301, configured to acquire echo planar imaging diffusion weighted imaging and fast spin echo diffusion weighted imaging from a magnetic resonance imaging device;

[0148] A data preprocessing module 302, configured to preprocess the echo planar imaging diffusion weighted imaging and the fast spin echo diffusion weighted imaging to obtain preprocessed images, where the preprocessing includes a training preprocessing method and an inference preprocessing method;

[0149] A distortion parameter prediction module 303, configured to input the preprocessed images into a trained distortion parameter prediction model to obtain distortion parameters;

[0150] A geometric distortion calibration module 304, configured to calibrate the initially acquired echo planar imaging diffusion weighted imaging through the distortion parameters to obtain distortion-free echo planar imaging diffusion weighted imaging.

[0151] All relevant contents of each step involved in the above method embodiments can be cited in the function descriptions of the corresponding functional modules, and will not be elaborated here.

[0152] In some other embodiments of the present invention, embodiments of the present invention disclose an electronic device 400, as Figure 7 shown. The electronic device may include: one or more processors 401; a memory 402; a display 403; one or more applications (not shown); and one or more computer programs 404. The above devices may be connected through one or more communication buses 405. The one or more computer programs 404 are stored in the memory 402 and configured to be executed by the one or more processors 401. The one or more computer programs 404 include instructions, and the instructions may be used to execute the steps in Figures 1 to 5 and the corresponding embodiments.

[0153] Through the description of the above embodiments, those skilled in the art can clearly understand that for the convenience and brevity of description, only the above division of each functional module is used as an example. In actual applications, the above functions may be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working processes of the above-described system, device, and unit may refer to the corresponding processes in the foregoing method embodiments and will not be elaborated here.

[0154] In each embodiment of the embodiments of the present invention, each functional unit may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0155] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in the embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes, such as flash memory, mobile hard disk, read-only memory, random access memory, magnetic disk, or optical disc.

[0156] As described above, it is only the specific implementation manner of the embodiments of the present invention, but the protection scope of the embodiments of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed in the embodiments of the present invention should be covered by the protection scope of the embodiments of the present invention. Therefore, the protection scope of the embodiments of the present invention should be subject to the protection scope of the claimed rights.

Claims

1. A method for correcting geometric distortion of MRI images, characterized in that, Including steps: Obtain echo planar imaging diffusion weighted imaging (EPI-DWI) and fast spin echo diffusion weighted imaging (FSE-DWI) from a magnetic resonance imaging (MRI) device; Preprocess the EPI-DWI and FSE-DWI to obtain preprocessed images, where the preprocessing includes a training preprocessing method and an inference preprocessing method; Input the preprocessed images into a trained distortion parameter prediction model to obtain distortion parameters; Calibrate the initially obtained EPI-DWI using the distortion parameters to obtain distortion-free EPI-DWI, including: Input the EPI-DWI and FSE-DWI for preprocessing to obtain preprocessed images; Perform a stitching operation on the preprocessed images obtained by preprocessing the EPI-DWI and the preprocessed images obtained by preprocessing the FSE-DWI; Successively obtain encoded features at three scales through three sliding window transformation modules and pooling layers; Respectively pass the encoded features at the three scales through a decoding module composed of a residual convolution, a rectified linear unit activation layer, and an upsampling layer to obtain decoded features; Stitch the 3 decoded features, then perform decoding through a convolutional layer and upsample by 4 times to output the distortion parameters.

2. The method according to claim 1, wherein Obtain the EPI-DWI and FSE-DWI from the MRI device, including: Configure the MRI device and perform diffusion weighted imaging using an echo planar imaging sequence to obtain EPI-DWI; Configure the MRI device and perform diffusion weighted imaging using a fast spin echo sequence to obtain FSE-DWI.

3. The method according to claim 1, characterized in that, The training preprocessing method includes image value range normalization, random transformation of image contrast, random rotation of the image, random flipping of the image, cutting out the effective region of the image, resampling the image resolution, and randomly cutting out fixed-size blocks of the image; The inference preprocessing method includes image value range normalization, cutting out the effective region of the image, and resampling the image resolution.

4. The method according to claim 1, characterized in that Before inputting the preprocessed images into the trained distortion parameter prediction model to obtain distortion parameters, train the distortion parameter prediction model, including: Obtain historical MRI data, including EPI-DWI and FSE-DWI data; Screen and preprocess the historical MRI data; Divide the sorted data into a training set, a validation set, and a test set. Among them, the training set is used for training the model, the validation set is used to adjust the model parameters and evaluate the model performance during training, and the test set is used to finally evaluate the generalization ability and correction effect of the model; During training, regularly evaluate the model using the validation set data and adjust the hyperparameters of the model according to the evaluation results; When the performance of the model on the validation set reaches a preset threshold or no longer improves, stop training and save the model parameters at this time.

5. The method according to claim 4, characterized in that, Before inputting the preprocessed images into the trained distortion parameter prediction model to obtain distortion parameters, train the distortion parameter prediction model, including: The loss function for model training, which includes three loss functions: structural similarity index, mutual information, and normalized cross - correlation. After weighting, the final loss function formula is as follows: ； W1, W2, and W3 are the weight coefficients of the structural similarity index, mutual information, and normalized cross - correlation respectively. SSIM is the structural similarity index loss function, MI is the mutual information loss function, and NCC is the normalized cross - correlation loss function; The optimizer for model training adopts a combination of adaptive momentum estimation with weight decay optimizer and look - ahead optimization.

6. The method according to claim 5, wherein The performance test of the trained model also includes the inference stage, which includes: It is necessary to annotate the initial echo - planar imaging diffusion - weighted imaging. Use the annotated echo - planar imaging diffusion - weighted imaging to test the trained model to obtain the corrected echo - planar imaging diffusion - weighted imaging. Calculate the error index between the corrected echo - planar imaging diffusion - weighted imaging and the true echo - planar imaging diffusion - weighted imaging.

7. The method according to claim 1, wherein Calibrate the initially acquired echo - planar imaging diffusion - weighted imaging through distortion parameters to obtain distortion - free echo - planar imaging diffusion - weighted imaging, including: Perform abnormal parameter correction and global Gaussian smoothing of the distortion parameters. According to the smoothed distortion parameters, re - sample the echo - planar imaging diffusion - weighted imaging image in three dimensions to achieve geometric distortion correction.

8. The method according to claim 6, characterized in that, The training pre - processing method and the inference pre - processing method are respectively for the training stage and the inference stage of the model.

9. A device for correcting geometric distortion of MRI images, characterized in that, Including: An acquisition module, used to acquire echo - planar imaging diffusion - weighted imaging and fast spin - echo diffusion - weighted imaging from a magnetic resonance imaging device; A data pre - processing module, used to pre - process the echo - planar imaging diffusion - weighted imaging and fast spin - echo diffusion - weighted imaging to obtain pre - processed images. Among them, the pre - processing includes a training pre - processing method and an inference pre - processing method; A distortion parameter prediction module, used to input the pre - processed image into the trained distortion parameter prediction model to obtain distortion parameters; A geometric distortion calibration module, used to calibrate the initially acquired echo - planar imaging diffusion - weighted imaging through distortion parameters to obtain distortion - free echo - planar imaging diffusion - weighted imaging, including: Input the echo - planar imaging diffusion - weighted imaging and fast spin - echo diffusion - weighted imaging for pre - processing to obtain pre - processed images; Perform a splicing operation on the pre - processed image obtained by pre - processing the echo - planar imaging diffusion - weighted imaging and the pre - processed image obtained by pre - processing the fast spin - echo diffusion - weighted imaging; Successively pass through three sliding window transformation modules and pooling layers to obtain encoded features at three scales; Respectively pass the encoded features at the three scales through a decoding module composed of a residual convolution, a rectified linear unit activation layer, and an up - sampling layer to obtain decoded features; Splice the 3 decoded features and then pass through a convolution layer for decoding and up - sampling 4 times to output the distortion parameters.

Images