A Blind Super-Resolution Reconstruction Method and System for Cardiac Magnetic Resonance Images Based on a Continuous-Time Conditional Diffusion Model

Abstract

This invention discloses a blind super-resolution reconstruction method and system for cardiac magnetic resonance images based on a continuous-time conditional diffusion model, relating to the field of blind super-resolution reconstruction of cardiac magnetic resonance images. The method includes acquiring a cardiac magnetic resonance image dataset and performing data augmentation and degradation processing on the cardiac magnetic resonance images in the dataset to obtain low-resolution images; constructing a continuous-time conditional diffusion model based on a cascaded residual attention network feature extractor, a continuous-time conditional diffusion module, a hybrid parameterized score predictor, and an image quality loss module; training the continuous-time conditional diffusion model parameters by setting an optimizer, training parameters, and hardware configuration; acquiring the source image of the cardiac magnetic resonance image to be tested, inputting it into the continuous-time conditional diffusion model, and generating a super-resolution reconstructed image. By improving the time efficiency and generation quality of the diffusion model, efficient and high-precision medical image super-resolution reconstruction is achieved.

Description

Technical Field

[0001] This invention relates to the field of blind super-resolution reconstruction technology for cardiac magnetic resonance images, and in particular to a method and system for blind super-resolution reconstruction of cardiac magnetic resonance images based on a continuous-time conditional diffusion model. Background Technology

[0002] Cardiac magnetic resonance imaging (MRI) is an important non-invasive medical imaging technique with significant clinical value in the assessment of cardiovascular diseases. High-resolution cardiac MRI images are crucial for the early detection of heart disease and improving diagnostic accuracy and efficiency. However, traditional diffusion models rely on Markov chain iterative sampling, resulting in a time-consuming super-resolution reconstruction process. Existing methods suffer from problems such as blurred details and unclear edges when dealing with complex degradation. Traditional methods also fail to adequately model the mapping relationship between low-resolution and high-resolution images, leading to poor structural consistency between the generated results and the real images. Therefore, super-resolution techniques that utilize low-resolution images for high-resolution reconstruction have become a research hotspot in the field of cardiac imaging.

[0003] Image super-resolution reconstruction aims to recover high-resolution images from low-resolution images. This task has attracted widespread attention due to its ill-posedness and broad practical application prospects. In recent years, deep learning-based super-resolution reconstruction techniques have significantly improved the clarity and detail of cardiac magnetic resonance imaging (MRI) images, driving medical diagnosis and treatment towards greater precision and intelligence. However, existing deep learning-based super-resolution methods rely on idealized degradation models, which differ significantly from the complex degradation processes in the real world, making it difficult to handle the complex degradation in real-world scenarios. Generative model-based super-resolution methods effectively address the ill-posedness problem by learning the distribution from low-resolution to high-resolution images, but they still suffer from high time costs and insufficient high-frequency feature extraction capabilities.

[0004] Therefore, a blind super-resolution reconstruction method and system for cardiac magnetic resonance images based on a continuous-time conditional diffusion model is provided to solve the above problems. Summary of the Invention

[0005] To address the aforementioned challenges, this invention provides a blind super-resolution reconstruction method and system for cardiac magnetic resonance images based on a continuous-time conditional diffusion model. By improving the time efficiency and generation quality of the diffusion model, it achieves efficient and high-precision medical image super-resolution reconstruction.

[0006] To achieve the above objectives, this invention provides a blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model, specifically including the following steps:

[0007] S1: Acquire the cardiac magnetic resonance image dataset collected by a medical scanner, and perform data enhancement and degradation processing on the cardiac magnetic resonance source images in the cardiac magnetic resonance image dataset, and obtain low-resolution images through bicubic downsampling;

[0008] S2: A continuous-time conditional diffusion model is constructed based on a cascaded residual attention network feature extractor, a continuous-time conditional diffusion module, a hybrid parameterized score predictor, and an image quality loss module.

[0009] The low-resolution image obtained in S1 is processed by a cascaded residual attention network feature extractor. The high-frequency features obtained are used as the conditional input of the continuous-time conditional diffusion model.

[0010] The cardiac magnetic resonance image is noise-enhanced and transformed into a Gaussian distribution using the forward stochastic differential equation in the continuous-time conditional diffusion module.

[0011] Inverse denoising is achieved through the probability flow ordinary differential equation in the continuous-time conditional diffusion module, and the parameterization strategy is dynamically adjusted by the hybrid parameterized fractional predictor to generate a super-resolution reconstructed image.

[0012] The perceptual consistency between the generated super-resolution reconstructed image and the cardiac magnetic resonance image is optimized by combining fractional matching loss and perceptual loss through the image quality loss module.

[0013] S3: Set the optimizer, training parameters and hardware configuration to train the continuous-time conditional diffusion model parameters obtained in S2, and obtain the trained continuous-time conditional diffusion model.

[0014] S4: Acquire the cardiac magnetic resonance image to be tested, input it into the continuous-time conditional diffusion model trained in S3, and generate a super-resolution reconstructed image corresponding to the cardiac magnetic resonance image to be tested.

[0015] Preferably, the data augmentation in S1 includes rotating, horizontally or vertically flipping, and scaling operations on the cardiac magnetic resonance image source;

[0016] The degradation generation in S1 involves downsampling the cardiac magnetic resonance source image using an isotropic Gaussian blur kernel to generate a low-resolution image.

[0017] Preferably, the cascaded residual attention network feature extractor in S2 includes a shallow feature extraction module, a multi-layer cascaded residual attention block module, and a feature fusion and upsampling module;

[0018] The shallow feature extraction module uses a 3×3 convolutional layer to extract shallow features from low-resolution images;

[0019] The cascaded residual attention block contains 10 cascaded residual blocks. Each residual block consists of 5 channel attention modules and 5 non-local channel attention modules. The channel attention modules generate channel weights through global average pooling to enhance important channel features. The non-local channel attention modules capture long-distance feature dependencies by calculating cross-channel global correlations.

[0020] The feature fusion and upsampling module adds shallow features to deep high-frequency features and upsamples them to the target resolution through pixel recombination.

[0021] Preferably, the forward stochastic differential equation in S2 is expressed as:

[0022] ;

[0023] in, For low-resolution images The mean of the bicubic interpolation upsampling The empirical variance of the training set The noise variance is linearly increasing, with β(0) = 0.1 and β(T) = 20.

[0024] Preferably, the probability flow ordinary differential equation in S2 is expressed as:

[0025] ;

[0026] in, It is a hybrid parameterized score predictor.

[0027] Preferably, the parameterization strategy of the hybrid parameterized score predictor in S2 includes: Parameterization Parameterization and hybrid interpolation;

[0028] Parametric prediction of noise components For use in high-noise areas;

[0029] Parametric prediction of clean image components For use in low-noise areas;

[0030] Hybrid interpolation dynamically adjusts interpolation coefficients The balance of the two parameterizations is expressed as:

[0031] .

[0032] Preferably, the hybrid parameterized score predictor in S2 adopts a U-Net architecture, with the input being a noisy image. Time step coding and conditional characteristics Noise is removed layer by layer by using residual blocks and the Mish activation function.

[0033] Preferably, the image quality loss module in S2 includes a score matching loss, a perceptual loss, and a total loss function;

[0034] The score matching loss minimizes the mean square error between the predicted score and the target score;

[0035] Perceptual loss uses a pre-trained VGG-19 network to extract features and calculates the feature space distance between the generated image and the real high-resolution image.

[0036] The total loss function combines score matching loss and perceptual loss to optimize the perceptual consistency between generated images and real images.

[0037] Preferably, the optimizer in S3 uses an adaptive moment estimation algorithm, and the training parameters include batch size, training epochs, diffusion time step, and hardware configuration settings.

[0038] A blind super-resolution reconstruction system for cardiac magnetic resonance images based on a continuous-time conditional diffusion model includes a data preparation module, a continuous-time conditional diffusion model construction module, a training module, and a super-resolution reconstruction module.

[0039] The data preparation module acquires the cardiac magnetic resonance image dataset collected by the medical scanner, and performs data enhancement and degradation processing on the cardiac magnetic resonance source images in the cardiac magnetic resonance image dataset, and obtains low-resolution images through bicubic downsampling;

[0040] The continuous-time conditional diffusion model construction module includes a cascaded residual attention network feature extractor, a continuous-time conditional diffusion module, a hybrid parameterized score predictor, and an image quality loss module.

[0041] The low-resolution image obtained by the data preparation module is processed by a cascaded residual attention network feature extractor. The high-frequency features obtained are used as the conditional input of the continuous-time conditional diffusion model.

[0042] The cardiac magnetic resonance image is noise-enhanced and transformed into a Gaussian distribution using the forward stochastic differential equation in the continuous-time conditional diffusion module.

[0043] Inverse denoising is achieved through the probability flow ordinary differential equation in the continuous-time conditional diffusion module, and the parameterization strategy is dynamically adjusted by the hybrid parameterized fractional predictor to generate a super-resolution reconstructed image.

[0044] The perceptual consistency between the generated super-resolution reconstructed image and the cardiac magnetic resonance image is optimized by combining fractional matching loss and perceptual loss through the image quality loss module.

[0045] The training module sets the optimizer, training parameters, and hardware configuration to train the parameters of the continuous-time conditional diffusion model, resulting in the trained continuous-time conditional diffusion model.

[0046] The super-resolution reconstruction module acquires the cardiac magnetic resonance image to be tested and inputs it into a trained continuous-time conditional diffusion model to generate a super-resolution reconstructed image corresponding to the cardiac magnetic resonance image to be tested. Therefore, this invention adopts the above-mentioned blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model. The continuous-time conditional diffusion model constructed by the cascaded residual attention network feature extractor, the continuous-time conditional diffusion module, the hybrid parameterized score predictor, and the image quality loss module solves the problems of long time consumption and insufficient quality and consistency of generated images when using iterative sampling. This reduces the time consumption of the diffusion model in super-resolution reconstruction, improves the detail and structural consistency of cardiac magnetic resonance image super-resolution reconstruction, and enhances the robustness of the model to complex degradation.

[0047] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating a blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model, as described in this invention.

[0049] Figure 2 This is a schematic diagram of the continuous-time conditional diffusion model in this invention;

[0050] Figure 3 This is a schematic diagram of the cascaded residual attention network feature extractor in this invention;

[0051] Figure 4 This is a schematic diagram of the residual block in this invention;

[0052] Figure 5 The images show the visual effects of different super-resolution reconstruction methods in super-resolution reconstruction of cardiac horizontal plane magnetic resonance images in embodiments of the present invention.

[0053] Figure 6 These are visual representations of the effects of different super-resolution reconstruction methods on coronary magnetic resonance imaging of the heart in embodiments of the present invention.

[0054] Figure 7 These are visual representations of the effects of different super-resolution reconstruction methods on super-resolution reconstruction of cardiac sagittal magnetic resonance images in embodiments of the present invention.

[0055] Figure 8These are visual representations of different residual models in super-resolution reconstruction of cardiac magnetic resonance images in embodiments of the present invention.

[0056] Figure 9 This is a schematic diagram of the ablation study results of different cascaded channel attention blocks and cascaded residual attention blocks in the embodiments of the present invention;

[0057] Figure 10 These are visual representations of different parameterized models in embodiments of the present invention. Detailed Implementation

[0058] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0059] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0060] The terms "comprising" or "including" as used in this invention mean that the element preceding the term encompasses the element listed after the term, and do not exclude the possibility of encompassing other elements. Terms such as "inner," "outer," "upper," and "lower" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. When the absolute position of the described object changes, the relative positional relationship may also change accordingly. In this invention, unless otherwise explicitly specified and limited, the term "attached" and similar terms should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can refer to a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication of two elements or the interaction relationship between two elements. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0061] Example

[0062] This invention provides a blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model, such as... Figure 1 As shown, the specific steps include:

[0063] S1: Acquire the cardiac magnetic resonance image dataset collected by a medical scanner, and perform data enhancement and degradation processing on the cardiac magnetic resonance source images in the cardiac magnetic resonance image dataset, and obtain low-resolution images through bicubic downsampling;

[0064] The datasets used are the AMRGCardiac MRI Atlas cardiac MRI dataset acquired by the Siemens Avanto scanner from the Oakland MRI Research Group and the Cardiac MRI dataset provided by Dr. Paul Babyn, Chief of Radiology, and Dr. Shi Joon Yoo, Chief of Cardiac Radiology, at Toronto Children's Hospital.

[0065] The data augmentation in this application involves rotating the original image by ±90°, flipping it horizontally / vertically, and scaling it by 0.8-1.2 times to expand the training samples in the dataset.

[0066] The degradation process downsamples the high-resolution image using an isotropic Gaussian blur kernel to generate a low-resolution image that simulates the real scene. The kernel size is 21×21 and the width ranges from [0.2 to 4.0].

[0067] S2: As Figure 2 As shown, a continuous-time conditional diffusion model (CCDM) is constructed based on a cascaded residual attention network feature extractor, a continuous-time conditional diffusion module, a hybrid parameterized score predictor, and an image quality loss module.

[0068] The low-resolution image obtained in S1 is processed by a cascaded residual attention network feature extractor. The high-frequency features obtained are used as the conditional input of the continuous-time conditional diffusion model.

[0069] like Figure 3-4 As shown, the cascaded residual attention network feature extractor includes a shallow feature extraction module, a multi-layer cascaded residual attention block module, and a feature fusion and upsampling module; wherein, the shallow feature extraction module uses 3×3 convolutional layers to extract shallow features from low-resolution images. The cascaded residual attention blocks consist of 10 cascaded residual blocks, each composed of 5 channel attention (CA) modules and 5 non-local channel attention (NCA) modules. The channel attention (CA) modules generate channel weights through global average pooling to enhance important channel features, while the non-local channel attention (NCA) modules capture long-distance feature dependencies by calculating cross-channel global correlations. The feature fusion and upsampling module adds shallow features to deep high-frequency features and upsamples them to the target resolution using PixelShuffle. After a low-resolution image is input, shallow features are progressively processed through the residual attention blocks, weighted and fused with global and local information to output high-frequency features. As a conditional input to the diffusion model.

[0070] The cardiac MRI source images are noise-enhanced and transformed into a Gaussian distribution using a forward stochastic differential equation in the continuous-time conditional diffusion module. The mean and variance of the data are preserved during the forward process, reducing the model training complexity. The forward process is represented as follows:

[0071] ;

[0072] in, For low-resolution images The mean of the bicubic interpolation upsampling The empirical variance of the training set The noise variance is linearly increasing, with β(0) = 0.1 and β(T) = 20.

[0073] Inverse denoising is achieved through the probability flow ordinary differential equation in the continuous-time conditional diffusion module, combined with a hybrid parameterized fractional predictor to dynamically adjust the parameterization strategy, thereby generating a super-resolution reconstructed image. High-resolution images are generated efficiently through the ordinary differential equation solver, reducing the iteration time of traditional Markov chains.

[0074] The formula for reverse sampling is expressed as: ;

[0075] in, It is a hybrid parameterized score predictor.

[0076] Hybrid parameterized score predictor parameterization strategies include Parameterization Parameterization and hybrid interpolation; Parametric prediction of noise components For use in high-noise areas; Parametric prediction of clean image components Used in low-noise regions; hybrid interpolation dynamically adjusts the interpolation coefficients. The balance of the two parameterizations is expressed as:

[0077] .

[0078] The hybrid parameterized score predictor uses a U-Net architecture and takes a noisy image as input. Time step coding and conditional characteristics Noise is removed layer by layer by using residual blocks and the Mish activation function.

[0079] The perceptual consistency between the generated super-resolution reconstructed image and the cardiac magnetic resonance image is optimized by combining fractional matching loss and perceptual loss through the image quality loss module.

[0080] Based on the fractional matching loss of the continuous-time conditional diffusion model, a feature space distance loss based on the VGG network is added to directly optimize the perceptual consistency between the generated image and the real image.

[0081] The image quality loss module includes fractional matching loss, perceptual loss, and total loss function; fractional matching loss Minimizing the mean square error between the predicted score and the target score is expressed as:

[0082] ;

[0083] Perceived loss Feature extraction is performed using a pre-trained VGG-19 network to generate images. Compared to true high-resolution images The feature space distance is expressed as:

[0084] ;

[0085] The total loss function combines score matching loss and perceptual loss to optimize the perceptual consistency between the generated image and the real image, and is expressed as:

[0086] .

[0087] S3: Set the optimizer, training parameters and hardware configuration to train the continuous-time conditional diffusion model parameters obtained in S2, and obtain the trained continuous-time conditional diffusion model.

[0088] (1) Optimizer: Adam algorithm, settings , The initial learning rate is 1e-4.

[0089] (2) Training parameters:

[0090] 1) Batch size is 16, training rounds are 200.

[0091] 2) The diffusion time step T=100, and the noise variance increases linearly according to VP-SDE.

[0092] (3) Hardware configuration: 2×NVIDIA GeForce RTX 3090 Ti GPU.

[0093] S4: Acquire the cardiac magnetic resonance image to be tested, input it into the continuous-time conditional diffusion model trained in S3, and generate a super-resolution reconstructed image corresponding to the cardiac magnetic resonance image to be tested. The specific steps are as follows:

[0094] (1) Input low-resolution image: initially upsampled to the target size by bicubic interpolation.

[0095] (2) Input the continuous-time conditional diffusion model.

[0096] 1) Feature extraction: Input to cascaded residual attention network, output conditional features .

[0097] 2) Probability stream sampling:

[0098] Initial noise The equations are inversely solved using a Runge-Kutta ordinary differential equation solver with a tolerance of 1e-4 to generate high-resolution images.

[0099] (3) Output results: The visual effect is optimized by post-processing, such as histogram equalization post-processing method.

[0100] Example 1

[0101] A blind super-resolution reconstruction system for cardiac magnetic resonance images based on a continuous-time conditional diffusion model includes a data preparation module, a continuous-time conditional diffusion model construction module, a training module, and a super-resolution reconstruction module.

[0102] The data preparation module acquires the cardiac magnetic resonance image dataset collected by the medical scanner, and performs data enhancement and degradation processing on the cardiac magnetic resonance source images in the cardiac magnetic resonance image dataset, and obtains low-resolution images through bicubic downsampling;

[0103] The continuous-time conditional diffusion model construction module includes a cascaded residual attention network feature extractor, a continuous-time conditional diffusion module, a hybrid parameterized score predictor, and an image quality loss module.

[0104] The low-resolution image obtained by the data preparation module is processed by a cascaded residual attention network feature extractor. The high-frequency features obtained are used as the conditional input of the continuous-time conditional diffusion model.

[0105] The cardiac magnetic resonance image is noise-enhanced and transformed into a Gaussian distribution using the forward stochastic differential equation in the continuous-time conditional diffusion module.

[0106] Inverse denoising is achieved through the probability flow ordinary differential equation in the continuous-time conditional diffusion module, and the parameterization strategy is dynamically adjusted by the hybrid parameterized fractional predictor to generate a super-resolution reconstructed image.

[0107] The perceptual consistency between the generated super-resolution reconstructed image and the cardiac magnetic resonance image is optimized by combining fractional matching loss and perceptual loss through the image quality loss module.

[0108] The training module sets the optimizer, training parameters, and hardware configuration to train the parameters of the continuous-time conditional diffusion model, resulting in the trained continuous-time conditional diffusion model.

[0109] The super-resolution reconstruction module acquires the cardiac magnetic resonance image to be tested, inputs it into the trained continuous-time conditional diffusion model, and generates a super-resolution reconstructed image corresponding to the cardiac magnetic resonance image to be tested.

[0110] Example 2

[0111] The effectiveness of the blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model provided in this application is verified as follows:

[0112] 1. Improved time efficiency:

[0113] This application uses probabilistic flow ordinary differential equation sampling, which reduces reconstruction time by more than 15% compared to traditional diffusion models such as DBSR. Experimental data show that CCDM takes 1.62 seconds and DBSR takes 1.90 seconds.

[0114] Model runtime is crucial for medical image super-resolution reconstruction. Therefore, this embodiment compares CCDM with IKC, SRDiff, and DBSR in terms of parameter count and runtime. As shown in Table 1, CCDM achieves the optimal PSNR value while exhibiting superior reconstruction time compared to DBSR. This indicates that CCDM significantly reduces computational complexity while maintaining high-quality cardiac MRI image reconstruction performance, thus achieving a better balance between super-resolution reconstruction performance and model computational complexity.

[0115] Table 1

[0116]

[0117] 2. Image quality optimization:

[0118] (1) Improved PSNR value: In the horizontal plane dataset with a blur kernel of 1.2, the PSNR value was improved to 32.57 dB, which is 0.03 dB higher than the existing best method DBSR 32.54 dB.

[0119] To further evaluate the super-resolution reconstruction performance of CCDM, it was compared with mainstream methods such as SRCNN, VDSR, DRRN, LapSRN, and A 2 A comparison was made between N, RCAN, SRMDNF, IKC, DASR, SRDiff, StableSR, and DBSR, and the results are shown in Table 2 below.

[0120] As shown in Table 2 below, with a Gaussian isotropic blur kernel of 1.2, CCDM achieves a PSNR of 32.57 dB for the horizontal plane dataset when the magnification factor is 4. Compared with other methods, CCDM significantly improves performance: compared to SRCNN, VDSR, DRRN, LapSRN, and A... 2N, RCAN, SRMDNF, IKC, DASR, SRDiff, StableSR, and DBSR showed higher values ​​of 4.12 dB, 3.69 dB, 3.46 dB, 3.32 dB, 3.20 dB, 1.86 dB, 1.83 dB, 0.38 dB, 0.19 dB, 0.08 dB, 0.31 dB, and 0.03 dB, respectively. This indicates that CCDM exhibits stronger robustness and generative capabilities in super-resolution reconstruction tasks under complex fuzzy kernel degradation conditions.

[0121] Furthermore, it can be observed that SRCNN, VDSR, DRRN, LapSRN, and A 2 N and RCAN exhibit lower PSNR values ​​on cardiac MRI horizontal, coronal, and sagittal datasets because these methods assume low-resolution images are based on bicubic downsampling degradation. In contrast, SRMDNF, IKC, DASR, and DBSR are designed for complex real-world degradation problems and better handle low-resolution image degradation caused by blur kernels, achieving higher PSNR values ​​on horizontal, coronal, and sagittal datasets.

[0122] Table 2

[0123]

[0124] (2) The perception indicators are significantly improved. The method provided in this application has an LPIPS of 0.0461 and an FID of 63.85.

[0125] To comprehensively evaluate the reconstruction performance of CCDM on low-resolution cardiac magnetic resonance images, PSNR, SSIM, LPIPS, and FID image evaluation metrics were used, and the results were compared with other blind super-resolution reconstruction models and diffusion models. As shown in Table 3, CCDM achieved the lowest LPIPS and FID values ​​while obtaining the highest PSNR and SSIM values, fully demonstrating its superior performance in low-resolution cardiac magnetic resonance image reconstruction. This indicates that CCDM can not only generate images with high perceptual quality but also surpass existing methods in structural similarity and pixel accuracy, demonstrating strong practical application value.

[0126] Table 3

[0127]

[0128] 3. Enhanced robustness:

[0129] This application can handle complex degradation under a Gaussian isotropic blur kernel with a width range of [0.2, 4.0], adapting to real medical imaging scenarios.

[0130] Compared to these state-of-the-art methods, CCDM achieves the highest PSNR values ​​on cardiac MRI horizontal, coronal, and sagittal datasets, demonstrating a significant performance improvement. Due to CCDM's conditional diffusion model network framework, its powerful generative capabilities not only predict blur kernels more accurately but also generate higher-quality super-resolution images, fully demonstrating its superiority under complex degradation conditions.

[0131] Figure 5-7 This paper presents a comparison of the visual effects of different super-resolution reconstruction methods in cardiac magnetic resonance imaging (MRI) images. It can be seen that the SRCNN and RCAN reconstructed cardiac MRI super-resolution images are relatively blurry, failing to clearly present important tissue details. Although SRMDNF, IKC, and DASR show better performance in the reconstruction process, the boundary details of their reconstructed images still appear insufficiently clear. In contrast, the images reconstructed by StableSR, SRDiff, DBSR, and CCDM are clearer. However, CCDM exhibits a smoother visual effect in the reconstruction, with less edge blurring and preserving more image details, such as pathological details like myocardial fibrosis and edema, fully demonstrating its superiority in cardiac MRI image super-resolution reconstruction.

[0132] Example 3

[0133] To further validate the effectiveness of CCDM, this embodiment conducts an ablation study on the cascaded residual attention network feature extractor, the hybrid parameter fractional predictor, and the image quality loss module. These experiments were conducted on a cardiac magnetic resonance imaging (MRI) horizontal dataset with a magnification factor of 4, evaluating the contribution of each module to super-resolution reconstruction performance. By analyzing the model performance under different configurations, the roles of each module in improving image quality, enhancing detail recovery, and accelerating the generation process can be more clearly understood, thus verifying the impact of these designs on the final model performance.

[0134] CCDM-Res refers to cascaded residual attention blocks without channel attention mechanisms and non-local channel attention mechanisms; CCDM CA refers to cascaded residual attention blocks with channel attention mechanisms; CCDM-NC refers to cascaded residual attention blocks with non-local channel attention mechanisms; and CCDM refers to cascaded residual attention blocks with both channel attention mechanisms and non-local channel attention mechanisms. Figure 8The image shows the visual effects of CCDM-Res, CCDM-CA, CCDM-NC, and CCDM in super-resolution reconstruction of cardiac magnetic resonance images. It can be seen that CCDM, which combines channel attention mechanism and non-local channel attention mechanism in cascaded residual attention block, can reconstruct clearer and more detailed cardiac magnetic resonance super-resolution images than CCDM-Res, CCDM-CA, and CCDM-NC.

[0135] Since the number of cascaded channel attention blocks and cascaded residual attention blocks has a significant impact on the reconstruction performance of CCDM, an ablation study was conducted on the number of these blocks in the feature extractor of the cascaded residual attention network. The results are as follows: Figure 9 As shown, from Figure 9 As can be seen, the PSNR value of CCDM increases with the increase in the number of cascaded channel attention blocks and cascaded residual attention blocks. However, when the number of these blocks reaches a certain threshold, the PSNR value of CCDM no longer increases, and even slightly decreases. Therefore, this application ultimately sets the number of cascaded channel attention blocks and cascaded residual attention blocks to 5 and 10, respectively, to obtain the best super-resolution reconstruction performance.

[0136] To investigate the different parameter choices in denoising networks and the impact of image quality loss on cardiac magnetic resonance image reconstruction, this embodiment conducts an ablation study on parameter selection. Regarding image quality loss, such as... Figure 10 As shown, the model trained with Lquality achieves better reconstruction performance than the model without Lquality, which fully demonstrates the importance of image quality loss for diffusion-based super-resolution reconstruction.

[0137] Therefore, this invention employs the aforementioned blind super-resolution reconstruction method and system for cardiac magnetic resonance images based on a continuous-time conditional diffusion model. High-frequency features are extracted from the low-resolution image using a cascaded residual attention network feature extractor. This serves as a conditional input to the diffusion model. The hybrid parameterized score predictor is combined with... The system incorporates time-step encoding and noisy images, dynamically adjusting parameterization strategies. Probabilistic flow ordinary differential equations utilize conditional scoring functions to rapidly generate super-resolution reconstructed images, while image quality loss further constrains the consistency between the generated results and the source images. By improving the time efficiency and generation quality of the diffusion model, efficient and high-precision medical image super-resolution reconstruction is achieved.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model, characterized in that, Specifically, the following steps are included: S1: Acquire the cardiac magnetic resonance image dataset collected by a medical scanner, and perform data enhancement and degradation processing on the cardiac magnetic resonance source images in the cardiac magnetic resonance image dataset, and obtain low-resolution images through bicubic downsampling; S2: A continuous-time conditional diffusion model is constructed based on a cascaded residual attention network feature extractor, a continuous-time conditional diffusion module, a hybrid parameterized score predictor, and an image quality loss module. The low-resolution image obtained in S1 is processed by a cascaded residual attention network feature extractor. The high-frequency features obtained are used as the conditional input of the continuous-time conditional diffusion model. The cardiac MRI source image is noise-enhanced and transformed into a Gaussian distribution using the forward stochastic differential equation in the continuous-time conditional diffusion module; the forward stochastic differential equation in S2 is expressed as: ； in, For low-resolution images The mean of the bicubic interpolation upsampling The empirical variance of the training set Linearly increasing noise variance, β(0)=0.1, β(T)=20; Inverse denoising is performed using the probability flow ordinary differential equation in the continuous-time conditional diffusion module, combined with dynamic adjustment of the parameterization strategy by a hybrid parameterized fractional predictor to generate a super-resolution reconstructed image; the probability flow ordinary differential equation in S2 is expressed as: ； in, For hybrid parameterized score predictors; The parameterization strategy of the hybrid parameterized score predictor in S2 includes: Parameterization Parameterization and hybrid interpolation; Parametric prediction of noise components For use in high-noise areas; Parametric prediction of clean image components For use in low-noise areas; Hybrid interpolation dynamically adjusts interpolation coefficients The balance of the two parameterizations is expressed as: ； The perceptual consistency between the generated super-resolution reconstructed image and the cardiac magnetic resonance image is optimized by combining fractional matching loss and perceptual loss through the image quality loss module. S3: Set the optimizer, training parameters and hardware configuration to train the continuous-time conditional diffusion model parameters obtained in S2, and obtain the trained continuous-time conditional diffusion model. S4: Acquire the cardiac magnetic resonance image to be tested, input it into the continuous-time conditional diffusion model trained in S3, and generate a super-resolution reconstructed image corresponding to the cardiac magnetic resonance image to be tested.

2. The blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model as described in claim 1, characterized in that: Data augmentation in S1 includes rotation, horizontal or vertical flipping, and scaling of cardiac magnetic resonance image sources; The degradation generation in S1 involves downsampling the cardiac magnetic resonance source image using an isotropic Gaussian blur kernel to generate a low-resolution image.

3. The blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model as described in claim 1, characterized in that: The cascaded residual attention network feature extractor in S2 includes a shallow feature extraction module, a multi-layer cascaded residual attention block module, and a feature fusion and upsampling module; The shallow feature extraction module uses a 3×3 convolutional layer to extract shallow features from low-resolution images; The cascaded residual attention block contains 10 cascaded residual blocks. Each residual block consists of 5 channel attention modules and 5 non-local channel attention modules. The channel attention modules generate channel weights through global average pooling to enhance important channel features. The non-local channel attention modules capture long-distance feature dependencies by calculating cross-channel global correlations. The feature fusion and upsampling module adds shallow features to deep high-frequency features and upsamples them to the target resolution through pixel recombination.

4. The blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model as described in claim 1, characterized in that: The hybrid parameterized score predictor in S2 uses a U-Net architecture, with a noisy image as input. Time step coding and conditional characteristics Noise is removed layer by layer by using residual blocks and the Mish activation function.

5. The blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model as described in claim 1, characterized in that: The image quality loss module in S2 includes fractional matching loss, perceptual loss, and total loss function; The score matching loss minimizes the mean square error between the predicted score and the target score; Perceptual loss uses a pre-trained VGG-19 network to extract features and calculates the feature space distance between the generated image and the real high-resolution image. The total loss function combines score matching loss and perceptual loss to optimize the perceptual consistency between generated images and real images.

6. The blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model as described in claim 1, characterized in that: The optimizer in S3 uses an adaptive moment estimation algorithm, and the training parameters include batch size, training epochs, diffusion time step, and hardware configuration settings.

7. A blind super-resolution reconstruction system for cardiac magnetic resonance images based on a continuous-time conditional diffusion model, the system being used to implement the blind super-resolution reconstruction method for cardiac magnetic resonance images based on a continuous-time conditional diffusion model as described in any one of claims 1-6, characterized in that: The blind super-resolution reconstruction system for cardiac magnetic resonance images based on the continuous-time conditional diffusion model includes a data preparation module, a continuous-time conditional diffusion model construction module, a training module, and a super-resolution reconstruction module. The data preparation module acquires the cardiac magnetic resonance image dataset collected by the medical scanner, and performs data enhancement and degradation processing on the cardiac magnetic resonance source images in the cardiac magnetic resonance image dataset, and obtains low-resolution images through bicubic downsampling; The continuous-time conditional diffusion model construction module includes a cascaded residual attention network feature extractor, a continuous-time conditional diffusion module, a hybrid parameterized score predictor, and an image quality loss module. The low-resolution image obtained by the data preparation module is processed by a cascaded residual attention network feature extractor. The high-frequency features obtained are used as the conditional input of the continuous-time conditional diffusion model. The cardiac magnetic resonance image is noise-enhanced and transformed into a Gaussian distribution using the forward stochastic differential equation in the continuous-time conditional diffusion module. Inverse denoising is achieved through the probability flow ordinary differential equation in the continuous-time conditional diffusion module, and the parameterization strategy is dynamically adjusted by the hybrid parameterized fractional predictor to generate a super-resolution reconstructed image. The perceptual consistency between the generated super-resolution reconstructed image and the cardiac magnetic resonance image is optimized by combining fractional matching loss and perceptual loss through the image quality loss module. The training module sets the optimizer, training parameters, and hardware configuration to train the parameters of the continuous-time conditional diffusion model, resulting in the trained continuous-time conditional diffusion model. The super-resolution reconstruction module acquires the cardiac magnetic resonance image to be tested, inputs it into the trained continuous-time conditional diffusion model, and generates a super-resolution reconstructed image corresponding to the cardiac magnetic resonance image to be tested.

Images