A dynamic medical imaging method, system, apparatus, and storage medium

Abstract

The embodiment of the specification provides a dynamic medical imaging method, system, device and storage medium, the method comprising: acquiring a plurality of first images corresponding to a plurality of phases of a target object, each first image corresponding to one phase of the plurality of phases; determining a reference image based on the first images corresponding to at least two phases of the plurality of phases; inputting the plurality of first images and the reference image into a trained image processing model to obtain a plurality of second images, each second image corresponding to one of the plurality of first images, and the image quality of the second image being higher than that of the corresponding first image.

Description

Technical Field

[0001] This specification relates to the field of medical imaging, and in particular to a dynamic medical imaging method, system, device, and storage medium. Background Technology

[0002] Dynamic medical imaging allows for the observation and assessment of dynamic changes in human organs and tissues over a given time period. Taking cardiac magnetic resonance imaging (MRI) cine imaging as an example, cardiac MRI cine imaging provides comprehensive information about the heart and plays a crucial role in cardiac diagnosis and management. However, to capture rapid real-time dynamic information, each phase of a cardiac MRI cine cannot fully acquire the signals required for image reconstruction, leading to a degraded quality of the reconstructed image. Current deep learning-based cardiac cine reconstruction techniques primarily utilize convolutional networks to design different feature extractors. However, convolutional neural networks can cause temporal and spatial blurring, making it difficult to recover high-frequency image details, thus resulting in a decrease in reconstruction quality.

[0003] Therefore, it is necessary to provide a dynamic medical imaging method that reconstructs high-quality images from the original acquired images, thereby providing clearer and more accurate reconstructed images and better utilizing the reconstructed images for diagnosis. Summary of the Invention

[0004] This specification provides one or more embodiments of a dynamic medical imaging method, the method comprising: acquiring a plurality of first images corresponding to a plurality of phases of a target object, each first image corresponding to one phase of the plurality of phases; determining a reference image based on at least two first images corresponding to the plurality of phases; inputting the plurality of first images and the reference image into a trained image processing model to obtain a plurality of second images, each second image corresponding to one of the plurality of first images, wherein the image quality of the second image is higher than that of the corresponding first image.

[0005] This specification provides one or more embodiments of a dynamic medical imaging system, the system including an acquisition module, a condition generation module, and an image processing module; the acquisition module is configured to acquire multiple first images corresponding to multiple phases of a target object, each first image corresponding to one phase of the multiple phases; the condition generation module is configured to determine a reference image based on at least two first images corresponding to the multiple phases; the image processing module is configured to input the multiple first images and the reference image into a trained image processing model to obtain multiple second images, each second image corresponding to one of the multiple first images, the image quality of the second image being higher than that of the corresponding first image.

[0006] This specification provides one or more embodiments of a dynamic medical imaging device, the device including at least one processor and at least one memory; the at least one memory is used to store computer instructions; the at least one processor is used to execute at least a portion of the computer instructions to implement a dynamic medical imaging method.

[0007] This specification provides one or more embodiments of a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions from the storage medium, the computer executes a dynamic medical imaging method. Attached Figure Description

[0008] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0009] Figure 1 These are schematic diagrams illustrating application scenarios of the dynamic medical imaging system according to some embodiments of this specification;

[0010] Figure 2 These are exemplary block diagrams of a dynamic medical imaging system according to some embodiments of this specification;

[0011] Figure 3 This is an exemplary flowchart of a dynamic medical imaging method according to some embodiments of this specification;

[0012] Figure 4 This is an exemplary schematic diagram illustrating the generation of a second image according to some embodiments of this specification;

[0013] Figure 5 These are exemplary schematic diagrams illustrating data consistency processing according to some embodiments of this specification;

[0014] Figure 6 This is an exemplary schematic diagram of the training process of an image processing model according to some embodiments of this specification;

[0015] Figure 7 This is an exemplary schematic diagram of the training process of an image processing model according to other embodiments of this specification;

[0016] Figure 8 This is an exemplary schematic diagram illustrating the generation of a second image according to other embodiments of this specification. Detailed Implementation

[0017] The accompanying drawings used in the description of the embodiments will be briefly introduced below. The drawings do not represent all embodiments.

[0018] The terms “system,” “device,” “unit,” and / or “module” as used herein are one method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0019] Unless the context clearly indicates an exception, words such as "a," "an," "a kind," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0020] Figure 1 These are schematic diagrams illustrating application scenarios of a dynamic medical imaging system according to some embodiments of this specification. For example... Figure 1 As shown, the application scenario 100 of the dynamic medical imaging system may include an imaging device 110, a first image 120, an image processing model 130, and a second image 140. Dynamic medical imaging may include cardiac magnetic resonance imaging (e.g., cardiac magnetic resonance cine imaging), or dynamic MRI imaging of other sites, and / or dynamic imaging of other modalities.

[0021] Imaging device 110 is configured to acquire a first image. In some embodiments, the imaging device may include at least one of a single-modal scanner and / or a multimodal scanner. A single-modal scanner may include a computed tomography (CT) scanner, a magnetic resonance imaging (MRI) scanner, an ultrasound scanner, or any combination thereof. A multimodal scanner may include an X-ray imaging-magnetic resonance imaging (X-MRI) scanner, a positron emission tomography-X-ray imaging (PET-X-ray) scanner, or any combination thereof.

[0022] The first image 120 refers to the undersampled image acquired by the imaging device. During the scanning process of the target object, the imaging device undersamples the K-space to obtain undersampled K-space data, and then reconstructs the undersampled K-space data (e.g., through inverse Fourier transform) to obtain the first image. The first image can be a two-dimensional image or a three-dimensional image.

[0023] K-space refers to the space used to store the frequency domain data of an image. In some embodiments, K-space includes a two-dimensional or three-dimensional matrix for storing image data.

[0024] The Fourier transform involves converting spatial domain (also known as image domain) data or signals into frequency domain data or signals. The inverse Fourier transform is the reverse process of the Fourier transform, which involves converting frequency domain data or signals into spatial domain data or signals.

[0025] Image processing model 130 refers to a model used to generate a second image. In some embodiments, the image processing model is a machine learning model. In some embodiments, the image processing model is a deep learning model based on a conditional diffusion model. The image processing model can communicate with an imaging device to process data and / or information acquired from the imaging device. For example, the image processing model can acquire a first image from the imaging device.

[0026] The second image 140 refers to a high-quality image generated after processing, reconstructing, or optimizing the first image. The image quality of the second image is higher than that of the corresponding first image. Image quality can be characterized by parameters such as resolution, signal-to-noise ratio, or contrast.

[0027] In a typical application scenario, the imaging device 110 acquires a first image 120 and sends the first image 120 to the image processing model 130. The image processing model 130 generates a second image 140 based on the first image 120 and the reference image.

[0028] Figure 2 This is an exemplary block diagram of a dynamic medical imaging system according to some embodiments of this specification.

[0029] In some embodiments, such as Figure 2 As shown, the dynamic medical imaging system 200 may include an acquisition module 210, a condition generation module 220, and an image processing module 230.

[0030] The acquisition module 210 is used to acquire a first image. In some embodiments, the acquisition module 210 is configured to acquire multiple first images corresponding to multiple periods of a target object. The acquisition module includes an imaging device, etc.

[0031] The condition generation module 220 is a module for determining a reference image. In some embodiments, the condition generation module 220 is configured to determine the reference image based on a first image corresponding to at least two of a plurality of phases.

[0032] Image processing module 230 is a module for acquiring second images. In some embodiments, image processing module 230 includes an image processing model. Image processing module 230 is configured to input multiple first images and reference images into a trained image processing model to obtain multiple second images.

[0033] In some embodiments, the dynamic medical imaging system 200 may also include a processor and a storage device.

[0034] The processor is configured to process data and / or information related to the dynamic medical imaging system 200. The storage device is configured to store data and / or instructions related to the dynamic medical imaging system 200. In some embodiments, the processor includes a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), or any combination thereof.

[0035] In some embodiments, the acquisition module 210, the condition generation module 220, the image processing module 230, and the storage device may be integrated into the processor.

[0036] For more information on the Dynamic Medical Imaging System 200 and its modules, please refer to [link / reference]. Figures 3-5 And its related descriptions.

[0037] It should be noted that the above description of the dynamic medical imaging system and its modules is for convenience only and should not be construed as limiting this specification to the scope of the embodiments described. It is understood that those skilled in the art, after understanding the principles of the system, may arbitrarily combine the various modules or construct subsystems connected to other modules without departing from these principles. In some embodiments, Figure 2 The acquisition module 210, condition generation module 220, and image processing module 230 disclosed herein can be different modules within a single system, or a single module can implement the functions of two or more of the aforementioned modules. For example, the modules can share a single storage module, or each module can have its own separate storage module. Such variations are all within the scope of protection of this specification.

[0038] Figure 3 This is an exemplary flowchart of a dynamic medical imaging method according to some embodiments of this specification. In some embodiments, process 300 may be executed by a dynamic medical imaging system. Figure 3 As shown, process 300 includes the following steps.

[0039] Step 310: Obtain multiple first images corresponding to multiple periods of the target object.

[0040] For an explanation of the first image, please refer to [link / reference]. Figure 1 And its related descriptions.

[0041] The target object refers to the object that needs to be medically imaged. For example, the target object includes the patient's body parts and organs (such as the heart, lungs, blood vessels, etc.).

[0042] A phase refers to a period of time during the physiological processes of a target object (e.g., heartbeat, respiration, blood flow). Taking the heart as an example, the phase includes different stages in the heartbeat process, such as systole and diastole. Systole is the period when the heart contracts and pumps blood. Diastole is the period when the heart relaxes and fills with blood, preparing for the next contraction.

[0043] In some embodiments, one first image corresponds to one phase among multiple phases. Multiple first images can dynamically display the movement of the heart during continuous beating.

[0044] Step 320: Determine a reference image based on the first image corresponding to at least two of the multiple phases.

[0045] The anatomical structure of the target object can be identified in the reference image.

[0046] In some embodiments, the dynamic medical imaging system uses a reference image as a condition for generating the image processing model. See [link to documentation] for an explanation of the generation conditions. Figure 4 And its related descriptions.

[0047] In some embodiments, the dynamic medical imaging system can combine pixel values ​​or voxel values ​​of at least two corresponding first images to obtain a reference image. The dynamic medical imaging system can also perform Fourier transform on the at least two corresponding first images to obtain at least two K-space datasets, combine the at least two K-space datasets to obtain reference K-space data, and perform inverse Fourier transform on the reference K-space data to obtain the reference image.

[0048] In some embodiments, the dynamic medical imaging system can integrate pixel values ​​or voxel values ​​of at least two corresponding first images to obtain a reference image. Integration includes summation, averaging, or weighted averaging. The weights of different first images in the weighted average can be preset based on experience. The pixel values ​​or voxel values ​​of the first images represent the grayscale values ​​at corresponding locations of pixels or voxels in the first image.

[0049] For example, the pixel values ​​of pixels at the same position in the first image I1 and the first image I2, corresponding to phases P1 and P2 respectively, can be integrated to obtain a reference image. For example, the pixel values ​​of the first row and first column of pixels in the first image I1 and the first row and first column of pixels in the first image I2 can be summed to obtain the pixel values ​​of the first row and first column of pixels in the reference image. Similarly, the pixel values ​​of the first row and second column of pixels in the first image I1 and the first row and second column of pixels in the first image I2 can be summed to obtain the pixel values ​​of the first row and second column of pixels in the reference image, and so on, to obtain the reference image.

[0050] A K-space dataset refers to a collection of K-space data corresponding to a first image. In some embodiments, a dynamic medical imaging system performs Fourier transforms on at least two corresponding first images to obtain at least two K-space datasets. One K-space dataset corresponds to one first image.

[0051] Reference K-space data refers to the K-space data corresponding to the reference image.

[0052] In some embodiments, the dynamic medical imaging system can combine at least two K-space datasets in various ways to obtain reference K-space data. For example, the dynamic medical imaging system can integrate K-space data at corresponding locations in each K-space dataset to obtain reference K-space data. Integration includes summation, averaging, or weighted averaging.

[0053] In some embodiments, the dynamic medical imaging system performs an inverse Fourier transform on reference K-space data to obtain a reference image.

[0054] A reference image is generated by integrating multiple phase images. This reference image serves as the generation condition for a deep learning model (image processing model) based on a conditional diffusion model, resulting in better denoising and enhanced image contrast, allowing doctors to more clearly observe minute anatomical structures. The generation condition acts as a constraint and guide for the model's output in the conditional diffusion model-based deep learning model. By inputting the reference image into the image processing model, it can synthesize information from different phases to generate a second image of higher quality. Compared to using a single phase image as a reference image, multi-phase fusion reduces noise, enhances contrast, and improves detail display, thus providing the image processing model with clearer and more accurate anatomical structures as guidance for its output, thereby improving the denoising effect and the overall image quality of the output image.

[0055] Step 330: Input multiple first images and reference images into the trained image processing model to obtain multiple second images.

[0056] For more information about the second image, please refer to [link / reference]. Figure 1 The corresponding description. For more information on image processing models, please refer to... Figure 1 The corresponding description.

[0057] In some embodiments, the image processing model can be a deep learning model based on the Conditional Diffusion Model (CDM).

[0058] In some embodiments, any one or combination of Convolutional Neural Networks (CNN) models, Convolutional Recurrent Neural Networks (CRNN) models, Transformers models, or other custom model structures can be applied to deep learning models based on conditional diffusion models to obtain image processing models.

[0059] CNN models can effectively extract spatial hierarchical features from images through convolutional layers, with low computational complexity, and can enhance image processing capabilities by combining attention mechanisms. CRNN models combine the spatial feature extraction capabilities of CNN models with the time-series processing capabilities of Recurrent Neural Networks (RNN) models, enabling them to better capture dependencies between different phases. Transformer models, utilizing self-attention mechanisms, excel at handling long-range dependencies and are suitable for processing large datasets and complex models.

[0060] In some embodiments, the input to the image processing model includes a first image and a reference image, and the output includes a second image. In some embodiments, the input to the image processing model may also include physiological signals (e.g., respiratory signals, electrocardiograms, etc.). When the imaging device acquires data in parallel, the input to the image processing model may also include a coil sensitivity map. The coil sensitivity map is used to reflect the sensitivity of the coils in the imaging device at different positions and orientations to received signals or generated magnetic fields.

[0061] In some embodiments, the input and output of the image processing model can be implemented in various ways. For example, single-channel input and output, where one first image and one reference image are input at a time, and a second image corresponding to the first image is output. Another example is multi-channel input and output, where multiple first images and reference images are input simultaneously, each channel processes one first image and one reference image, and a second image corresponding to the first image is output, with multiple channels processing in parallel.

[0062] This application utilizes a deep learning model based on the conditional diffusion model to process undersampled images and generate higher-quality second images. By incorporating additional generation conditions (reference images) and iteratively removing noise from the undersampled images, the deep learning model based on the conditional diffusion model outputs clearer and more detailed second images, achieving higher denoising efficiency and better results.

[0063] It should be noted that the above description of process 300 is for illustrative purposes only and does not limit the scope of this specification. Those skilled in the art can make various modifications and changes to process 300 under the guidance of this specification. However, these modifications and changes remain within the scope of this specification.

[0064] Figure 4 This is an exemplary schematic diagram illustrating the generation of a second image according to some embodiments of this specification.

[0065] In some embodiments, for each first image 410, the dynamic medical imaging system can utilize a trained image processing model to perform multiple rounds of image processing based on the first image 410 and the reference image 420. Each round of image processing includes: denoising the input image of the current round of image processing based on the reference image 420 to obtain an output image (such as output image 430-1, output image 430-2, ..., output image 430-n, where n is the number of rounds of image processing), the input image being the first image 410 or the output image of the previous round of image processing, and generating a second image 440 corresponding to the first image 410 based on the output image of the last round of image processing (such as output image 430-n).

[0066] In some embodiments, the dynamic medical imaging system uses a reference image as a generation condition for the image processing model. Generation conditions refer to the conditions used to guide the image processing model in generating a second image.

[0067] In some embodiments, such as Figure 4 As shown, the input for the first round of image processing is the first image 410, the input for the second round of image processing is the output image of the previous round, and the output image of the last round of image processing is used as the output of the image processing model, namely the second image 440. Image processing includes denoising the input image. Denoising methods include any feasible methods such as mean filtering and bilateral filtering.

[0068] In some embodiments, the dynamic medical imaging system can utilize a trained image processing model to perform multiple rounds of image processing based on a first image and a reference image to obtain an output image. For example, the dynamic medical imaging system uses the output image of the last round of image processing as the second image.

[0069] In some embodiments, for each round of image processing, the dynamic medical imaging system can denoise the input image of the current round of image processing based on a reference image to obtain a denoised image. Based on the first image, data consistency processing is performed on the denoised image to obtain the output image of the current round of image processing. The dynamic medical imaging system performs data consistency processing on the denoised image obtained from the last round of image processing to obtain a second image.

[0070] Data consistency processing refers to the process of maintaining consistency between different image data during image reconstruction. In some embodiments, data consistency processing includes any feasible method such as calculating data expressions or adding constraints.

[0071] In some embodiments, the process of the dynamic medical imaging system performing data consistency processing on a denoised image based on a first image includes: performing a Fourier transform on the first image to obtain first K-space data, performing a Fourier transform on the denoised image to obtain second K-space data, obtaining third K-space data based on the first K-space data and the second K-space data, and performing an inverse Fourier transform on the third K-space data to obtain an output image.

[0072] The first K-space data refers to the K-space data obtained by performing a Fourier transform on the first image.

[0073] The second K-space data refers to the K-space data obtained by performing a Fourier transform on the denoised image.

[0074] The third K-space data refers to the K-space data corresponding to the output image. In some embodiments, the dynamic medical imaging system can merge the first K-space data and the second K-space data to obtain the third K-space data. For example, the dynamic medical imaging system can merge the first K-space data and the second K-space data using a preset formula to obtain the third K-space data.

[0075] For example, the preset formula is represented by the following equation (1): K r =k t′ (1-M)+k i′ M (1)

[0076] Among them, K r Represents the third K-space data, k t′ This represents the second K-space data, k i′ This represents the first K-space data. M represents k. i′ The corresponding matrix M contains the positions of its elements and the first K-space data k. i′ Each data point in the M-k matrix corresponds to a specific position. i′ The position corresponding to the location containing data is set to 1, and the position in M ​​is related to k. i′ The position corresponding to a position with no data is set to 0. 1-M represents k. i′ The corresponding other matrix, the positions of each element in 1-M and the first K-space data k i′ Each data point in the sequence corresponds to a specific position, where 1-M corresponds to k. i′ The position corresponding to the location containing data is set to 0, and the position in 1-M is related to k. i′The position corresponding to a position with no data is set to 1. k t′ (1-M) indicates that k is retained. t′ In and k i′ Data at the same position where there is no data, k i′ M indicates that k is retained. i′ Data in locations containing data.

[0077] Understandably, k t′ (1-M)+k i′ M can be understood as k tρ and k i′ The retained data is combined; that is, for a position in K space, if this position is in k... iρ If there is data, then the data at this position is k. iρ The data at the corresponding position in the middle, if this position is in k iρ If there is no data in the table, then the data corresponding to this position is k. t′ The data at the corresponding position in the middle.

[0078] In some embodiments, the dynamic medical imaging system can perform an inverse Fourier transform on the third K-space data obtained by the above method to obtain an output image.

[0079] As an example, such as Figure 5 As shown, for the first image x i′ Perform a Fourier transform to obtain the first K-space data k i′ For the denoised image x t′ Perform a Fourier transform to obtain the second K-space data k t′ Based on the first K-space data k i′ Second K-space data k t′ Obtain the third K-space data k t′ (1-M)+k i′ M, the inverse Fourier transform of the third K-space data is used to obtain the output image x. t .

[0080] In some embodiments, data consistency processing can be omitted, and the dynamic medical imaging system can denoise the input image of the current image processing based on the reference image to directly obtain the output image of the current image processing.

[0081] Introducing data consistency processing during the denoising process ensures the consistency between the reconstructed second image and the original data (first image), avoiding over-smoothing or distortion, thus preserving more detail information, significantly improving image quality, and providing clearer diagnostic information. Data consistency processing, as a constraint, makes the reconstruction process more stable.

[0082] Through multiple rounds of image processing, noise and artifacts in the image are gradually reduced, while the signal-to-noise ratio, contrast, and detail display are progressively enhanced, ultimately generating a high-quality second image. Simultaneously, multiple rounds of image processing can progressively optimize image quality, avoiding over-smoothing or distortion caused by a single processing step. Each round of processing improves upon the previous one, ensuring a steady improvement in image quality.

[0083] Figure 8 This is an exemplary schematic diagram illustrating the generation of a second image according to other embodiments of this specification.

[0084] In some embodiments, taking phase 1 as an example, the dynamic medical imaging system can utilize a trained image processing model to perform multiple rounds of image processing based on a first image containing noise (such as Gaussian noise) and a reference image (integrated phase). Each round of image processing includes: denoising the input image of this round of image processing based on the reference image to obtain a denoised image G. θ Then, data consistency processing (DC) is performed on the denoised image to obtain the output image. The input image is the first image or the output image of the previous round of image processing. Based on the output image of the last round of image processing, a second image corresponding to the phase of the first image (such as phase 1) is generated. (About...) Figure 8 For details, please refer to Figures 3-5 The description in the text.

[0085] Figure 6 This is an exemplary schematic diagram illustrating the training process of an image processing model according to some embodiments of this specification.

[0086] In some embodiments, the training process of the image processing model includes: acquiring training samples, determining a sample reference image based on sample first images corresponding to at least two sample periods among a plurality of sample periods, and training an initial image processing model based on the sample reference image and a plurality of sample second images to obtain the image processing model.

[0087] In some embodiments, the training samples include at least one set of sample data. Each set of sample data includes multiple first images and multiple second images corresponding to multiple sample periods of the target object. Each first image corresponds to one sample period from the multiple sample periods, and each second image corresponds to one first image. The image quality of the second image is higher than that of the corresponding first image. The definitions of the target object and the sample period are similar to those of the target object and the period, and will not be repeated here.

[0088] The first image of the sample refers to the undersampled image of the target object.

[0089] The second sample image refers to the fully sampled image acquired from the sample target object. A fully sampled image is an image obtained by the imaging device fully sampling the K-space during the scanning process of the sample target object, obtaining fully sampled K-space data, and then reconstructing the fully sampled K-space data (e.g., by inverse Fourier transform).

[0090] In some embodiments, the dynamic medical imaging system can scan the target object of the sample within multiple sample phases, use the fully sampled images corresponding to the multiple sample phases as the second sample images, delete some data from the fully sampled K-space data corresponding to the multiple second sample images to obtain undersampled K-space data corresponding to the multiple sample phases, and perform inverse Fourier transform on the undersampled K-space data corresponding to the multiple sample phases to obtain the first sample images.

[0091] In some embodiments, the dynamic medical imaging system can perform standardization processing on the first sample image and the second sample image. For example, the dynamic medical imaging system can normalize the gray values ​​of different images in the training samples, so that the gray values ​​of the images are limited to a certain range, such as [0, 1], [0, 255], etc.

[0092] A sample reference image is an image generated by combining sample first images from at least two sample phases. Sample reference images and reference images are essentially similar. The method for determining a sample reference image based on a sample first image is the same as the method for determining a reference image based on a first image; its implementation is described in step 210 regarding the method for determining a reference image based on a first image.

[0093] In some embodiments, the dynamic medical imaging system can train an initial image processing model based on a sample reference image and multiple sample second images to obtain an image processing model.

[0094] In some embodiments, the dynamic medical imaging system can perform a forward diffusion process on a sample second image, that is, add noise to a sample second image multiple times to obtain multiple noisy images. The dynamic medical imaging system can select a target noisy image from the multiple noisy images and perform a reverse diffusion process on the target noisy image, that is, input the target noisy image and the sample reference image into an initial image processing model to obtain the predicted noise output by the initial image processing model, determine the loss value of the loss function based on the predicted noise and the noise corresponding to the target noisy image, and update the initial image processing model based on the loss value.

[0095] A noisy image is an image obtained by adding noise to a second sample image.

[0096] In some embodiments, the dynamic medical imaging system adds noise to a sample second image multiple times, including: randomly adding noise to the sample second image to generate a first noisy image, randomly adding noise to the first noisy image to generate a second noisy image, and repeating the above process to generate multiple noisy images, so that the final generated noisy image approximates a simple distribution. Here, a simple distribution refers to a probability distribution with a simple form and intuitive characteristics, such as a Gaussian distribution.

[0097] The target noisy image refers to the noisy image input to the initial image processing model. In some embodiments, the target noisy image can be obtained in various ways, such as random selection.

[0098] Predictive noise refers to the noise added to the preceding noisy image when the initial image processing model predicts the generated target noisy image based on the input target noisy image and the sample reference image. For example, if the target noisy image is the 8th noisy image, then the predicted noise is the noise predicted by the initial image processing model to be added to the 7th noisy image based on the input target noisy image and the sample reference image. The preceding noisy image refers to the noisy image preceding the target noisy image.

[0099] In some embodiments, the dynamic medical imaging system determines a first loss value for a loss function based on predicted noise and actual noise added to a preceding noisy image when the target noisy image is generated. For example, the dynamic medical imaging system constructs a first loss function using the predicted noise and the actual noise corresponding to the target noisy image, and obtains the corresponding first loss value. The actual noise corresponding to the target noisy image refers to the actual noise added to the preceding noisy image when the target noisy image is actually generated. The dynamic medical imaging system updates the initial image processing model based on the first loss value.

[0100] In some embodiments, the dynamic medical imaging system may also determine an intermediate image based on the predicted noise and the target noisy image, and compare the intermediate image with at least one of the preceding noisy image, the sample reference image, and the sample second image of the target noisy image to determine the loss value.

[0101] An intermediate image refers to the image obtained after processing the noisy target image. In some embodiments, a dynamic medical imaging system can subtract predicted noise from the noisy target image to obtain an intermediate image.

[0102] In some embodiments, the dynamic medical imaging system constructs a second loss function based on an intermediate image and at least one of a preceding noisy image, a sample reference image, and a sample second image of the target noisy image, and obtains a corresponding second loss value. Specifically, the dynamic medical imaging system compares the intermediate image with at least one of a preceding noisy image, a sample reference image, and a sample second image of the target noisy image to construct the second loss function and obtain a corresponding second loss value.

[0103] By comparing the intermediate image with at least one of multiple images, the determined loss value can be more accurate, which is beneficial for subsequent updates to the model parameters.

[0104] In some embodiments, the dynamic medical imaging system can determine an intermediate image based on the predicted noise and the target noisy image, perform data consistency processing on the intermediate image based on the target noisy image to obtain a sample denoised image, compare the sample denoised image with at least one of the preceding noisy image, the sample reference image and the sample second image, construct a third loss function, and obtain the corresponding third loss value.

[0105] A sample denoised image refers to an image obtained after performing data consistency processing on an intermediate image.

[0106] The method of performing data consistency processing on intermediate images based on the target noisy image is similar to the method of performing data consistency processing on denoised images based on the first image. The implementation method can be found in [link to relevant documentation]. Figure 4 and Figure 5 The method of performing data consistency processing on the denoised image based on the first image.

[0107] In some embodiments, the dynamic medical imaging system constructs a third loss function based on at least one of the sample denoised image, the preceding denoised image, the sample reference image, and the sample second image, and obtains the corresponding third loss value.

[0108] The dynamic medical imaging system determines a target loss value based on a first loss value and a second loss value, or based on a first loss value and a third loss value, and updates the initial image processing model according to the target loss value. For example, the dynamic medical imaging system determines the target loss value by taking a weighted average of the first loss value and the second loss value, or the first loss value and the third loss value.

[0109] In some embodiments, after the current backdiffusion process is completed, it can be determined whether a termination condition is met. Termination conditions include loss function convergence, the number of executions of the backdiffusion process reaching a threshold, etc. If the termination condition is met, the updated model is output as the trained image processing model. If the termination condition is not met, the next round of backdiffusion process continues.

[0110] The next round of reverse diffusion can be performed by randomly selecting another target noisy image from among the multiple noisy images. Alternatively, the reverse diffusion process can be performed sequentially from the final noisy image with a near-simple distribution to the first noisy image. In this case, the preceding noisy image of the current noisy image is selected for the next round of reverse diffusion. In some embodiments, after the forward and reverse diffusion processes for a sample second image have been completed, another sample second image can be selected to perform the forward and reverse diffusion processes.

[0111] As an example, such as Figure 6 As shown, a set of sample data in the training samples includes multiple sample first images corresponding to phases 1-P of the target object. and multiple sample second images A sample reference image x is generated by combining the first image of at least two sample phases. R The dynamic medical imaging system can produce a second image of the sample corresponding to phase 1. Perform a forward diffusion process, that is, perform forward diffusion on the second image of the sample. Noise is added multiple times to obtain N noisy images. A dynamic medical imaging system can select a target noisy image from N noisy images. Adding noise to the target image Perform a reverse diffusion process, that is, add noise to the target image. and sample reference image x R Input the initial image processing model and obtain the prediction noise S′ output by the initial image processing model. n Based on the predicted noise S′ n The noise S corresponding to the target noisy image n The loss value of the loss function is determined, and the initial image processing model is updated based on the loss value.

[0112] Optionally, a dynamic medical imaging system can extract noise from a target image. Subtract the prediction noise S′ n Determine the intermediate image, based on the target noisy image. Data consistency processing is performed on the intermediate image to obtain the sample denoised image x′. The sample denoised image x′ is then compared with the preceding denoised image. Sample reference image x R and the second image of the sample The comparison is performed to obtain the loss value, and the initial image processing model is updated based on the loss value.

[0113] By using the target noisy image to perform data consistency processing on the intermediate image, not only can the original data in the target noisy image be preserved, but the resulting denoised sample image can also be made closer to the true value, improving the accuracy of the loss value and further enhancing the fidelity of the output data of the image processing model.

[0114] By using both the noisy image and a sample reference image as input, the image processing model can leverage the high-quality prior information provided by the sample reference image to better learn how to separate noise from the noisy image. This allows the model to predict noise more accurately and reduces the possibility of misclassification. By calculating the loss value, the image processing model can evaluate its accuracy in predicting noise and adjust its parameters accordingly, gradually optimizing the model's performance. Through continuous parameter updates, the image processing model can gradually improve its ability to predict noise, ultimately generating high-quality denoised images.

[0115] The reference images provide high-quality prior information for the image processing model, helping it to better learn how to recover a high-quality second image from a low-quality first image. By training the image processing model, image quality can be improved while enhancing the model's generalization ability and robustness, enabling its widespread application in medical imaging fields such as MRI, CT, and PET, significantly improving imaging results and clinical application value.

[0116] Figure 7 This is an exemplary schematic diagram of the training process of an image processing model according to other embodiments of this specification.

[0117] In some embodiments, such as Figure 7 As shown, the training process of the image processing model includes: collecting multiple undersampled (sample first image) and fully sampled images (sample second image) from N phases (e.g., phase 0, phase 1, ..., phase N-1), and performing data standardization on them. Noise is progressively added to the fully sampled images to generate multiple noisy images G. θ This results in the generated noisy image including Gaussian noise. Undersampled images from N phases are integrated (e.g., summed or averaged) and used as the generation condition (sample reference image) for the diffusion model. Based on this generation condition, noise reduction is performed progressively, starting with Gaussian noise, training the network to generate fully sampled images of different phases from this generation condition. Regarding... Figure 7 For details, please refer to Figure 6 The description in the text.

[0118] Conditional diffusion models are used in dynamic imaging to generate multiple noisy images through a forward diffusion process, making the training samples more diverse. By iteratively removing noise from noisy images, conditional diffusion models can produce clear and detailed results in image reconstruction tasks. By incorporating additional conditional information, conditional diffusion models can generate images that meet specific requirements or features. Compared to some adversarial networks, the training process of conditional diffusion models is more stable and less susceptible to training instability. They can also generate reasonable results with a wider range of initial noise levels and are less prone to significant degradation in generation quality due to fluctuations in input noise levels. Conditional diffusion models lose information during forward diffusion and then recover it during back diffusion, and do not directly rely on the specific features of the original training samples, thus exhibiting strong resistance to overfitting.

[0119] The aforementioned dynamic imaging methods can be applied to dynamic reconstruction of cardiac MRI. Therefore, this application proposes a cardiac MRI cine reconstruction method based on a conditional diffusion generative network, aiming to recover high-quality cardiac cine from highly undersampled k-space data. Cardiac MRI cine imaging provides comprehensive information about the heart and plays a crucial role in cardiac diagnosis and management. However, to capture rapid real-time dynamic information, each phase of the MRI cine cannot fully acquire the signals required for image reconstruction. Existing technologies have shortcomings in terms of reconstruction speed and accuracy, which may lead to a decrease in cine quality and loss of information. This application proposes using a deep learning network based on a conditional diffusion model, which can fully utilize information from different phases and introduce a data consistency term to ensure the authenticity of the generated image, rapidly recovering high-quality cardiac MRI cine from highly undersampled images, enabling rapid and high-precision reconstruction of cardiac cine from newly scanned patients. This method can significantly improve the clinical applicability and accuracy of cardiac MRI cine, helping physicians to better observe and diagnose, and contributing to a better assessment of cardiac function and health status.

[0120] For ease of description, this application uses dynamic magnetic resonance imaging as an example. It is understood that the dynamic medical imaging method of this application can also be applied to other modalities of dynamic imaging, such as dynamic CT imaging. The application of the dynamic medical imaging method in dynamic CT imaging is essentially similar to its application in dynamic MRI imaging.

[0121] In CT dynamic imaging, the first image refers to the sparsely sampled image acquired by the imaging device. During the scanning process of the target object, the imaging device performs sparse sampling to obtain sparsely sampled projection data, and then reconstructs this projection data (e.g., filtered backprojection) to obtain the first image. In sparse sampling, the number of projection angles acquired during the scan is less than in full sampling. For example, in full sampling, projection data is acquired every 1° or 0.5° within a range of 360° or 180° + θ (beam fan angle). In sparse sampling, projection data is acquired only for a portion of the projection angles corresponding to those in full sampling.

[0122] In dynamic CT imaging, the first sample image refers to a sparsely sampled image acquired of the target object. The second sample image refers to a fully sampled image acquired of the target object. A fully sampled image is an image obtained by the imaging device fully sampling the target object during scanning, obtaining fully sampled projection data, and then reconstructing the fully sampled projection data (e.g., filtered backprojection or iterative reconstruction). For example, within a range of 360° or 180° + θ (beam fan angle), projection data is acquired every 1° or 0.5° to obtain fully sampled projection data. The dynamic medical imaging system can scan the target object within multiple sample phases, using the fully sampled images corresponding to multiple sample phases as the second sample images. Then, projection data corresponding to certain projection angles is deleted from the fully sampled projection data corresponding to multiple second sample images to obtain sparse projection data corresponding to multiple sample phases. Filtered backprojection is then performed on this sparse projection data to obtain multiple first sample images.

[0123] In the embodiments described in this specification, the order of the steps is interchangeable unless otherwise specified, and steps may be omitted. Other steps may also be included in the operation process.

[0124] The embodiments in this specification are merely illustrative and not intended to limit the scope of this specification. Various modifications and alterations that can be made by those skilled in the art under the guidance of this specification remain within its scope.

[0125] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are sometimes modified by the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. Although the numerical ranges and parameters used to confirm their breadth in some embodiments of this specification are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

Claims

1. A dynamic medical imaging method, characterized in that, The method includes: Obtain multiple first images corresponding to multiple periods of the target object, where each first image corresponds to one of the multiple period phases; A reference image is determined based on the first image corresponding to at least two of the plurality of phases; The plurality of first images and the reference image are input into a trained image processing model to obtain a plurality of second images. Each second image corresponds to one of the plurality of first images, and the image quality of the second image is higher than that of the corresponding first image.

2. The method as described in claim 1, characterized in that, The determination of the reference image based on the first image corresponding to at least two of the plurality of phases includes: The reference image is obtained by combining the pixel values ​​or voxel values ​​of the first images corresponding to at least two periods; or Perform a Fourier transform on the first images corresponding to the at least two periods to obtain at least two K-space datasets; The at least two K-space datasets are combined to obtain reference K-space data; The reference image is obtained by performing an inverse Fourier transform on the reference K-space data.

3. The method as described in claim 1, characterized in that, The step of inputting the plurality of first images and the reference image into the trained image processing model to obtain a plurality of second images includes: For each of the first images Using the trained image processing model, multiple rounds of image processing are performed based on the first image and the reference image, wherein each round of image processing includes: Based on the reference image, the input image of this round of image processing is denoised to obtain the output image, wherein the input image is the first image or the output image of the previous round of image processing; Based on the output image of the last round of image processing, the second image corresponding to the first image is generated.

4. The method as described in claim 3, characterized in that, The step of denoising the input image in this round of image processing based on the reference image to obtain the output image includes: The input image for this round of image processing is denoised based on the reference image to obtain a denoised image; Based on the first image, data consistency processing is performed on the denoised image to obtain the output image of this round of image processing.

5. The method as described in claim 1, characterized in that, The image processing model is a deep learning model based on the conditional diffusion model.

6. The method as described in claim 5, characterized in that, The training process of the image processing model includes: Acquire training samples, which include multiple sample first images and multiple sample second images corresponding to multiple sample periods of the sample target object. Each sample first image corresponds to one sample period of the multiple sample periods, and each sample second image corresponds to one sample first image. The image quality of the sample second image is higher than that of the corresponding sample first image. Based on the first images of samples corresponding to at least two sample periods among the plurality of sample periods, a sample reference image is determined; The initial image processing model is trained based on the sample reference image and the multiple sample second images to obtain the image processing model.

7. The method as described in claim 6, characterized in that, The step of training the initial image processing model based on the sample reference image and the plurality of sample second images to obtain the image processing model includes: By adding noise multiple times to a second image of a sample, multiple noisy images are obtained. Select a target noisy image from the plurality of noisy images; The target noisy image and the sample reference image are input into the initial image processing model to obtain the predicted noise output by the initial image processing model; Based on the predicted noise and the noise corresponding to the target noisy image, the loss value of the loss function is determined; The initial image processing model is updated based on the loss value to obtain the image processing model.

8. A dynamic medical imaging system, characterized in that, The system includes an acquisition module, a condition generation module, and an image processing module; The acquisition module is configured to acquire multiple first images corresponding to multiple periods of the target object, and each first image corresponds to one of the multiple period phases; The condition generation module is configured to determine a reference image based on a first image corresponding to at least two of the plurality of phases; The image processing module is configured to input the plurality of first images and the reference image into a trained image processing model to obtain a plurality of second images, each second image corresponding to one of the plurality of first images, and the image quality of the second image is higher than that of the corresponding first image.

9. A dynamic medical imaging device, characterized in that, The device includes at least one processor and at least one memory; The at least one memory is used to store computer instructions; The at least one processor is configured to execute at least a portion of the computer instructions to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions. When the computer reads the computer instructions from the storage medium, the computer executes the method as described in any one of claims 1 to 7.

Images