A system and method for generating contrast-free angiography based on mask guidance

Abstract

The present invention relates to a mask-guided, contrast-free angiography generation system and method. A parameter optimization device optimizes the parameters of a conversion device using a loss function that emphasizes contrast-enhanced shadows, improving the conversion device's ability to capture complex image details. The conversion device then integrates a Gaussian noise image into a contrast-free computed tomography image, and a computed tomography angiography is generated through a fusion of feature sampling and progressive denoising. This ensures sufficient consistency and realism in the resulting computed tomography angiography, improving the sharpness and fidelity of the generated vascular structures.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a mask-guided contrast-free angiography generation system and method. Background Art

[0002] Computed tomography angiography (CTA), based on the use of iodinated contrast agents (ICAs), can provide high-resolution and three-dimensional vascular images. Its biggest advantage over non-contrast computed tomography images (NCCT) is that CTA can provide clearer and more detailed vascular anatomical images, thereby enhancing the contrast between blood vessels and surrounding tissues.

[0003] In order to convert contrast-free computed tomography images into computed tomography angiography, the prior art discloses a RegGAN model architecture that introduces regularization techniques to enhance image generation quality, improve structural consistency, and reduce artifacts.

[0004] However, this model architecture has difficulty generating high-quality images with sufficient consistency and realism in complex imaging scenarios. Specifically, the generated computed tomography angiography contains artifacts, which manifest as irregular vascular shapes, imprecise boundaries, and unnatural textures, thereby weakening the clinical practicality of the generated images. Summary of the Invention

[0005] The technical problem to be solved by the present invention is how to overcome the technical defect that the existing RegGAN model architecture is difficult to generate high-quality images with sufficient consistency and realism. In order to overcome this technical defect, the present invention provides a mask-guided contrast-free angiography generation system and method, specifically including a mask-guided contrast-free angiography generation system and a mask-guided contrast-free angiography generation method.

[0006] The present invention provides a mask-guided contrast-free angiography generation system, comprising:

[0007] The noise generating device is configured to generate a Gaussian noise image using a random Gaussian noise iterative algorithm;

[0008] an acquisition device configured to acquire a non-contrast computed tomography image;

[0009] a conversion device, electrically connected to both the noise generating device and the acquisition device, configured to integrate the Gaussian noise image into the contrast-free CT image and obtain a CT angiogram by a fusion method of feature sampling and stepwise denoising;

[0010] A parameter optimization device is electrically connected to the conversion device and is configured to generate a binary contrast enhancement mask using a data set based on a subtraction strategy, substitute the binary contrast enhancement mask into a loss function for highlighting the vascular area affected by the contrast agent to optimize the parameters of the conversion device using the loss function, obtain model parameters and transmit them to the conversion device.

[0011] The present invention discloses a mask-guided contrast-free angiography generation system, comprising a noise generation device, an acquisition device, a conversion device, and a parameter optimization device. The parameter optimization device optimizes the parameters of the conversion device using a loss function that emphasizes contrast-enhanced shadows, thereby enhancing the conversion device's ability to capture complex image details. The conversion device integrates a Gaussian noise image into a contrast-free computed tomography image and obtains a computed tomography angiography through a fusion of feature sampling and stepwise denoising, ensuring sufficient consistency and realism in the resulting computed tomography angiography. This integration of a loss function that emphasizes contrast-enhanced shadows into a diffusion process based on a Gaussian noise image generated using a random Gaussian noise iterative algorithm not only effectively converts the contrast-free computed tomography image into a higher-quality computed tomography angiography, but also stabilizes the parameter optimization process, improving the sharpness and fidelity of the generated vascular structure, thereby overcoming the technical shortcomings of existing RegGAN model architectures.

[0012] In a possible implementation, the iterative formula of the random Gaussian noise iterative algorithm is as follows:

[0013] ,

[0014] ,

[0015] ,

[0016] Where,

[0017] is a composite number, representing the number of diffusion steps initially planned;

[0018] is a composite number The root cause of

[0019] Represents the iteration to Step noise;

[0020] Represents the iteration to The noise addition term of the step;

[0021] Represents an arbitrarily small number;

[0022] Represents the noise obtained The noise is obtained under the condition The conditional probability distribution of ;

[0023] represents the identity matrix;

[0024] represents Gaussian distribution;

[0025] Represents the minimum value of index progress;

[0026] Represents the maximum value of the exponential progress.

[0027] When executing the above iterative formula, as the time step increases, the diffusion step will be much larger than 1, thereby reducing the number of diffusion steps, and eventually reducing the number of diffusion steps from T to T / k, greatly improving the noise generation efficiency.

[0028] In one possible embodiment, the conversion device is a communication network structure composed of multiple antagonistic diffusion generation modules connected in series. The antagonistic diffusion generation modules are configured to convert an input image into an output image through a feature sampling method of encoding and decoding and a Bayesian rule sampling denoising method. This not only enables the integration of a Gaussian noise image into a contrast-free computed tomography image, but also further ensures that the obtained computed tomography angiography has sufficient consistency and realism.

[0029] In one possible implementation, the adversarial diffusion generation module includes an encoder, a decoder, a Bayesian posterior sampling submodule, a contrast enhancement attention guidance submodule, and a contrast perception discrimination submodule;

[0030] The input end of the encoder in the first adversarial diffusion generation module is electrically connected to the output end of the noise generation device and the output end of the acquisition device, and the input ends of the encoders in the remaining adversarial diffusion generation modules are electrically connected to the output end of the Bayesian posterior sampling submodule in the previous adversarial diffusion generation module;

[0031] In each of the adversarial diffusion generation modules, the input end of the decoder is electrically connected to the output end of the encoder, the input end of the Bayesian posterior sampling submodule is electrically connected to the output end of the decoder, the input end of the contrast perception discrimination submodule is electrically connected to the output end of the Bayesian posterior sampling submodule and the output end of the parameter optimization device at the same time, the input end of the contrast enhancement attention guidance submodule is electrically connected to the output end of the decoder and the output end of the parameter optimization device at the same time, and the input end of the encoder is also electrically connected to the output end of the contrast perception discrimination submodule and the output end of the contrast enhancement attention guidance submodule at the same time.

[0032] In one possible embodiment, the encoder is configured to perform a convolution operation on its input image to extract anatomical structure and texture information, incorporate temporal information into the extraction process, and output an encoded image;

[0033] The decoder is configured to perform deconvolution denoising on the encoded image to obtain an estimated denoised image;

[0034] The Bayesian posterior sampling submodule is configured to denoise the estimated denoised image using a Bayesian sampling rule to obtain a denoised result;

[0035] The contrast enhancement attention guidance submodule is configured to, after the parameter optimization device is started, perform element-wise multiplication and ReLU activation on the estimated denoised image and the binary contrast enhancement mask in sequence to generate an attention map, and optimize the parameters of the encoder and decoder in the adversarial diffusion generation module in which it is located as a target to obtain the minimum value of the distance between the attention map and the binary contrast enhancement mask;

[0036] The contrast perception discrimination submodule is configured to optimize the parameters of the encoder, decoder and Bayesian posterior sampling submodule in the adversarial diffusion generation module in which it is located, with the goal of obtaining the minimum value of the non-saturated adversarial loss function with gradient penalty after the parameter optimization device is started.

[0037] The adversarial diffusion generation module with the above-mentioned structure and functions can integrate Gaussian noise images into contrast-free computed tomography images through the encoder and decoder, and denoise them through the Bayesian posterior sampling submodule. Combined with the included contrast enhancement attention guidance submodule and angiography perception discrimination submodule, it can independently optimize the parameters of the encoder, decoder and Bayesian posterior sampling submodule of each module, thereby further ensuring that the obtained computed tomography angiography has sufficient consistency and realism.

[0038] In a possible implementation, the parameter optimization device includes:

[0039] an unsupervised mask generation module, the output of which is electrically connected to the inputs of all the contrast enhancement attention guidance submodules and the input of the contrast perception discrimination submodule, and is configured to generate a binary contrast enhancement mask using the dataset based on a subtraction strategy;

[0040] A mask-guided feature alignment module, whose input end is electrically connected to the output end of the unsupervised mask generation module and the output end of the Bayesian posterior sampling submodule in the adversarial diffusion generation module located at the end, is configured to substitute the binary contrast enhancement mask into the loss function to optimize the parameters of the conversion device using the loss function, obtain model parameters and transmit them to all the encoders, the decoders and the Bayesian posterior sampling submodules.

[0041] In the parameter optimization device with the above structure, a supervised mask generation module uses the data set to generate a binary contrast enhancement mask, and the mask-guided feature alignment module is used to achieve model parameter optimization, which stabilizes the parameter optimization process and further improves the sharpness and fidelity of the generated vascular structure.

[0042] In a possible implementation, the calculation formula of the subtraction strategy is:

[0043] ,

[0044] Where,

[0045] Represents the first Binary contrast enhancement mask of data;

[0046] Represents the first Computed tomography angiography of 1000 data;

[0047] Represents the first Contrast-free computed tomography images of 1000 images;

[0048] Represents binarization processing.

[0049] The binary contrast enhancement mask constructed by the above formula is actually an unsupervised contrast enhancement mask. Combined with the introduced contrast perception technology, it can focus on clinically relevant contrast areas and reduce the impact of pixel misalignment, enhancing the accuracy of vascular structure and making the generated image have a more accurate vascular structure.

[0050] In one possible implementation, the loss function is calculated as follows:

[0051] ,

[0052] ,

[0053] ,

[0054] ,

[0055] ,

[0056] ,

[0057] Where,

[0058] represents the loss function;

[0059] 、 、 and Both represent weight factors;

[0060] stands for element-wise multiplication;

[0061] represents the target PatchNCE loss function;

[0062] represents the source PatchNCE loss function;

[0063] represents the attention loss function;

[0064] represents the combined discriminator loss function;

[0065] Represents the number of data in the dataset;

[0066] represents the anchor point complement determined on the image output by the conversion device during parameter optimization;

[0067] represents a positive sample patch defined on the input image of the conversion device when performing parameter optimization;

[0068] represents a negative sample patch defined on the input image of the conversion device during parameter optimization;

[0069] represents the first Attention map;

[0070] represents the denoising result obtained by the Bayesian posterior sampling submodule during parameter optimization;

[0071] represents the binary contrast enhancement mask;

[0072] Represents the InfoNCE loss function;

[0073] represents the non-saturating adversarial loss function with gradient penalty.

[0074] The contrast perception technology corresponding to the above loss function, when combined with the binary contrast enhancement mask, can enhance the accuracy of vascular structure on the basis of focusing on clinically relevant contrast areas and reducing the impact of pixel misalignment, so that the generated image has a more accurate vascular structure.

[0075] Another technical solution of the present invention is to provide a method for generating contrast-free angiography based on mask guidance, comprising the following steps:

[0076] S1: generating a binary contrast enhancement mask using a data set by a parameter optimization device, and substituting the binary contrast enhancement mask into a loss function for highlighting a vascular region affected by a contrast agent to optimize parameters of a conversion device using the loss function, obtaining model parameters, and transmitting the obtained model parameters to the conversion device;

[0077] S2: a noise generating device generates a Gaussian noise image using a random Gaussian noise iterative algorithm, and an acquisition device acquires a contrast-free computed tomography image;

[0078] S3: The Gaussian noise image is integrated into the contrast-free CT image by the conversion device, and CT angiography is obtained by a fusion method of feature sampling and stepwise denoising.

[0079] The mask-guided contrast-free angiography generation method disclosed in the present invention first optimizes the parameters of a conversion device using a loss function that emphasizes contrast-enhanced shadows, thereby improving the conversion device's ability to capture complex image details. The conversion device then integrates a Gaussian noise image into the contrast-free computed tomography image, and obtains a computed tomography angiography through a fusion of feature sampling and stepwise denoising. Ultimately, this ensures sufficient consistency and realism in the resulting computed tomography angiography. This integration of a loss function that emphasizes contrast-enhanced shadows into a diffusion process based on a Gaussian noise image generated by a random Gaussian noise iterative algorithm not only effectively facilitates the conversion of contrast-free computed tomography images into higher-quality computed tomography angiography, but also stabilizes the parameter optimization process, improving the sharpness and fidelity of the generated vascular structures, thereby overcoming the technical shortcomings of existing RegGAN model architectures.

[0080] In a possible implementation, step S1 includes the following steps:

[0081] S101: Retrieving multiple sets of contrast-free computed tomography images and their corresponding computed tomography angiography images from a database to obtain the data set;

[0082] S102: inputting all data in the data set into the parameter optimization device and the conversion device;

[0083] S103: generating a binary contrast enhancement mask using the data set through an unsupervised mask generation module in the parameter optimization device;

[0084] S104: Processing the data set by the encoder and decoder of each anti-diffusion generation module in the conversion device to obtain an estimated denoised image corresponding to each anti-diffusion generation module;

[0085] S105: performing element-wise multiplication and ReLU activation on the estimated denoised image and the binary contrast enhancement mask in sequence through the contrast enhancement attention guidance submodule of each adversarial diffusion generation module in the conversion device to generate an attention map corresponding to each adversarial diffusion generation module;

[0086] S106: Optimizing parameters of the encoder and decoder in the adversarial diffusion generation module using the attention map obtained in step S105 through each of the contrast enhancement attention guidance submodules;

[0087] S107: Obtaining denoising results corresponding to the respective adversarial diffusion generation modules according to the estimated denoised image through the Bayesian posterior sampling submodules of the respective adversarial diffusion generation modules in the conversion device;

[0088] S108: Optimizing parameters of the encoder, decoder, and Bayesian posterior sampling submodule in each of the anti-diffusion generation modules in the conversion device according to the denoising results obtained in step S107 by using the contrast perception discrimination submodule of each anti-diffusion generation module in the conversion device;

[0089] S109: Substitute the binary contrast enhancement mask into the loss function through the mask-guided feature alignment module in the parameter optimization device to optimize the parameters of the conversion device using the loss function, obtain model parameters and transmit them to all the encoders, the decoders and the Bayesian posterior sampling sub-modules.

[0090] In the above-mentioned parameter optimization process, by introducing and integrating the binary contrast enhancement mask, not only the original loss function is expanded and the areas most significantly affected by the contrast agent in the final image obtained by the conversion device are emphasized, but also the conversion device can be guided to prioritize the vascular areas most affected by the contrast agent, thereby improving the accuracy and continuity of these key anatomical features. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 Schematic diagram of the structure of a mask-guided contrast-free angiography generation system disclosed in an embodiment of the present invention;

[0092] Figure 2 This is a schematic diagram of the operating principle of the conversion device disclosed in an embodiment of the present invention;

[0093] Figure 3 Schematic diagram of the structure of the anti-diffusion generation module disclosed in an embodiment of the present invention;

[0094] Figure 4 is a schematic diagram of a parameter optimization device disclosed in an embodiment of the present invention;

[0095] Figure 5 A flow chart of the method disclosed in an embodiment of the present invention;

[0096] Figure 6 This is a flow chart of step S1 disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0097] First, those skilled in the art should understand that these embodiments are merely used to explain the technical principles of the embodiments of the present application and are not intended to limit the scope of protection of the embodiments of the present application. Those skilled in the art may adjust them as needed to suit specific application scenarios.

[0098] In the description of the embodiments of the present application, it should be noted that, unless otherwise expressly specified or limited, the terms "electrically connected" and "establishing an electrical connection relationship" should be understood in a broad sense, that is, it should be understood that two or more devices have an electrical relationship, which can be achieved through a wire connection, a wireless connection, or a combination of the two; and can be directly connected or indirectly connected through an intermediate medium. For those skilled in the art, the specific meanings of the above terms in the embodiments of the present application can be understood in specific circumstances.

[0099] In the description of the embodiments of the present application, it should be noted that, unless otherwise clearly stipulated and limited, any reference to parameter optimization of a module or device through or using a certain loss function refers to a series of processes in which the model parameters of this module or device are used as the parameters to be determined, and the optimal value of the loss function is obtained as the goal, and the specific values ​​of the parameters to be determined are obtained by using an optimization model solving algorithm (such as a particle swarm optimization algorithm, a greedy algorithm, Newton's method, and a reverse gradient method), and the specific values ​​of the parameters to be determined are used as the model parameters of this module or device.

[0100] In the embodiments of the present application, unless otherwise expressly specified or limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. Furthermore, a first feature being "above," "above," and "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0101] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0102] See also Figures 1 to 6 The present application discloses a mask-guided contrast-free angiography generation system. Figure 1 This is a schematic diagram of the structure of the angiography generation system, which includes a noise generation device, an acquisition device, a conversion device, and a parameter optimization device. The conversion device is electrically connected to both the noise generation device and the acquisition device, and the parameter optimization device is electrically connected to the conversion device. The acquisition device is configured to acquire non-contrast computed tomography images (NCCT images are referred to in all figures as non-contrast computed tomography images). Single arrows in the figure indicate that data can only be transmitted in one direction, in the direction of the arrow, while double arrows indicate that data can be transmitted in both directions between the two connected devices.

[0103] In the angiography generation system, the noise generation device is configured to generate a Gaussian noise image using a random Gaussian noise iterative algorithm, i.e., pure Gaussian noise. Specifically, in this embodiment, the iterative formula of the random Gaussian noise iterative algorithm is as follows:

[0104] ,

[0105] ,

[0106] ,

[0107] Where,

[0108] is a composite number, representing the number of diffusion steps initially planned;

[0109] is a composite number The root cause of Divide by 1 and Factors other than represent the iterative diffusion step length;

[0110] Represents the iteration to Step noise;

[0111] Represents the iteration to The noise addition term of the step;

[0112] Represents an arbitrarily small number;

[0113] Represents the noise obtained The noise is obtained under the condition The conditional probability distribution of ;

[0114] represents the identity matrix;

[0115] represents Gaussian distribution;

[0116] Represents the minimum value of index progress;

[0117] Represents the maximum value of the exponential progress.

[0118] In the angiography generation system, a conversion device is configured to fuse a Gaussian noise image into a contrast-free computed tomography image, and obtain a computed tomography angiography (the computed tomography angiography is referred to as CTA image in all figures) by fusing feature sampling and stepwise denoising. Figure 2 For the operating principle diagram of the conversion device, see Figure 2 In this embodiment, the conversion device comprises a communication network structure composed of multiple adversarial diffusion generation modules connected in series. The adversarial diffusion generation modules are configured to convert the input image into the output image through encoding and decoding feature sampling and Bayesian rule sampling denoising. The number of adversarial diffusion generation modules should be equal to the number of noise addition steps in the noise generation device. This embodiment uses four adversarial diffusion generation modules, each corresponding to a denoising step in the diffusion model.

[0119] See also Figure 3 and Figure 4 In this embodiment, the adversarial diffusion generation module includes an encoder, a decoder, a Bayesian posterior sampling submodule, a contrast enhancement attention guidance submodule, and a contrast perception discrimination submodule. The input of the encoder in the first adversarial diffusion generation module is electrically connected to both the output of the noise generation device and the output of the acquisition device. The input of the encoders in the remaining adversarial diffusion generation modules is electrically connected to the output of the Bayesian posterior sampling submodule in the previous adversarial diffusion generation module. In each adversarial diffusion generation module, the input of the decoder is electrically connected to the output of the encoder, the input of the Bayesian posterior sampling submodule is electrically connected to the output of the decoder, the input of the contrast perception discrimination submodule is electrically connected to both the output of the Bayesian posterior sampling submodule and the output of the parameter optimization device, the input of the contrast enhancement attention guidance submodule is electrically connected to both the output of the decoder and the output of the parameter optimization device, and the input of the encoder is also electrically connected to both the output of the contrast perception discrimination submodule and the output of the contrast enhancement attention guidance submodule.

[0120] In the adversarial diffusion generation module, the encoder is configured to perform a convolution operation on its input image (if it is the encoder in the first adversarial diffusion generation module, the input image is a Gaussian noise image and a contrast-free computed tomography image) to extract anatomical structure and texture information, and incorporate temporal information in the extraction process to output an encoded image; the decoder is configured to perform deconvolution denoising on the encoded image to obtain an estimated denoised image; the Bayesian posterior sampling submodule is configured to denoise the estimated denoised image by using the Bayesian sampling rule to obtain a denoised result; the contrast enhancement attention guidance submodule is configured to, after the parameter optimization device is started, perform element-wise multiplication and ReLU activation on the estimated denoised image and the binary contrast enhancement mask in sequence to generate an attention map, and then optimize the parameters of the encoder and decoder in the adversarial diffusion generation module in which it is located with the goal of obtaining the minimum value of the distance between the attention map and the binary contrast enhancement mask; the contrast perception discrimination submodule is configured to, after the parameter optimization device is started, perform parameter optimization on the encoder, decoder and Bayesian posterior sampling submodule in the adversarial diffusion generation module in which it is located with the goal of obtaining the minimum value of the non-saturated adversarial loss function with gradient penalty.

[0121] In the angiography generation system, a parameter optimization device is configured to generate a binary contrast enhancement mask using a data set based on a subtraction strategy, substitute the binary contrast enhancement mask into a loss function for highlighting the vascular area affected by the contrast agent to optimize the parameters of the conversion device using the loss function, obtain the model parameters and transmit them to the conversion device.

[0122] See also Figure 4 In this embodiment, the parameter optimization device includes an unsupervised mask generation module and a mask-guided feature alignment module, wherein the output end of the unsupervised mask generation module is electrically connected to the input ends of all contrast enhancement attention guidance sub-modules and the input end of the contrast perception discrimination sub-module at the same time, and the input end of the mask-guided feature alignment module is electrically connected to the output end of the unsupervised mask generation module and the output end of the Bayesian posterior sampling sub-module in the adversarial diffusion generation module located at the end.

[0123] In the parameter optimization device, the unsupervised mask generation module is configured to generate a binary contrast enhancement mask using the data set based on a subtraction strategy. In this embodiment, the calculation formula of the subtraction strategy is:

[0124] ,

[0125] Where,

[0126] Represents the first Binary contrast enhancement mask of data;

[0127] Represents the first Computed tomography angiography of 1000 data;

[0128] Represents the first Contrast-free computed tomography images of 1000 images;

[0129] represents the binarization process. In this embodiment, its definition is:

[0130] ,

[0131] Among them represents a threshold value set by a threshold technology verified by empirical experiments, and in this embodiment, the value is 20.

[0132] In the parameter optimization device, the mask-guided feature alignment module is configured to substitute the binary contrast enhancement mask into the loss function to optimize the parameters of the conversion device using the loss function, obtain the model parameters and pass them to all encoders, decoders and Bayesian posterior sampling submodules. In this embodiment, the loss function is calculated as:

[0133] ,

[0134] ,

[0135] ,

[0136] ,

[0137] ,

[0138] ,

[0139] Where,

[0140] represents the loss function;

[0141] 、 、 and Both represent weight factors; in this embodiment, 、 、 、 ;

[0142] stands for element-wise multiplication;

[0143] represents the target PatchNCE loss function;

[0144] represents the source PatchNCE loss function;

[0145] represents the attention loss function;

[0146] represents the combined discriminator loss function;

[0147] Represents the number of data in the dataset;

[0148] represents the anchor point complement determined on the image output by the conversion device during parameter optimization;

[0149] represents the positive sample patch defined on the input image of the converter during parameter optimization;

[0150] Represents the negative sample patch defined on the input image of the converter during parameter optimization;

[0151] Represents the first Attention map;

[0152] Represents the denoising result obtained by the Bayesian posterior sampling submodule during parameter optimization;

[0153] represents binary contrast enhancement mask;

[0154] Represents the InfoNCE loss function;

[0155] represents the non-saturating adversarial loss function with gradient penalty.

[0156] The following will further disclose the method of using the mask-guided contrast-free angiography generation system in this embodiment. The flowchart of the method is as follows: Figure 5 As shown, the method includes the following steps:

[0157] S1: A parameter optimization device is used to generate a binary contrast enhancement mask using a data set, and the binary contrast enhancement mask is substituted into a loss function for highlighting the vascular area affected by the contrast agent to optimize the parameters of the conversion device through the loss function, obtain the model parameters and transmit them to the conversion device.

[0158] See also Figure 4 and Figure 6In this embodiment, step S1 includes the following steps:

[0159] S101: Retrieving multiple sets of contrast-free computed tomography images and their corresponding computed tomography angiography images from a database to obtain a data set.

[0160] S102: Input all data in the data set into the parameter optimization device and the conversion device.

[0161] S103: Generate a binary contrast enhancement mask using the data set through an unsupervised mask generation module in the parameter optimization device.

[0162] S104: Processing the dataset through the encoder and decoder of each adversarial diffusion generation module in the conversion device to obtain an estimated denoised image corresponding to each adversarial diffusion generation module. In this process, the estimated denoised image corresponding to the adversarial diffusion generation module is a function of the model parameters of the encoder and decoder of the adversarial diffusion generation module.

[0163] S105: Generate an attention map corresponding to each adversarial diffusion generation module by performing element-wise multiplication and ReLU activation on the estimated denoised image and the binary contrast enhancement mask through the contrast enhancement attention guidance submodule of each adversarial diffusion generation module in the conversion device. The attention map corresponding to the adversarial diffusion generation module is a function of the model parameters of the encoder and decoder of the adversarial diffusion generation module.

[0164] S106: Optimize the parameters of the encoder and decoder in each of the adversarial diffusion generation modules using the attention map obtained in step S105 through each of the contrast-enhanced attention guidance submodules. Specifically, optimize the parameters of the encoder and decoder in each of the adversarial diffusion generation modules based on the minimum distance between the attention map and the binary contrast-enhanced mask.

[0165] S107: Obtaining a denoising result corresponding to each adversarial diffusion generation module based on the estimated denoised image using the Bayesian posterior sampling submodule of each adversarial diffusion generation module in the conversion device. The denoising result corresponding to the adversarial diffusion generation module is a function of the model parameters of the encoder, decoder, and Bayesian posterior sampling submodule of the adversarial diffusion generation module.

[0166] S108: Parameter optimization is performed on the contrast perception discrimination submodule of each adversarial diffusion generation module in the conversion device using the denoising results obtained in step S107, for the encoder, decoder, and Bayesian posterior sampling submodule in the adversarial diffusion generation module to which it belongs. Specifically, the denoising results are substituted into a non-saturated adversarial loss function with a gradient penalty, and the parameters of the encoder, decoder, and Bayesian posterior sampling submodule in the adversarial diffusion generation module to which it belongs are optimized with the goal of obtaining the minimum value of the non-saturated adversarial loss function with a gradient penalty.

[0167] S109: Substitute the binary contrast enhancement mask into the loss function through the mask-guided feature alignment module in the parameter optimization device to optimize the parameters of the conversion device using the loss function, obtain the model parameters and pass them to all encoders, decoders and Bayesian posterior sampling submodules.

[0168] During parameter optimization, the coefficients are determined as: Based on ablation studies, gradient stability and detail preservation are balanced. This embodiment also performs data preprocessing when acquiring the dataset in step S101. For data preprocessing, the ratio of window width to window level is set according to clinical guidelines. Computed tomography angiography uses a vascular window of [600, 200] to enhance contrast visualization, while non-contrast computed tomography images use a mediastinal window of [400, 50] to distinguish soft tissue. Simultaneously, SimpleITK is used to implement rigid affine registration based on mutual information maximization.

[0169] S2: A Gaussian noise image is generated by a noise generating device using a random Gaussian noise iterative algorithm, and a contrast-free computed tomography image is acquired by an acquisition device.

[0170] S3: A Gaussian noise image is integrated into a contrast-free computed tomography image through a conversion device, and a computed tomography angiography is obtained by a fusion method of feature sampling and stepwise denoising.

[0171] In this embodiment, the iterative diffusion step size , the noise scheduling is determined by Definition, total number of diffusion steps For the mask-guided PatchNCE loss function, patch embeddings are compared using cosine similarity, temperature The converted set was trained for 50 epochs (approximately 120 hours) on four NVIDIA RTX4090 GPUs. The dataset was split strictly following the patient-level separation: 100 patients were used for training, 25 patients for validation, and 25 patients for testing, ensuring no data leakage. Inference maintained the same diffusion parameters as training. , in order to preserve the random dynamics.

[0172] The following describes in detail the technical effects of the mask-guided, contrast-free angiography generation system in this embodiment. In this embodiment, the angiography generation system is compared with existing image-to-image conversion devices, including GAN-based frameworks such as Pix2Pix, CycleGAN, CTA-GAN, UNIT, MUNIT, and NICE-GAN, as well as the diffusion-based model SynDiff.

[0173] Quantitative evaluation results show that the angiography generation system outperforms these devices in multiple indicators. Specifically, the PSNR value of the angiography generation system is 24.74, which is significantly improved compared to Pix2Pix (21.28), CycleGAN (22.97), and other methods such as CTA-GAN (23.25) and UNIT (22.60). This shows that the angiography generation system performs well in restoring image details and fidelity, and the generated images are closer to the target images in quality. In terms of the SSIM indicator, the angiography generation system reached 86.43, showing better structural similarity than Pix2Pix (74.44) and CycleGAN (80.42), indicating that the generated images are not only highly similar to the target images in terms of overall structure, but also perform better in terms of texture and detail preservation. In terms of the NMAE metric, the angiography generation system achieved a value of 0.0317, lower than Pix2Pix (0.0453), CycleGAN (0.0382), and other comparison methods, indicating superior brightness and intensity alignment with the target image. This further demonstrates the strong performance of the angiography generation system in maintaining brightness consistency and reconstructing details. In terms of the MSSSIM metric, the angiography generation system achieved a value of 92.66, significantly higher than CycleGAN (76.52) and CTA-GAN (79.71). This demonstrates the angiography generation system's ability to preserve multi-scale structural features and provide higher image quality at multiple resolution levels. The improvement in MSSSIM highlights the angiography generation system's ability to effectively capture and maintain structural details, which are critical for accurate visual interpretation. In terms of the MI metric, the angiography generation system achieved a value of 1.325, surpassing SynDiff (1.243) and CycleGAN (1.237). This indicates that the proposed angiography generation system performs better in preserving the mutual information between the generated and target images, reflecting its superior ability to preserve important image features and information during the generation process.

[0174] These quantitative results clearly demonstrate the beneficial effects of the proposed angiography generation system, especially in terms of PSNR, SSIM, NMAE, MSSSIM, and MI, fully validating its superior performance in detail preservation and image consistency.

[0175] In summary, the mask-guided contrast-free angiography generation system disclosed in this embodiment, by providing a noise generation device, an acquisition device, a conversion device, and a parameter optimization device, can optimize the parameters of the conversion device using a loss function that emphasizes contrast-enhanced shadows, thereby improving the conversion device's ability to capture complex image details. The conversion device can then integrate a Gaussian noise image into the contrast-free CT image, and obtain a CT angiogram through a fusion of feature sampling and stepwise denoising, ensuring sufficient consistency and realism in the obtained CT angiogram. This integration of a loss function that emphasizes contrast-enhanced shadows into a diffusion process based on a Gaussian noise image generated by a random Gaussian noise iterative algorithm not only effectively ensures the conversion of the contrast-free CT image into a high-quality CT angiogram, but also stabilizes the parameter optimization process, improving the sharpness and fidelity of the generated vascular structure, thereby overcoming the technical shortcomings of existing RegGAN model architectures.

[0176] In the description of the embodiments of the present application, it should be noted that in the description of the present application, terms such as "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or component must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present application.

[0177] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "in the present embodiment", "specific example", or "some examples" means that the specific features, mechanisms, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0178] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A mask-guided contrast-free angiography generation system, characterized in that: include: The noise generating device is configured to generate a Gaussian noise image using a random Gaussian noise iterative algorithm; an acquisition device configured to acquire a non-contrast computed tomography image; a conversion device, electrically connected to both the noise generating device and the acquisition device, configured to integrate the Gaussian noise image into the contrast-free CT image and obtain a CT angiogram by a fusion method of feature sampling and stepwise denoising; a parameter optimization device electrically connected to the conversion device and configured to generate a binary contrast enhancement mask using the data set based on a subtraction strategy, substitute the binary contrast enhancement mask into a loss function for highlighting the vascular region affected by the contrast agent, optimize the parameters of the conversion device using the loss function, obtain model parameters, and transmit the obtained model parameters to the conversion device; The conversion device is a communication network structure composed of a plurality of adversarial diffusion generation modules connected in series, wherein the adversarial diffusion generation modules are configured to convert an input image into an output image by encoding and decoding feature sampling and Bayesian rule sampling and denoising; The adversarial diffusion generation module includes an encoder, a decoder, a Bayesian posterior sampling submodule, a contrast enhancement attention guidance submodule and a contrast perception discrimination submodule; The input end of the encoder in the first adversarial diffusion generation module is electrically connected to the output end of the noise generation device and the output end of the acquisition device, and the input ends of the encoders in the remaining adversarial diffusion generation modules are electrically connected to the output end of the Bayesian posterior sampling submodule in the previous adversarial diffusion generation module; In each of the adversarial diffusion generation modules, the input end of the decoder is electrically connected to the output end of the encoder, the input end of the Bayesian posterior sampling submodule is electrically connected to the output end of the decoder, the input end of the contrast perception discrimination submodule is electrically connected to the output end of the Bayesian posterior sampling submodule and the output end of the parameter optimization device, the input end of the contrast enhancement attention guidance submodule is electrically connected to the output end of the decoder and the output end of the parameter optimization device, and the input end of the encoder is also electrically connected to the output end of the contrast perception discrimination submodule and the output end of the contrast enhancement attention guidance submodule; The parameter optimization device comprises: an unsupervised mask generation module, the output of which is electrically connected to the inputs of all the contrast enhancement attention guidance submodules and the input of the contrast perception discrimination submodule, and is configured to generate a binary contrast enhancement mask using the dataset based on a subtraction strategy; a mask-guided feature alignment module, the input of which is electrically connected to the output of the unsupervised mask generation module and the output of the Bayesian posterior sampling submodule in the final adversarial diffusion generation module, and configured to substitute the binary contrast enhancement mask into the loss function to optimize the parameters of the conversion device using the loss function, obtain model parameters, and transmit them to all of the encoders, the decoders, and the Bayesian posterior sampling submodules; The calculation formula of the loss function is: Where, represents the loss function; λ SP ,λ TP ,λ attn and λ adv Both represent weight factors; ⊙ represents element-by-element multiplication; represents the target PatchNCE loss function; represents the source PatchNCE loss function; represents the attention loss function; represents the combined discriminator loss function; n represents the number of data in the data set; represents the anchor point complement determined on the image output by the conversion device during parameter optimization; represents a positive sample patch defined on the input image of the conversion device when performing parameter optimization; represents a negative sample patch defined on the input image of the conversion device during parameter optimization; F i represents the i-th attention map generated by the contrast-enhanced attention guidance submodule during parameter optimization; represents the denoising result obtained by the Bayesian posterior sampling submodule during parameter optimization; m represents the binary contrast enhancement mask; Represents the InfoNCE loss function; represents the non-saturating adversarial loss function with gradient penalty.

2. The mask-guided contrast-free angiography generation system according to claim 1, characterized in that: The iterative formula of the random Gaussian noise iterative algorithm is as follows: Where, T is a composite number, representing the number of diffusion steps initially planned; k is the fundamental factor of the composite number T; x t Represents the noise at the T-th step of iteration; γ t Represents the noise addition item at the t-th step of iteration; ∈ represents an arbitrarily small number; q(x t |x t-k ) represents the noise x t-k Under the condition of obtaining noise x t The conditional probability distribution of ; I represents the identity matrix; represents Gaussian distribution; Represents the minimum value of index progress; Represents the maximum value of the exponential progress.

3. The mask-guided contrast-free angiography generation system according to claim 1 or 2, characterized in that: The encoder is configured to perform a convolution operation on its input image to extract anatomical structure and texture information, incorporate temporal information into the extraction process, and output an encoded image; The decoder is configured to perform deconvolution denoising on the encoded image to obtain an estimated denoised image; The Bayesian posterior sampling submodule is configured to denoise the estimated denoised image using a Bayesian sampling rule to obtain a denoised result; The contrast enhancement attention guidance submodule is configured to, after the parameter optimization device is started, perform element-wise multiplication and ReLU activation on the estimated denoised image and the binary contrast enhancement mask in sequence to generate an attention map, and optimize the parameters of the encoder and decoder in the adversarial diffusion generation module in which it is located as a target to obtain the minimum value of the distance between the attention map and the binary contrast enhancement mask; The contrast perception discrimination submodule is configured to optimize the parameters of the encoder, decoder and Bayesian posterior sampling submodule in the adversarial diffusion generation module in which it is located, with the goal of obtaining the minimum value of the non-saturated adversarial loss function with gradient penalty after the parameter optimization device is started.

4. The mask-guided contrast-free angiography generation system according to claim 3, characterized in that: The calculation formula of the subtraction strategy is: m i =δ(CTA i -NCCT i ), Where, m i A binary contrast enhancement mask representing the i-th data in the dataset; CTA i A computed tomography angiogram representing the i-th data in the data set; NCCT i a contrast-free computed tomography image representing the i-th data in the dataset; δ represents the binarization process.

5. A method for generating contrast-free angiography based on mask guidance, characterized in that: The mask-guided contrast-free angiography generation system according to any one of claims 1 to 4 comprises the following steps: S1: generating a binary contrast enhancement mask using a data set by a parameter optimization device, and substituting the binary contrast enhancement mask into a loss function for highlighting a vascular region affected by a contrast agent to optimize parameters of a conversion device using the loss function, obtaining model parameters, and transmitting the obtained model parameters to the conversion device; S2: a noise generating device generates a Gaussian noise image using a random Gaussian noise iterative algorithm, and an acquisition device acquires a contrast-free computed tomography image; S3: The Gaussian noise image is integrated into the contrast-free CT image by the conversion device, and CT angiography is obtained by a fusion method of feature sampling and stepwise denoising.

6. The mask-guided contrast-free angiography generation method according to claim 5, characterized in that: The step S1 comprises the following steps: S101: Retrieving multiple sets of contrast-free computed tomography images and their corresponding computed tomography angiography images from a database to obtain the data set; S102: inputting all data in the data set into the parameter optimization device and the conversion device; S103: generating a binary contrast enhancement mask using the data set through an unsupervised mask generation module in the parameter optimization device; S104: Processing the data set by the encoder and decoder of each anti-diffusion generation module in the conversion device to obtain an estimated denoised image corresponding to each anti-diffusion generation module; S105: performing element-wise multiplication and ReLU activation on the estimated denoised image and the binary contrast enhancement mask in sequence through the contrast enhancement attention guidance submodule of each adversarial diffusion generation module in the conversion device to generate an attention map corresponding to each adversarial diffusion generation module; S106: Optimizing parameters of the encoder and decoder in the adversarial diffusion generation module using the attention map obtained in step S105 through each of the contrast enhancement attention guidance submodules; S107: Obtaining denoising results corresponding to the respective adversarial diffusion generation modules according to the estimated denoised image through the Bayesian posterior sampling submodules of the respective adversarial diffusion generation modules in the conversion device; S108: Optimizing parameters of the encoder, decoder, and Bayesian posterior sampling submodule in each of the anti-diffusion generation modules in the conversion device according to the denoising results obtained in step S107 by using the contrast perception discrimination submodule of each anti-diffusion generation module in the conversion device; S109: Substitute the binary contrast enhancement mask into the loss function through the mask-guided feature alignment module in the parameter optimization device to optimize the parameters of the conversion device using the loss function, obtain model parameters and transmit them to all the encoders, the decoders and the Bayesian posterior sampling sub-modules.

Images