Multi-domain perceptual contrast enhanced computed tomography image synthesis method and system and electronic equipment

Abstract

According to the multi-domain perceptual contrast enhanced computed tomography image synthesis method and system and the electronic equipment, efficient synthesis from non-contrast enhanced images to high-quality contrast enhanced images is successfully achieved by adopting the 3D auto-encoder and the de-noising diffusion probability model, the calculation process is optimized, and the image quality is improved. Due to the introduction of multi-domain perceptual regularization, fine-grained enhancement is carried out on the image at different frequency bands, and the detail expressive force and texture consistency of the synthesized image are remarkably improved. Besides, automation of an image synthesis process is realized through a deep learning framework, manual operation is reduced, processing time is shortened, and a rapid, accurate and reliable technical solution is provided for the field of medical image processing. The technical innovation provides a new path for the generation of the CECT image, and promotes the development of the medical imaging technology.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a multi-domain perception contrast-enhanced computed tomography (CECT) image synthesis method, system and electronic equipment for acute pancreatitis assessment. Background Art

[0002] Acute pancreatitis is a severe inflammatory disease characterized by edema, hemorrhage, and necrosis of pancreatic tissue. The disease may rapidly worsen, leading to multi-organ dysfunction and high mortality. Therefore, early and accurate diagnosis is essential for effective treatment.

[0003] Traditional diagnostic methods rely on contrast-enhanced computed tomography (CECT), an imaging technique that can provide high-resolution images of the pancreas and its surrounding structures. CECT uses a contrast agent containing iodine to enhance the contrast of the image, thereby more clearly showing the vascular and tissue structures of the pancreas. However, this technology has some limitations: (1) Side effects of contrast agents: The use of iodine contrast agents may cause allergic reactions or other side effects and is not suitable for some patients (such as those with kidney disease or those who are allergic to iodine). (2) Radiation exposure: CECT scans involve higher radiation doses, posing a potential risk to the patient's health. (3) Cost and accessibility: CECT equipment is expensive and not easily available in some areas or medical institutions.

[0004] To overcome these limitations, researchers have been exploring methods to synthesize CECT images using non-contrast enhanced computed tomography (NCCT) images. This method aims to generate CECT-like images from NCCT images through software algorithms, thereby avoiding the use of contrast agents and reducing radiation exposure. Existing image synthesis methods, including those based on generative adversarial networks (GANs) and diffusion models, have made some progress in the field of image synthesis, but there are still some challenges, such as unstable training, insufficient image quality, and the inability to accurately simulate specific features of CECT images. Summary of the invention

[0005] The purpose of this invention is to provide an improved CECT image synthesis method that can reduce the need for invasive examinations of patients while providing high-quality imaging results to support accurate diagnosis of acute pancreatitis. By using advanced image processing technology and machine learning algorithms, it aims to improve the accuracy and efficiency of diagnosis while reducing health risks and medical costs for patients.

[0006] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0007] A method for synthesizing multi-domain perception contrast-enhanced computed tomography images comprises the following steps:

[0008] S1. Obtain a pancreatitis medical image dataset through medical image data collection and preprocessing, and divide it into a training set, a validation set, and a test set;

[0009] S2, compress and reconstruct the image using a 3D autoencoder to obtain the latent representation;

[0010] S3, constructing a denoising diffusion probability model to generate the distribution of the latent representation;

[0011] S4, optimizing a specific target by encoding conditional codes to generate a synthetic image with high consistency with the actual CECT image;

[0012] S5. Design multi-domain-aware regularization to ensure the consistency of the generated synthetic images in different frequency spaces.

[0013] As a preferred solution of the multi-domain perception contrast enhanced computed tomography image synthesis method of the present invention, wherein: the step S1 specifically includes the following steps:

[0014] S101. Collect pancreatic NCCT-CECT paired data, which consist of 220 pairs of CT images. The CT images are acquired using a Siemens Definition AS 40 scanner and scanned using a voltage of 120 kV and a current of 303 mA. The voxel spacing of each NCCT and CECT is 0.78×0.78×5 mm. 3 , the image matrix size is 512 × 512, containing 19 to 81 slices;

[0015] S102, randomly dividing the paired CT images into a training set, a validation set and a test set, wherein the training set accounts for 80% and is used for model learning; the validation set accounts for 10% and is used for supervising and optimizing the model; and the test set accounts for 10% and is used for evaluating model performance;

[0016] S103. Standardize the data set, adjust its size to 128×128×16, and normalize it to the range of [0,255] to reduce data storage pressure and improve data practicality.

[0017] As a preferred solution of the multi-domain perception contrast enhanced computed tomography image synthesis method of the present invention, wherein: step S2 specifically includes the following steps:

[0018] S201, constructing a three-dimensional autoencoder to encode the voxel space of the CECT image into a latent representation;

[0019] S202, CECT image x i Through the encoder q Φ (z|x i ) is compressed into latent variables, and the latent representation z is obtained by sampling;

[0020] S203, using decoder p θ (z|x i ) reconstruct the latent representation z to generate a reconstructed CECT image, and optimize the latent representation through reconstruction consistency learning to make it close to the data distribution;

[0021] S204. Introduce a patch-based inverse regularization mechanism to enhance the recognition of details of the pancreatitis area and improve the local contour and texture of the generated image.

[0022] As a preferred solution of the multi-domain perception contrast enhanced computed tomography image synthesis method of the present invention, the implementation method of step S204 is:

[0023]

[0024] Among them, q Φ represents the encoder, p θ Represents a decoder, represents the discriminator, and Represent the real image and the reconstructed image respectively;

[0025] As a preferred solution of the multi-domain perception contrast enhanced computed tomography image synthesis method of the present invention, wherein: step S3 specifically includes the following steps:

[0026] S301, constructing a denoising diffusion probability model, converting the original data into noisy data by gradually introducing Gaussian noise, and achieving denoising by predicting the noise distribution;

[0027] S302, the denoising diffusion probability model gradually adds Gaussian noise to the original data by establishing a Markov chain to learn the distribution of the noise;

[0028] S303, encode the NCCT image into conditional embedding through the conditional extractor, and use q Φ The generated latent representation and the noise prediction network p θ Conduct joint optimization;

[0029] S304, reconstructing a high-quality CECT image through an inverse diffusion process, and decoding the inferred latent representation into an image output.

[0030] As a preferred solution of the multi-domain perceptual contrast enhanced computed tomography image synthesis method of the present invention, wherein: in step S3, the optimal paradigm of the denoising diffusion probability model for:

[0031]

[0032] Among them, x t represents the data sample after adding noise at time step t, ∈ represents the actual Gaussian noise added, ∈ θ (x t ,t) represents the prediction result of the model for the noise at time step t, and t and ∈ are random samples.

[0033] As a preferred solution of the multi-domain perception contrast enhanced computed tomography image synthesis method of the present invention, wherein: the step S4 specifically includes the following steps:

[0034] S401, by inheriting the encoder q from the encoder Φ , obtain the potential condition code for conditional generation and reduce redundant calculation;

[0035] S402, passing the potential conditional code zc as input to the 3D U-Net backbone network to guide the noise prediction model ∈ θ Perform denoising generation;

[0036] S403. During the conditional extraction process, avoid additional training of the DDPM model to improve generation efficiency.

[0037] As a preferred solution of the multi-domain perception contrast enhanced computed tomography image synthesis method of the present invention, the implementation method of step S403 is:

[0038]

[0039] where ∈ represents Gaussian noise sampled from a standard normal distribution N(0,1), ∈ θ (z t ,t,z c ) represents the model’s prediction of the noise at time step t, z t is the representation of data x at time step t, z c is from the conditional distribution q φ (z|x i ) represents the potential representation of multi-domain images.

[0040] As a preferred solution of the multi-domain perception contrast enhanced computed tomography image synthesis method of the present invention, wherein: step S5 specifically includes the following steps:

[0041] S501, mapping the synthesized image to different frequency spaces through a preset mask to enhance the consistency of the generated image in each frequency domain;

[0042] S502, optimizing the generated image through multi-domain perception regularization and combining the spatial information of the dual domains to ensure the balance between the local details and the global features of the reconstructed image;

[0043] S503. A balance is maintained between consistency and multi-domain awareness through a dynamic trade-off factor λ, thereby achieving high-quality generation and detail optimization of the synthetic image.

[0044] As a preferred solution of the multi-domain perception contrast enhanced computed tomography image synthesis method of the present invention, the implementation method of step S501 is:

[0045]

[0046] in, Represents the conditional probability distribution p θ (x i All reconstructed images sampled from |z) The expected value of and They represent the real image and the reconstructed image respectively. The subscript h represents the image after being filtered by a high-pass filter, and the subscript l represents the image after being filtered by a low-pass filter.

[0047] As a preferred solution of the multi-domain perceptual contrast-enhanced computed tomography image synthesis method of the present invention, wherein: the complete optimization target Q in step S503 satisfies:

[0048]

[0049] in, represents the overall optimization objective, including reconstruction loss and multi-domain aware regularization loss; Reconstruction loss measures the difference between the model reconstructed image and the original image; λ represents the trade-off coefficient, which controls the balance between the reconstruction loss and the multi-domain-aware regularization loss, and its value is 1.5.

[0050] As a preferred solution of the multi-domain perceptual contrast-enhanced computed tomography image synthesis method described in the present invention, wherein: the step S3 also includes a conditional extractor for encoding a non-contrast-enhanced computed tomography (NCCT) image into a conditional embedding to guide the DDPM to generate a specific CECT image.

[0051] To solve the above technical problem, according to another aspect of the present invention, the present invention provides the following technical solution:

[0052] A multi-domain perception contrast-enhanced computed tomography image synthesis system, comprising:

[0053] The data acquisition and preprocessing module acquires the pancreatitis medical image dataset through medical image data acquisition and preprocessing, and divides it into a training set, a validation set, and a test set;

[0054] The image compression and reconstruction module uses a 3D autoencoder to compress and reconstruct the image to obtain a latent representation;

[0055] A denoising diffusion generation module constructs a denoising diffusion probability model to generate the distribution of the potential representation;

[0056] The conditional generation module optimizes specific targets by encoding conditional codes to generate synthetic images with high consistency with actual CECT images;

[0057] Multi-domain-aware regularization module,Multi-domain-aware regularization is designed to ensure the consistency of the generated synthetic images in different frequency spaces.

[0058] To solve the above technical problem, according to another aspect of the present invention, the present invention provides the following technical solution:

[0059] An electronic device comprises: a processor, a memory, an internal bus, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the multi-domain perception contrast-enhanced computer tomography image synthesis method.

[0060] A readable storage medium stores a program or instruction, which, when executed by a processor, implements the multi-domain perception contrast-enhanced computed tomography image synthesis method.

[0061] The beneficial effects of the present invention are as follows:

[0062] The present invention proposes a method, system and electronic device for synthesizing multi-domain-aware contrast-enhanced computed tomography images. By adopting a 3D autoencoder and a denoising diffusion probability model, it successfully realizes the efficient synthesis from non-contrast-enhanced to high-quality contrast-enhanced images, optimizes the calculation process and improves the image quality. The introduction of multi-domain-aware regularization significantly improves the detail expression and texture consistency of the synthesized image by performing fine-grained enhancement of the image in different frequency bands. In addition, the present invention realizes the automation of the image synthesis process through a deep learning framework, reduces manual operations, shortens processing time, and provides a fast, accurate and reliable technical solution for the field of medical image processing. These technological innovations provide a new path for the generation of CECT images and promote the development of medical imaging technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0064] Figure 1 is a flow chart of the steps of a method for synthesizing multi-domain-aware contrast-enhanced computed tomography images;

[0065] Figure 2 It is a schematic diagram of the framework structure of a multi-domain perception contrast-enhanced computed tomography image synthesis method;

[0066] Figure 3 It is a block diagram of the structure of the multi-domain perception contrast-enhanced computed tomography image synthesis system;

[0067] Figure 4 It is a structural block diagram of an electronic device according to an embodiment of the present application.

[0068] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0069] The following will be described clearly and completely in conjunction with the technical solutions in the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] The present invention proposes a multi-domain perception contrast-enhanced computed tomography (CECT) image synthesis method, which can solve the patient health risk problem caused by the use of iodine contrast agents in traditional CECT scans. Specifically, through the multi-domain perception CECT image synthesis method, an image that is highly consistent with the actual CECT image can be synthesized from a non-contrast enhanced computed tomography (NCCT) image, achieving the purpose of disease diagnosis without the use of iodine contrast agents, while reducing the radiation dose to the patient. In addition, this method significantly improves the efficiency and accuracy of image synthesis through an automated image processing process, providing an innovative technical support for medical imaging diagnosis.

[0071] like Figure 1 As shown, an embodiment of the present invention provides a multi-domain perception contrast enhanced computer tomography image synthesis method, comprising the following steps:

[0072] S1. Through medical image data collection and preprocessing, a pancreatitis medical image dataset is obtained and divided into a training set, a validation set, and a test set.

[0073] This step is a step of collecting and preprocessing medical image data and dividing the data set. An optional method of preprocessing medical image data, collecting image data, and dividing the data set may include the following steps:

[0074] S101. Collect pancreatic NCCT-CECT paired data, which consist of 220 pairs of CT images. The CT images are acquired using a Siemens Definition AS 40 scanner and scanned using a voltage of 120 kV and a current of 303 mA. The voxel spacing of each NCCT and CECT is 0.78×0.78×5 mm. 3 , the image matrix size is 512 × 512, containing 19 to 81 slices;

[0075] S102, randomly dividing the paired CT images into a training set, a validation set and a test set, wherein the training set accounts for 80% and is used for model learning; the validation set accounts for 10% and is used for supervising and optimizing the model; and the test set accounts for 10% and is used for evaluating model performance;

[0076] S103. Standardize the data set, adjust its size to 128×128×16, and normalize it to the range of [0,255] to reduce data storage pressure and improve data practicality.

[0077] S2, compress and reconstruct the image using a 3D autoencoder to obtain the latent representation;

[0078] An alternative framework could be Figure 2 As shown, in the first training stage, an autoencoder constraint builds a latent representation of an image. Multi-domain perception captures the consistency of different frequency spaces. Subsequently, DDPM learns the distribution of the later representation. We inherit the encoder of the autoencoder as a conditional extractor, which is considered appropriate. Then, we introduce the fusion of multiple domains in the conditional embedding to obtain sufficient expressions. During inference, DDPM generates a specific synthetic CECT image by sampling the generated NCCTimage and the conditional code, which is reconstructed into the image space by the decoder of the autoencoder. It includes the following steps:

[0079] S201, constructing a three-dimensional autoencoder to encode the voxel space of the CECT image into a latent representation;

[0080] S202, CECT image x i Through the encoder q Φ(z|x i ) is compressed into latent variables, and the latent representation z is obtained by sampling;

[0081] S203, using decoder p θ (z|x i ) reconstruct the latent representation z to generate a reconstructed CECT image, and optimize the latent representation through reconstruction consistency learning to make it close to the data distribution;

[0082] S204, introducing a patch-based inverse normalization mechanism to enhance the recognition of details of the pancreatitis region and improve the local contour and texture of the generated image. The implementation method of step S204 is:

[0083]

[0084] Among them, q Φ represents the encoder, p θ Represents a decoder, represents the discriminator, and Represent the real image and the reconstructed image respectively;

[0085] S3, constructing a denoising diffusion probability model (DDPM) to generate the distribution of the latent representation;

[0086] The step S3 specifically includes the following steps:

[0087] S301, constructing a denoising diffusion probability model, converting the original data into noisy data by gradually introducing Gaussian noise, and achieving denoising by predicting the noise distribution;

[0088] S302, the denoising diffusion probability model gradually adds Gaussian noise to the original data by establishing a Markov chain to learn the distribution of the noise;

[0089] S303, encode the NCCT image into conditional embedding through the conditional extractor, and use q Φ The generated latent representation and the noise prediction network p θ Conduct joint optimization;

[0090] S304, reconstructing a high-quality CECT image through an inverse diffusion process, and decoding the inferred latent representation into an image output.

[0091] In step S3, the optimal paradigm of the denoising diffusion probability model is for:

[0092]

[0093] Among them, x trepresents the data sample after adding noise at time step t, ∈ represents the actual Gaussian noise added, ∈ θ (x t ,t) represents the prediction result of the model for the noise at time step t, and t and ∈ are random samples.

[0094] The step S3 also includes a conditional extractor for encoding a non-contrast enhanced computed tomography (NCCT) image into a conditional embedding to guide the DDPM to generate a specific CECT image.

[0095] S4, optimizing a specific target by encoding conditional codes to generate a synthetic image with high consistency with the actual CECT image;

[0096] The step S4 specifically includes the following steps:

[0097] S401, by inheriting the encoder q from the encoder Φ , obtain the potential condition code for conditional generation and reduce redundant calculation;

[0098] S402, passing the potential conditional code zc as input to the 3D U-Net backbone network to guide the noise prediction model ∈ θ Perform denoising generation;

[0099] S403: During the condition extraction process, avoid additional training of the DDPM model to improve generation efficiency. The implementation method of step S403 is as follows:

[0100]

[0101] where ∈ represents Gaussian noise sampled from a standard normal distribution N(0,1), ∈ θ (z t ,t,z c ) represents the model’s prediction of the noise at time step t, z t is the representation of data x at time step t, z c is from the conditional distribution q φ (z|x i ) represents the potential representation of multi-domain images.

[0102] S5. Design multi-domain-aware regularization to ensure the consistency of the generated synthetic images in different frequency spaces.

[0103] The step S5 specifically includes the following steps:

[0104] S501, mapping the synthesized image to different frequency spaces through a preset mask to enhance the consistency of the generated image in each frequency domain, the implementation method is:

[0105]

[0106] in, Represents the conditional probability distribution p θ (x i All reconstructed images sampled from |z) The expected value of and They represent the real image and the reconstructed image respectively. The subscript h represents the image after being filtered by a high-pass filter, and the subscript l represents the image after being filtered by a low-pass filter.

[0107] S502, optimizing the generated image through multi-domain perception regularization and combining the spatial information of the dual domains to ensure the balance between the local details and the global features of the reconstructed image;

[0108] S503. A balance is maintained between consistency and multi-domain awareness through a dynamic trade-off factor λ, thereby achieving high-quality generation and detail optimization of the synthetic image.

[0109] As a preferred solution of the multi-domain perceptual contrast-enhanced computed tomography image synthesis method of the present invention, wherein:

[0110] As a preferred solution of the multi-domain perceptual contrast-enhanced computed tomography image synthesis method of the present invention, wherein: the complete optimization target Q in step S503 satisfies:

[0111]

[0112] in, represents the overall optimization objective, including reconstruction loss and multi-domain aware regularization loss; Reconstruction loss measures the difference between the model reconstructed image and the original image; λ represents the trade-off coefficient, which controls the balance between the reconstruction loss and the multi-domain-aware regularization loss, and its value is 1.5.

[0113] like Figure 3 As shown, another embodiment of the present invention provides a multi-domain perception contrast enhanced computed tomography (CECT) image synthesis system, comprising:

[0114] The data acquisition and preprocessing module acquires the pancreatitis medical image dataset through medical image data acquisition and preprocessing, and divides it into a training set, a validation set, and a test set;

[0115] The image compression and reconstruction module uses a 3D autoencoder to compress and reconstruct the image to obtain a latent representation;

[0116] A denoising diffusion generation module constructs a denoising diffusion probability model to generate the distribution of the potential representation;

[0117] The conditional generation module optimizes specific targets by encoding conditional codes to generate synthetic images with high consistency with actual CECT images;

[0118] Multi-domain-aware regularization module,Multi-domain-aware regularization is designed to ensure the consistency of the generated synthetic images in different frequency spaces.

[0119] Optionally, the data acquisition and preprocessing module includes:

[0120] In the data acquisition submodule, the system collects paired NCCT (non-contrast enhanced CT) and CECT (contrast enhanced CT) image data from multiple clinical data sources. The data is acquired through a standard CT scanner such as the Siemens Definition AS 40 scanner, and the CT scanning parameters are set to 120kV voltage and 303mA current to ensure scanning quality.

[0121] In the data preprocessing submodule, the system standardizes the image data, resizes it to 128×128×16, and normalizes the pixel values ​​to the range of [0,255] so that subsequent model training can be processed uniformly.

[0122] The data partitioning submodule divides the collected data set into a training set (80%), a validation set (10%), and a test set (10%) according to a certain ratio to ensure the reliability and generalization ability of the model.

[0123] Optionally, the image compression and reconstruction module includes:

[0124] Encoding submodule: By constructing a three-dimensional autoencoder, the high-dimensional spatial information of the CECT image is encoded into a low-dimensional latent representation. The encoding process extracts the global and local features of the image, reduces redundant data and compresses the image information.

[0125] Reconstruction submodule: The latent representation is reconstructed through the decoder to generate a reconstructed image with a high similarity to the original CECT image. This process ensures the quality of the reconstructed image by learning the feature distribution of the CECT image.

[0126] Optimization submodule: In order to improve the reconstruction accuracy, the system adopts a consistency learning mechanism. By optimizing the potential representation to match the real data distribution and introducing a patch-based inverse regularization mechanism, the system can enhance the recognition of pancreatic regional details and ensure that the reconstructed image has high consistency in both details and global information.

[0127] Optionally, the denoising diffusion generation module includes:

[0128] Noise introduction submodule: During the training process of the diffusion model, Gaussian noise is gradually introduced into the original image to convert it into noisy data at different stages.

[0129] Noise prediction submodule: Using the Markov chain model, the system gradually predicts the noise distribution at different stages and learns how to denoise at each step, gradually approaching a clean image.

[0130] Conditional encoding submodule: Through the conditional extractor, the NCCT image is encoded into a conditional embedding and combined with the latent variables in the diffusion process to guide the system to generate CECT images with specific medical features. This process generates latent representations through joint optimization, ensuring that the synthesized image reflects the real image characteristics in both global structure and details.

[0131] Optionally, the condition generation module includes:

[0132] Conditional coding submodule: This submodule generates latent conditional codes (zc) by inheriting the encoder part from the encoder, which is used to guide the subsequent generation process and reduce redundant calculations.

[0133] Noise removal submodule: Based on the latent conditional code, the noise in the image is gradually removed through the 3D U-Net network structure to generate a clean and high-quality CECT image.

[0134] Efficiency optimization submodule: To improve generation efficiency, the system avoids additional training of the DDPM model and speeds up the image generation process by reducing redundant calculations. This method improves the operating efficiency of the overall model, allowing high-quality images to be generated quickly in clinical applications.

[0135] Optionally, the multi-domain aware regularization module includes:

[0136] Frequency domain mapping submodule: Through the preset mask, the generated CECT image is mapped to different frequency spaces to enhance its multi-domain information perception ability. This mapping process can ensure that the image maintains consistency in high-frequency and low-frequency information.

[0137] Detail optimization submodule: Combines the spatial information of the dual domains to optimize the details of the generated image, ensuring its balance between local structure and global features, and improving the clarity and contrast of the image.

[0138] Trade-off control submodule: The dynamic trade-off factor λ is used to regulate between consistency and multi-domain awareness to achieve the optimal fusion of information from different domains, thereby achieving the ideal effect in terms of details and overall perception. Finally, by optimizing the objective function Q, the system can generate high-quality synthetic images, and its optimization objective Q satisfies:

[0139]

[0140] in, represents the overall optimization objective, including reconstruction loss and multi-domain aware regularization loss; Reconstruction loss measures the difference between the model reconstructed image and the original image; λ represents the trade-off coefficient, which controls the balance between the reconstruction loss and the multi-domain-aware regularization loss, and its value is 1.5.

[0141] The present invention provides a method for synthesizing multi-domain-aware contrast-enhanced computed tomography images, which has significant technical innovation and practicality. By adopting a 3D autoencoder and a denoising diffusion probability model (DDPM), the method effectively realizes the automated synthesis from non-contrast-enhanced to high-quality contrast-enhanced images. The multi-domain-aware regularization technology enhances image details and texture consistency in different frequency bands, significantly improving the quality of the synthesized image. In addition, the present invention optimizes computational efficiency, reduces manual intervention and processing time, and provides a fast and accurate technical solution for medical image processing, which has important clinical application value.

[0142] The medical image synthesis system in the embodiment of the present invention may be a device, or may be a chip, an integrated circuit, etc. The medical image synthesis system may be an Android, iOS or other operating system, which is not specifically limited in the embodiment of the present application.

[0143] like Figure 4 As shown, another embodiment of the present invention provides an electronic device, the structural block diagram of which is as follows Figure 4 As shown, it includes a processor, a memory, an internal bus, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, each process of the above-mentioned image synthesis method embodiment is implemented. In addition, it may also include other required hardware such as a network interface, etc.

[0144] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the above-mentioned medical synthesis method embodiment is implemented.

[0145] The electronic device may be a computer chip or a physical implementation, or a product with a certain function. A typical implementation device is a computer, such as a personal computer, a smart phone, a tablet computer, a wearable device, or any combination of devices. The processor is a processor of an electronic device, and the readable storage medium includes a computer readable storage medium, such as a computer read-only memory, a random access memory, an optical disk, etc.

[0146] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for synthesizing multi-domain perceptual contrast-enhanced computed tomography images, characterized in that: The steps include: S1. Obtain a pancreatitis medical image dataset through medical image data collection and preprocessing, and divide it into a training set, a validation set, and a test set; S2, compress and reconstruct the image using a 3D autoencoder to obtain the latent representation; S3, constructing a denoising diffusion probability model to generate the distribution of the latent representation; S4, optimizing a specific target by encoding conditional codes to generate a synthetic image with high consistency with the actual CECT image; S5. Design multi-domain-aware regularization to ensure the consistency of the generated synthetic images in different frequency spaces.

2. The method for synthesizing multi-domain perceptual contrast-enhanced computed tomography images according to claim 1, characterized in that: The step S1 specifically includes the following steps: S101, collect pancreatic NCCT-CECT paired data; S102, randomly dividing the paired data into a training set, a validation set and a test set; S103, standardize the data set.

3. The method for synthesizing multi-domain perceptual contrast-enhanced computed tomography images according to claim 1, characterized in that: The step S2 specifically includes the following steps: S201, constructing a three-dimensional autoencoder to encode the voxel space of the CECT image into a latent representation; S202, CECT image x i Through the encoder q Φ (z|x i ) is compressed into latent variables, and the latent representation z is obtained by sampling; S203, using decoder p θ (z|x i ) reconstruct the latent representation z to generate a reconstructed CECT image, and optimize the latent representation through reconstruction consistency learning to make it close to the data distribution; S204. Introduce a patch-based inverse regularization mechanism to enhance the recognition of details of the pancreatitis area and improve the local contour and texture of the generated image.

4. The method for synthesizing multi-domain perceptual contrast-enhanced computed tomography images according to claim 1, characterized in that: The step S3 specifically includes the following steps: S301, constructing a denoising diffusion probability model, converting the original data into noisy data by gradually introducing Gaussian noise, and achieving denoising by predicting the noise distribution; S302, the denoising diffusion probability model gradually adds Gaussian noise to the original data by establishing a Markov chain to learn the distribution of the noise; S303, encode the NCCT image into conditional embedding through the conditional extractor, and use q Φ The generated latent representation and the noise prediction network p θ Conduct joint optimization; S304, reconstructing a high-quality CECT image through an inverse diffusion process, and decoding the inferred latent representation into an image output.

5. The method for synthesizing multi-domain perceptual contrast-enhanced computed tomography images according to claim 1, characterized in that: The step S4 specifically includes the following steps: S401, by inheriting the encoder q from the encoder Φ , obtain the potential condition code for conditional generation and reduce redundant calculation; S402, passing the potential conditional code zc as input to the 3D U-Net backbone network to guide the noise prediction model ∈ θ Perform denoising generation; S403. During the conditional extraction process, avoid additional training of the DDPM model to improve generation efficiency.

6. The method for synthesizing multi-domain perceptual contrast-enhanced computed tomography images according to claim 1, characterized in that: The step S5 specifically includes the following steps: S501, mapping the synthesized image to different frequency spaces through a preset mask to enhance the consistency of the generated image in each frequency domain; S502, optimizing the generated image through multi-domain perception regularization and combining the spatial information of the dual domains to ensure the balance between the local details and the global features of the reconstructed image; S503. A balance is maintained between consistency and multi-domain awareness through a dynamic trade-off factor λ, thereby achieving high-quality generation and detail optimization of the synthetic image.

7. The method for synthesizing multi-domain perceptual contrast-enhanced computed tomography images according to claim 1, characterized in that: The step S3 also includes a conditional extractor for encoding the non-contrast enhanced computed tomography image into a conditional embedding to guide the DDPM to generate a specific CECT image.

8. A multi-domain perceptual contrast-enhanced computed tomography image synthesis system, characterized in that: include: The data acquisition and preprocessing module acquires the pancreatitis medical image dataset through medical image data acquisition and preprocessing, and divides it into a training set, a validation set, and a test set; The image compression and reconstruction module uses a 3D autoencoder to compress and reconstruct the image to obtain a latent representation; A denoising diffusion generation module constructs a denoising diffusion probability model to generate the distribution of the potential representation; The conditional generation module optimizes specific targets by encoding conditional codes to generate synthetic images with high consistency with actual CECT images; Multi-domain-aware regularization module,Multi-domain-aware regularization is designed to ensure the consistency of the generated synthetic images in different frequency spaces.

9. An electronic device, characterized in that: include: A processor, a memory, an internal bus, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the multi-domain perception contrast-enhanced computed tomography image synthesis method described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, which, when executed by a processor, implements the multi-domain perception contrast-enhanced computed tomography image synthesis method described in any one of claims 1 to 7.

Images