Search

CN-121982162-A - PET image generation method, system, device and storage medium based on 3D wavelet transformation Brownian bridge diffusion

CN121982162ACN 121982162 ACN121982162 ACN 121982162ACN-121982162-A

Abstract

The invention relates to the technical field of hospital image analysis, and provides a PET image generation method, a system, a device and a storage medium based on 3D wavelet transformation Brownian bridge diffusion, which solve the technical problems of how to improve the accuracy and reliability of PET images in the existing nasopharyngeal carcinoma medical image analysis process. The invention is suitable for nasopharyngeal carcinoma or other tumors, or other medical image analysis processes.

Inventors

  • HE YANPING
  • ZHU JUN
  • YANG WENQI
  • YANG YIFAN
  • LIU BINGSHUO
  • Pu Yanbo
  • WANG CHANGMIAO
  • LI YONGHONG
  • YAN YAN
  • LI YOU

Assignees

  • 哈尔滨工业大学（威海）

Dates

Publication Date
20260505
Application Date
20260407

Claims (9)

  1. 1. The PET image generation method based on 3D wavelet transformation Brownian bridge diffusion is characterized by comprising the following steps: step (1), preprocessing the MRI image data and the PET image data which are respectively acquired to obtain a preprocessed MRI three-dimensional image And pre-processed PET three-dimensional image ; Step (2), preprocessing the MRI three-dimensional image by adopting a 3D Haar wavelet transform algorithm Decomposition into low frequency components And a high frequency component The PET three-dimensional image after pretreatment Decomposition into low frequency components And a high frequency component ; Step (3) of combining the low frequency components Low frequency component Respectively defined as low frequency components Low frequency component Will high frequency component High frequency component Respectively defined as high frequency components High frequency component ; Will low frequency component As a bridging start point, a low frequency component As bridging end point, performing low-frequency Brownian bridge diffusion treatment to obtain low-frequency bridge branch ; Calculating high frequency residual : High frequency component As bridging starting point, high frequency residual error As bridging end point, high-frequency Brownian bridge diffusion treatment is carried out to obtain residual information Computing high frequency bridge branches : ; Step (4), branching the low-frequency bridge by adopting an inverse wavelet transformation algorithm And high frequency bridge branching Fusion to obtain final three-dimensional synthetic PET image 。
  2. 2. The PET image generation method based on 3D wavelet transform brongridge diffusion according to claim 1, wherein the preprocessing process of step (1) is: Firstly, registering MRI image data and PET image data; secondly, resampling the MRI image data and the PET image data according to the consistent resolution; Finally, carrying out normalization processing on the MRI image data and the PET image data after resampling processing to obtain an MRI three-dimensional image after preprocessing And pre-processed PET three-dimensional image 。
  3. 3. The method for generating a PET image based on 3D wavelet transform brongridge diffusion according to claim 2, wherein the registration process registers MRI image data with PET image data with a CT image as a registration reference, the CT image being obtained while acquiring PET image data.
  4. 4. A PET image generation method based on 3D wavelet transform brongbridge diffusion according to claim 2 or 3, wherein the diffusion process is optimized with a joint loss function, the joint loss function being: ; In the formula, The weight is represented by a weight that, The weight is represented by a weight that, The weight is represented by a weight that, For the low frequency branch loss, For the high-frequency branch loss, For the reconstruction of the loss of the image domain, For enhancing the high frequency boundary and texture expression, For constraining final three-dimensional synthetic PET images And (3) consistency with the PET image data in the step (1) in the image space.
  5. 5. The PET image generation system based on the 3D wavelet transformation Brownian bridge diffusion is characterized by comprising a preprocessing module, a data decomposition module, a double-bridge diffusion generation module and a joint reconstruction optimization module; The preprocessing module is configured to preprocess the MRI image data and the PET image data which are respectively acquired to obtain an MRI three-dimensional image after preprocessing And pre-processed PET three-dimensional image ; The data decomposition module is configured to apply a 3D Haar wavelet transform algorithm to the preprocessed MRI three-dimensional image Decomposition into low frequency components And a high frequency component The PET three-dimensional image after pretreatment Decomposition into low frequency components And a high frequency component ; The double-bridge diffusion generation module is configured to generate a low-frequency component Low frequency component Respectively defined as low frequency components Low frequency component Will high frequency component High frequency component Respectively defined as high frequency components High frequency component ; Will low frequency component As a bridging start point, a low frequency component As bridging end point, performing low-frequency Brownian bridge diffusion treatment to obtain low-frequency bridge branch ; Calculating high frequency residual : High frequency component As bridging starting point, high frequency residual error As bridging end point, high-frequency Brownian bridge diffusion treatment is carried out to obtain residual information Computing high frequency bridge branches : ; The joint reconstruction optimization module is configured to branch the low frequency bridge by adopting an inverse wavelet transform algorithm And high frequency bridge branching Fusion to obtain final three-dimensional synthetic PET image 。
  6. 6. The 3D wavelet transform brongn-spread based PET image generation system of claim 5, wherein the preprocessing module performs preprocessing by: Firstly, registering MRI image data and PET image data; secondly, resampling the MRI image data and the PET image data according to the consistent resolution; Finally, carrying out normalization processing on the MRI image data and the PET image data after resampling processing to obtain an MRI three-dimensional image after preprocessing And pre-processed PET three-dimensional image 。
  7. 7. The 3D wavelet transform brongn-spread based PET image generation system of claim 6, wherein the registration process registers MRI image data with PET image data using CT images as registration references, the CT images being acquired while acquiring PET image data.
  8. 8. A storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method according to any of claims 1-4.
  9. 9. A PET image generating device based on 3D wavelet transformation Brownian bridge diffusion, which is characterized by comprising a memory and a processor; the memory is used for storing programs; the processor being adapted to execute the program for carrying out the steps of the method according to any one of claims 1-4.

Description

PET image generation method, system, device and storage medium based on 3D wavelet transformation Brownian bridge diffusion Technical Field The invention relates to the technical field of hospital image analysis, in particular to a PET image generation method, system, device and storage medium based on 3D wavelet transformation Brownian bridge diffusion. Background Nasopharyngeal carcinoma is a malignant tumor of head and neck with high incidence of the apparent region, especially in areas such as south China, southeast Asia, etc. Because of the complex anatomy structures such as the skull base bone, the cranial nerve duct, the parapharyngeal gap, the large blood vessel and the like, the tumor is often spread in a stepwise manner, so that the accurate evaluation of the affected range becomes a key premise for making a treatment scheme, drawing a radiotherapy target area and judging prognosis. Currently, prognostic prediction tasks rely primarily on Magnetic Resonance Imaging (MRI) in combination with patient clinical indicators for comprehensive assessment. Positron Emission Tomography (PET) can provide metabolism and function information, and then the PET image is combined with the MRI image, so that the method has a remarkable effect on prognosis prediction of nasopharyngeal carcinoma. However, acquisition and fusion of PET images still face a number of limitations. On the one hand, the injection of radioisotopes is required in the PET imaging process, which can place an ionizing radiation burden on the patient, and can present a potential long-term risk to nasopharyngeal carcinoma patients who need multiple follow-up evaluations, and on the other hand, the PET image acquisition cost is high. In addition, PET images have problems such as overcomplete, blurring of boundaries, distortion of boundaries, drift of details, loss of details, local artifacts, etc., such as the lack of good recurrence of details such as lymph nodes. Therefore, how to improve the accuracy and reliability of PET images is a technical problem to be solved by those skilled in the art. Disclosure of Invention The application provides a PET image generation method, a system, a device and a storage medium based on 3D wavelet transformation Brownian bridge diffusion, which aim at solving the technical problems of how to improve the accuracy and the reliability of a PET image in the existing nasopharyngeal carcinoma medical image analysis process. In a first aspect of the present application, there is provided a PET image generation method based on 3D wavelet transform brongn bridge diffusion, comprising the steps of: step (1), preprocessing the MRI image data and the PET image data which are respectively acquired to obtain a preprocessed MRI three-dimensional image And pre-processed PET three-dimensional image; Step (2), preprocessing the MRI three-dimensional image by adopting a 3D Haar wavelet transform algorithmDecomposition into low frequency componentsAnd a high frequency componentThe PET three-dimensional image after pretreatmentDecomposition into low frequency componentsAnd a high frequency component; Step (3) of combining the low frequency componentsLow frequency componentRespectively defined as low frequency componentsLow frequency componentWill high frequency componentHigh frequency componentRespectively defined as high frequency componentsHigh frequency component; Will low frequency componentAs a bridging start point, a low frequency componentAs bridging end point, performing low-frequency Brownian bridge diffusion treatment to obtain low-frequency bridge branch; Calculating high frequency residual:High frequency componentAs bridging starting point, high frequency residual errorAs bridging end point, high-frequency Brownian bridge diffusion treatment is carried out to obtain residual informationComputing high frequency bridge branches:; Step (4), branching the low-frequency bridge by adopting an inverse wavelet transformation algorithmAnd high frequency bridge branchingFusion to obtain final three-dimensional synthetic PET image。 Preferably, the pretreatment process of step (1) is: Firstly, registering MRI image data and PET image data; secondly, resampling the MRI image data and the PET image data according to the consistent resolution; Finally, carrying out normalization processing on the MRI image data and the PET image data after resampling processing to obtain an MRI three-dimensional image after preprocessing And pre-processed PET three-dimensional image。 Preferably, the registration process registers the MRI image data with the PET image data using the CT image obtained while the PET image data is acquired as a registration reference. Preferably, the diffusion process is optimized using a joint loss function, which is: ; In the formula, The weight is represented by a weight that,The weight is represented by a weight that,The weight is represented by a weight that,For the low frequency branch loss,For the high-frequency b