Search

CN-122023596-A - Single-view projection rapid fluorescence tomography reconstruction system based on dual-path condition guided diffusion model

CN122023596ACN 122023596 ACN122023596 ACN 122023596ACN-122023596-A

Abstract

The invention provides a single-view projection rapid fluorescence fault reconstruction system based on a dual-path condition-guided diffusion model, which comprises a view acquisition module, a model construction module and a model optimization module, wherein the view acquisition module is used for generating an original data set consisting of a fluorescence three-dimensional distribution image and a single-view fluorescence projection measurement image and acquiring single-view fluorescence projection data in a real experiment, the model construction module is used for defining a condition model frame, a diffusion model frame and a dual-path condition-guided mechanism to obtain an initial single-view projection reconstruction model, and the model optimization module is used for training and updating the initial reconstruction model by utilizing the original data set and setting and optimizing super parameters to obtain the single-view projection reconstruction model, and then the single-view projection reconstruction model is verified and evaluated by a test set. The method solves the problem of the exacerbation of the pathological condition caused by the extremely sparse single-view data and the overlapping of the depth information, and can realize the rapid dynamic living body three-dimensional imaging of target monitoring target molecules and high-quality reconstruction.

Inventors

  • ZHANG GUANGLEI
  • Cai Ruxin

Assignees

  • 北京航空航天大学

Dates

Publication Date
20260512
Application Date
20260202

Claims (6)

  1. 1. A dual-path condition guided diffusion model-based single-view projection rapid fluorescence tomographic reconstruction system, comprising: The view acquisition module is used for acquiring an original data set consisting of a fluorescence three-dimensional distribution image and a single-view fluorescence projection measurement image and acquiring single-view fluorescence projection data in a real experiment; The model construction module is used for defining a conditional model frame for extracting conditional features from boundary fluorescence projection measurement, a diffusion model frame for generating fluorescence three-dimensional distribution by iterative denoising and a dual-path conditional guiding mechanism for connecting the two model frames to obtain an initial single-view projection reconstruction model; The model optimization module is used for training and updating and super-parameter setting and optimizing the initial single-view projection reconstruction model by utilizing the original data set to obtain a single-view projection reconstruction model, and verifying and evaluating the single-view projection reconstruction model by the test set.
  2. 2. The dual-path condition guided diffusion model-based single-view projection rapid fluorescence tomographic reconstruction system according to claim 1, wherein the view acquisition module comprises: The data set unit is used for calculating fluorescence projection image data by utilizing a forward model according to a simulated fluorescence target set by a computer so as to obtain an original data set consisting of a fluorescence three-dimensional distribution image and a single-view fluorescence projection measurement image, and dividing the original data set into a training set and a verification set according to the proportion of 9:1; The single-view fluorescence projection unit is used for generating a test set of experiments through computer simulation, acquiring a real fluorescence tomographic three-dimensional image of the test set by utilizing a CT image aiming at a real physical imitation or living mouse experiment, and acquiring single-view fluorescence projection data by utilizing a fluorescence camera.
  3. 3. The dual-path condition guided diffusion model-based single-view projection rapid fluorescence tomographic reconstruction system according to claim 1, wherein the model building module comprises: The feature extraction unit is used for combining the jump connection of the condition 2D coding module, the first 3D decoding module and the fault lifting representation to obtain a condition model frame, and extracting the features of the measured fluorescence projection data by adopting a lightweight 4-layer U-Net based on the condition model frame to obtain a condition model, wherein the basic unit of each coding module or decoding module is set to be ResNet structure, and the ResNet structure is utilized to obtain local and global condition features; The iterative denoising unit is used for combining the 3D coding module, the second 3D decoding module and the skip splice to obtain a diffusion model frame, adopting a lightweight 4-layer U-Net based on the diffusion model frame, integrating time step embedding for ResNet units of each coding module or decoding module to iteratively denoise and reconstruct fluorescence three-dimensional distribution to obtain a diffusion model; The image reconstruction unit is used for combining local guidance based on multi-feature fusion and global guidance based on difference-common conversion to obtain a dual-path condition guidance mechanism, transmitting the local and global condition features extracted by the condition model to the diffusion model, and generating a fluorescent molecular tomographic reconstruction three-dimensional image conforming to a physical rule under the condition guidance of the diffusion model so as to obtain an initial single-view projection reconstruction model.
  4. 4. A dual-path condition guided diffusion model based single-view projection fast fluorescence tomographic reconstruction system according to claim 3, wherein the tomographic elevation representation mechanism comprises a first stage and a second stage; The first stage comprises the steps of extracting cross-channel space features from single-view two-dimensional projection by utilizing depth separable convolution, and carrying out channel conversion according to feature depth to realize feature extraction and preliminary lifting, wherein the depth separable convolution consists of depth convolution and point-by-point convolution; The second stage comprises the steps of compressing the lifted features into channel statistics by using global average pooling, converting the channel statistics into excitation factors for scaling the features by using a full connection layer and Sigmoid, and performing shape transformation on the features to obtain final lifted volume feature representation.
  5. 5. A dual-path condition guided diffusion model-based single-view projection rapid fluorescence tomographic reconstruction system according to claim 3, wherein: The local guidance based on multi-feature fusion is used for fusion of double branches of a space and a channel, and in the initialization stage of a diffusion model, multi-feature fusion is carried out on shallow noise features, local condition features and time step embedded features, so that fine local guidance with strong time correlation is provided for diffusion generation; The global guidance based on the difference-common conversion is used for combining deep noise characteristics with global condition characteristics positioned in a condition model middle layer through a difference information injection module and a pair of common information injection modules so as to ensure the reasonability of global structural reconstruction.
  6. 6. The dual-path condition guided diffusion model-based single-view projection rapid fluorescence tomographic reconstruction system according to claim 1, wherein the model optimization module comprises: The optimizing unit is used for training and updating the initial single projection reconstruction model by utilizing the training set, performing model loss calculation by utilizing the verification set and a loss function in the training process, performing super-parameter setting and optimization by utilizing an Adam optimizer and learning rate linear attenuation, and completing model optimizing to obtain a single view projection reconstruction model; And the verification unit is used for verifying and evaluating the single-view projection reconstruction model through a test set.

Description

Single-view projection rapid fluorescence tomography reconstruction system based on dual-path condition guided diffusion model Technical Field The invention relates to the technical field of optical molecular images, in particular to a single-view projection rapid fluorescence tomographic reconstruction system based on a dual-path condition guided diffusion model. Background Fluorescent molecular tomography (Fluorescence Molecular Tomography, FMT) is an important non-invasive three-dimensional (3D) optical imaging technique. By measuring the fluorescent signal on the surface of living organisms and reconstructing the three-dimensional distribution and concentration of the in vivo fluorescent probes, it is possible to quantitatively monitor and visualize the molecular activity in living animal tissues. Compared with structural imaging technologies such as CT, MRI, photoacoustic and the like and functional imaging technologies such as PET and the like, the FMT technology has the advantages of no radiation, low cost, high specificity, high sensitivity and the like. In recent years, FMT has been widely used in various biomedical fields including cardiovascular and cerebrovascular diseases, tumor detection and evaluation, drug development, intra-operative navigation, and the like. Because of the need to capture rapid biological molecule dynamics, FMTs must run at extremely high speeds, often requiring only a single angle of projection image to be taken to shorten the image acquisition time. However, exchanging extremely sparse projection data for time resolution exacerbates the underqualification and morbidity of the inverse problem. Although the emerging second near infrared (NIR-II) based FMT framework (1000-1700 nm) alleviates morbidity to some extent by reducing scatter, the scattering effect still dominates and impedes the fundamental improvement in imaging quality. To address the severe morbidity present in finite projection FMT, researchers have developed two main reconstruction strategies. The first is a traditional iterative regularization method, which solves the inverse problem by introducing regularization terms or sparse priors. Although these methods are theoretically stringent, they suffer from the disadvantages of high computational cost, poor condition numbers, manual parameter adjustment, and the need for ≡3 projections. In addition, their performance is limited by the inherent severe morbidity of the inverse problem, and the fidelity and stability of the reconstruction is often poor. The second is a deep learning method, which directly learns the mapping from measured data to fluorescence distribution through a neural network, such as MAP-PGAN and SVRNet. They can better conform to optical nonlinearity and alleviate morbidity, significantly improving the positioning accuracy and shape recovery capability of the limited/single view FMT. However, the current technology still has the defects of insufficient spatial resolution (more than or equal to 2 mm), limited generalization and the like. Disclosure of Invention In order to overcome the defects of the prior art, the invention aims to provide a single-view projection rapid fluorescence tomographic reconstruction system based on a dual-path condition guided diffusion model, which solves the problem of exacerbation of disease caused by extremely sparse single-view data and overlapping depth information, and can realize rapid dynamic living body three-dimensional imaging of target monitoring target molecules and high-quality reconstruction. In order to achieve the purpose, the invention provides a single-view projection rapid fluorescence tomographic reconstruction system based on a double-path condition-guided diffusion model, which comprises: The view acquisition module is used for acquiring an original data set consisting of a fluorescence three-dimensional distribution image and a single-view fluorescence projection measurement image and acquiring single-view fluorescence projection data in a real experiment; The model construction module is used for defining a conditional model frame for extracting conditional features from boundary fluorescence projection measurement, a diffusion model frame for generating fluorescence three-dimensional distribution by iterative denoising and a dual-path conditional guiding mechanism for connecting the two model frames to obtain an initial single-view projection reconstruction model; The model optimization module is used for training and updating and super-parameter setting and optimizing the initial single-view projection reconstruction model by utilizing the original data set to obtain a single-view projection reconstruction model, and verifying and evaluating the single-view projection reconstruction model by the test set. Optionally, the view acquisition module includes: The data set unit is used for calculating fluorescence projection image data by utilizing a forward model according to a simulated fluo