CN117437152B - PET iterative reconstruction method and system based on diffusion model - Google Patents
PET iterative reconstruction method and system based on diffusion model Download PDFInfo
- Publication number
- CN117437152B CN117437152B CN202311764922.5A CN202311764922A CN117437152B CN 117437152 B CN117437152 B CN 117437152B CN 202311764922 A CN202311764922 A CN 202311764922A CN 117437152 B CN117437152 B CN 117437152B
- Authority
- CN
- China
- Prior art keywords
- image
- pet
- diffusion model
- diffusion
- input image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000009792 diffusion process Methods 0.000 title claims abstract description 86
- 238000000034 method Methods 0.000 title claims abstract description 65
- 230000008569 process Effects 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims description 29
- 230000006870 function Effects 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 230000001351 cycling effect Effects 0.000 claims description 2
- 238000003384 imaging method Methods 0.000 abstract description 13
- 230000009467 reduction Effects 0.000 abstract description 6
- 238000003672 processing method Methods 0.000 abstract 1
- 238000002600 positron emission tomography Methods 0.000 description 85
- 238000012545 processing Methods 0.000 description 9
- 210000000538 tail Anatomy 0.000 description 9
- 238000011282 treatment Methods 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000012879 PET imaging Methods 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000001902 propagating effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 229910052802 copper Inorganic materials 0.000 description 1
- 239000010949 copper Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 239000012216 imaging agent Substances 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 230000002503 metabolic effect Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 230000001991 pathophysiological effect Effects 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000002560 therapeutic procedure Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10104—Positron emission tomography [PET]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2211/00—Image generation
- G06T2211/40—Computed tomography
- G06T2211/424—Iterative
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention relates to a PET iterative reconstruction method and system based on a diffusion model, which mainly solve the problems of large noise interference and low imaging quality in the existing PET reconstruction system. According to the invention, a diffusion model is introduced into an iterative reconstruction process, and during each iterative reconstruction, an input image and a corresponding forward projection image are spliced and input into a trained diffusion model for gradual denoising, so that a denoised input image is obtained; reconstructing an output image based on the denoised input image; according to the PET chord graph data processing method, the diffusion model is utilized to reduce noise of PET chord graph data, so that noise interference in an original signal is reduced, and data quality is improved; the reconstruction process is optimized by introducing a diffusion model noise reduction unit in the original iterative reconstruction step, so that the imaging quality of the reconstructed image is further improved. Compared with the prior art, the method adopts an iterative reconstruction method based on a diffusion model, can effectively reduce noise interference, improves image imaging quality, and has wide application prospect.
Description
Technical Field
The invention relates to the field of medical images, in particular to a PET reconstruction system based on a diffusion model and a construction method.
Background
PET (positron emission tomography) is an advanced medical imaging technique, and is widely used in clinical medical research and therapy because of its capability of realizing noninvasive, accurate and high-resolution imaging of internal tissues and organs of the human body. PET imaging technology uses a radioisotope-labeled molecular imaging agent to realize imaging display through reactions such as metabolic conversion of a labeled substance in a human body. Compared with the traditional imaging technology, such as X-ray and MRI, PET has higher sensitivity and specificity, can detect the slight change of pathophysiological activity, and helps doctors to diagnose and treat diseases more accurately.
However, the existing PET reconstruction system has disadvantages in terms of imaging quality and data processing efficiency due to physical characteristics of PET imaging itself and noise interference in the imaging process. Among them, noise interference is one of the main factors affecting PET image quality. In the process of acquiring the original data, problems such as noise, distortion, blurring and the like of an imaging result can be caused by the interference of various factors such as electronic noise, scattering, interaction and the like, so that diagnosis and treatment effects are further affected.
Currently, some methods have emerged to solve the problem of noise interference, for example, using filtering, denoising, downsampling, etc. However, these methods have some drawbacks in practice, such as difficulty in suppressing noise, causing information loss, and the like, and new methods are required to improve the accuracy and reliability of PET imaging.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a PET iterative reconstruction method and a PET iterative reconstruction system based on a diffusion model, which are used for carrying out noise reduction treatment on original chord image data of a PET image by means of the characteristics of average filtering effect, edge retention, expansibility and the like of the diffusion model in image noise reduction, and fusing the process into the traditional iterative reconstruction, thereby reducing the noise level of the PET reconstructed image and being better for clinical use.
The aim of the invention is realized by the following technical scheme:
a PET iterative reconstruction method based on a diffusion model comprises the following steps in each PET iterative reconstruction process:
splicing and inputting the input image and the corresponding forward projection image into a trained diffusion model to perform gradual denoising, so as to obtain a denoised input image;
reconstructing an output image based on the denoised input image;
judging whether a preset iteration stopping condition is met, and if so, taking the output image after the iteration as a final PET iteration reconstruction image; otherwise, taking the output image after the current iteration as the input image of the next iteration, and executing the next iteration;
the initial value of the input image is PET original chord graph data; the corresponding forward projection image is obtained by reconstructing PET original chord graph data once and then projecting forward.
Further, the trained diffusion model is obtained by training based on noise diffusion learning by using a training data set; wherein each sample of the training dataset comprises PET raw chordal image data and a corresponding forward projection image.
Further, the sizes of the input image and the forward projection image are matched with the size of the trained diffusion model network structure.
Further, training based on noise diffusion learning is specifically:
carrying out diffusion on PET original chord image data in each sample of a training data set by utilizing random Gaussian noise to obtain a diffusion image, splicing the diffusion image with a corresponding forward projection image to serve as input of a diffusion model, carrying out network back propagation by a minimized loss function to complete gradient updating, and cycling until the model converges or reaches training times to obtain a trained diffusion model; the loss function comprises the deviation loss of the prediction noise and the random Gaussian noise output by the diffusion model.
Further, the loss function is specifically:
；
wherein,prediction noise indicative of output +.>Representing true random gaussian noise, Σ represents the cumulative summation over all pixels; n represents the number of all pixels in the image and i represents the pixel number.
Further, the network structure adopted by the diffusion model is U-net, FCN, deepLab or PSPNet.
A diffusion model-based PET iterative reconstruction system comprising:
the denoising module is used for splicing and inputting the input image and the corresponding forward projection image into the trained diffusion model to perform gradual denoising, so as to obtain a denoised input image; the initial value of the input image is PET original chord graph data; the corresponding forward projection image is obtained by carrying out primary reconstruction on PET original chord graph data and then forward projection;
the reconstruction module is used for reconstructing and obtaining an output image based on the denoised input image;
the output module is used for judging whether a preset iteration stop condition is met, and if so, taking the output image after the iteration as a final PET iteration reconstruction image; otherwise, the output image after the current iteration is used as the input image of the next iteration, and the next iteration is executed.
Further, the system further comprises:
and the training module is used for training based on noise diffusion learning by utilizing the training data set to obtain a diffusion model.
An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of iterative reconstruction of PET based on a diffusion model when executing the computer program.
A storage medium containing computer executable instructions that when executed by a computer processor implement the one diffusion model-based PET iterative reconstruction method.
The beneficial effects of the invention are as follows: the diffusion model is combined into the traditional PET iterative reconstruction, so that the influence of noise can be effectively reduced, and the quality and accuracy of images are improved; detail information can be better reserved, so that reconstructed images are clearer and richer, and accurate diagnosis and disease assessment of doctors are facilitated; the spatial resolution of the PET reconstructed image can be improved; the convergence speed can be increased, the iteration times can be reduced, and the reconstruction time can be shortened.
Drawings
FIG. 1 is a flow chart of a PET iterative reconstruction method based on a diffusion model provided by the invention;
FIG. 2 is a flowchart of a training process of a diffusion model according to an embodiment of the present invention;
FIG. 3 is a flowchart of a PET iterative reconstruction method based on a diffusion model provided by an embodiment of the invention;
FIG. 4 is a flowchart of a diffusion model noise reduction process according to an embodiment of the present invention;
FIG. 5 is a block diagram of a PET iterative reconstruction system based on a diffusion model provided by an embodiment of the invention;
fig. 6 is a block diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present application as detailed in the accompanying claims.
The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application.
As used in this application and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items.
It should be understood that although the terms first, second, third, etc. may be used herein to describe various information, these information should not be limited by these terms. These terms are only used to distinguish one type of information from another. For example, a first message may also be referred to as a second message, and similarly, a second message may also be referred to as a first message, without departing from the scope of the present application. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.
In order to solve the problems of noise interference, low imaging quality and the like in the existing PET reconstruction system, the invention provides a PET iterative reconstruction method and system based on a diffusion model. The method is mainly characterized in that in the PET iterative reconstruction process, the data of the PET original chord chart are continuously optimized, and the noise in the original chord chart data is reduced, so that the noise of a reconstructed image is reduced, and the purposes of improving the imaging quality and the data processing efficiency are achieved. The invention provides a PET iterative reconstruction method based on a diffusion model, specifically, as shown in figure 1, in each PET iterative reconstruction process, the method comprises the following steps:
splicing and inputting the input image and the corresponding forward projection image into a trained diffusion model to perform gradual denoising, so as to obtain a denoised input image;
reconstructing an output image based on the denoised input image;
judging whether a preset iteration stopping condition is met, and if so, taking the output image after the iteration as a final PET iteration reconstruction image; otherwise, taking the output image after the current iteration as the input image of the next iteration, and executing the next iteration;
the initial value of the input image is PET original chord graph data; the corresponding forward projection image is obtained by reconstructing PET original chord graph data once and then projecting forward.
The method of the invention takes the forward projection image as the condition information, the prior constraint can help to limit the possibility space of the reconstructed image, effectively constrain the noise and the artifact of the reconstructed image and improve the noise reduction effect; the forward projection image contains the general shape and structure information of the PET scanning object, so that the reconstructed image can be ensured to remove noise while the structure of the object is maintained, and the detail information in the image can be maintained; the range of the knowledge space can be reduced, so that the algorithm approaches the optimal solution more quickly, and the convergence process of the reconstruction method is accelerated; and the occurrence of artifacts, noise and false structures in the reconstructed image can be effectively reduced, so that the image quality and accuracy are improved.
The present invention will be described in detail with reference to examples and drawings.
A PET iterative reconstruction method based on a diffusion model comprises two stages:
the first stage: training phase, comprising the following steps:
step one: constructing a training data set, wherein each sample of the training data set comprises PET original chord graph data and a corresponding forward projection image; wherein the forward projection image is used as a priori information.
As a preferred solution, in order to make the generalization ability of the training model stronger, the data set may be expanded by acquiring low-dose data in a downsampling manner. Therefore, the step needs to acquire low-dose PET original chord chart data and normal-dose PET original chord chart data and forward projection images thereof; the specific method for obtaining the medicine is as follows:
1. original normal dose PET raw List-mode data L norm Normal dose PET raw List-mode data L using a downsampling tool norm Downsampling to half the data volume in time series, treated as low dose PET raw List-mode data L low The method comprises the steps of carrying out a first treatment on the surface of the Wherein, the down sampling mode according to the time sequence can be set according to the requirement, the PET original List-mode data L of normal dosage is obtained in the embodiment norm The low dose data is simulated by collecting selected data at intervals of n seconds over n seconds and then n seconds over the original time series. In this embodiment, n=5.
2. Original List-mode data L of PET of normal dose norm And low dose PET sourceList-mode data L low Respectively projecting the PET original chord graph data S to the data frequency domain to obtain normal dose PET original chord graph data S norm And low dose PET raw chordal image data S low ；
3. Normal dose PET raw chordal map data S using conventional iterative reconstruction methods norm And low dose PET raw chordal image data S low Respectively carrying out iterative reconstruction to obtain a standard image I of normal dose after reconstruction norm And low dose standard image I low The method comprises the steps of carrying out a first treatment on the surface of the The iterative reconstruction method can adopt a traditional iterative reconstruction method, such as an OSEM iterative reconstruction method, and the number of iterations is set to be 3 in the embodiment;
4. standard image I of normal dose using corresponding forward tool norm And low dose standard image I low Forward projection operation is carried out to obtain a forward projection image S of normal dose corresponding to the frequency domain Pnorm And a low dose forward projection image S Plow The method comprises the steps of carrying out a first treatment on the surface of the Raw chordal plot data S of normal dose PET norm And low dose PET raw chordal image data S low Forward projection images S of the respective and corresponding normal doses Pnorm And a low dose forward projection image S Plow And combining to obtain a sample, and constructing to obtain a training data set.
As another preferred solution, the PET original chord chart data and the corresponding forward projection image may be segmented to form each sample, so as to increase the data volume and simultaneously remove the limitation of the video memory of the display card, the segmented image size is matched with the actual network structure size, the segmentation mode may be performed by a sliding window mode, and the segmentation of the image size of 256×256×1 is taken as an example, and specifically as follows:
using a 256 x 1 rectangular frame, smoothly moving on PET original chord graph data and forward projection images, wherein the x-axis step length is 128, the y-axis step length is 128, the z-axis step length is 1, recording position coordinates, and obtaining an image with the image size of 256 x 1 in each step to obtain PET original chord graph data slices I '' Scut And corresponding forward projection image slice I' Pcut And each sample is constructed.
Step two: constructing a conditional diffusion model network and training:
(2.1) designing a diffusion model based on an end-to-end deep convolutional neural network, in this embodiment, UNet is used as the end-to-end deep convolutional neural network, and other convolutional neural networks can be used as the base network for subsequent noise prediction.
(2.2) the diffusion model in the present invention is a conditional diffusion model, as shown in FIG. 2, in which PET raw chord chart data is sliced I' Scut As input data, the corresponding forward projection image slice I 'is taken' Pcut The condition data is used as input data and used for training a condition diffusion model; the time step of the conditional diffusion model is t=1, 2, …, T; in the training process, for any step number t, data I 'is input' Scut Is based on superposition of random Gaussian noise related to t stepsObtaining a diffusion image I 'of the t step' Scut,t The formula is->Wherein beta is t The T-th value in the β vector set is linearly preset according to the size of T, in this embodiment t=1000. />Represented in given input data I' Scut Under the condition of (1) the diffusion image I 'of the t-th step' Scut,t Distribution of (I), i.e. at I' Scut The probability distribution of the image obtained after randomly adding noise on the basis of the above; />Mean value of +.>Variance is beta t I, a multidimensional Gaussian distribution for describing the distribution at a given I' Scut And under the condition of random noise, the distribution of the diffusion image obtained in the step t; i represents an identity matrix indicating the same standard deviation beta for each dimension t 。
Diffusion image I 'of the t step' Scut,t And corresponding forward projection image slice I' Pcut Adding the two channels into a diffusion model based on an end-to-end deep convolutional neural network to perform prediction noise, and outputting the prediction noiseAnd true random Gaussian noise->Comparing, calculating deviation Loss of the two as Loss of network training, feeding back to complete gradient updating, and circulating until the model converges or the training times are reached to obtain a trained diffusion model; where the bias Loss is calculated using MSE Loss in this embodiment: />. Wherein (1)>Prediction noise indicative of output +.>Representing true random gaussian noise, Σ represents the cumulative summation over all pixels; n represents the number of all pixels in the image and i represents the pixel number.
And in the second stage, PET iterative reconstruction is carried out based on a trained diffusion model, wherein prediction noise is gradually obtained by utilizing the trained diffusion model through a conditional back sampling process, and denoising is carried out on an input image based on the prediction noise, as shown in fig. 3, and specifically the method comprises the following steps of:
step one: acquiring PET original chord graph data to be reconstructed; likewise, the PET raw List-mode data L to be reconstructed norm Projecting to a data frequency domain to obtain PET original chord image data to be reconstructed;
step two: acquiring a forward projection image; carrying out iterative reconstruction on PET original chord image data to be reconstructed for one time to obtain a reconstructed image, and carrying out forward projection operation on the reconstructed image by using a corresponding forward projection tool to obtain a forward projection image of a corresponding frequency domain;
similarly, dividing the PET original chord image data and the corresponding forward projection image into 256-1 image-sized slices in a sliding window mode;
step three: as shown in fig. 4, each PET original chord image data slice is taken as an initial value of an input image and a corresponding forward projection image slice are spliced and input into a trained diffusion model, wherein the sampling steps of the model are set to be t=t, T-1, …,1,0, an input image which is subjected to denoising through prediction noise output by the diffusion model based on an end-to-end deep convolutional neural network is taken as the next input of the trained diffusion model, gradual denoising is performed until t=0, and a final denoised PET chord image data slice is obtained; according to the reverse operation mode of the segmented image, splicing each final PET chord graph data slice after noise reduction into complete PET chord graph data after noise reduction by using a sliding window scheme for image reconstruction; the specific operation is as follows:
and slicing the obtained PET chord graph data with the size of 256 x 1, and placing the PET chord graph data back into a matrix of original PET original chord graph data according to the position coordinates, wherein the covering operation is performed during the slicing, and the average value operation is performed on the covering part at the moment to obtain the final chord graph data, namely the complete denoised input image.
Step four: reconstructing an output image based on the complete denoised input image;
judging whether a preset iteration stopping condition is met, and if so, taking the output image after the iteration as a final PET iteration reconstruction image; otherwise, taking the output image after the current iteration as the input image of the next iteration, and executing the next iteration according to the steps III to IV. In this embodiment, the upper limit of the number of iterations is set to 3.
Corresponding to the embodiment of the PET iterative reconstruction method based on the diffusion model, the invention also provides an embodiment of the PET iterative reconstruction system based on the diffusion model.
Referring to fig. 5, a PET iterative reconstruction system based on a diffusion model according to an embodiment of the present invention includes:
the denoising module is used for splicing and inputting the input image and the corresponding forward projection image into the trained diffusion model to perform gradual denoising, so as to obtain a denoised input image; the initial value of the input image is PET original chord graph data; the corresponding forward projection image is obtained by carrying out primary reconstruction on PET original chord graph data and then forward projection;
the reconstruction module is used for reconstructing and obtaining an output image based on the denoised input image;
the output module is used for judging whether a preset iteration stop condition is met, and if so, taking the output image after the iteration as a final PET iteration reconstruction image; otherwise, the output image after the current iteration is used as the input image of the next iteration, and the next iteration is executed.
The denoising module performs denoising treatment on PET original chord graph data by using a diffusion model, so that noise interference in the original data is reduced, and the influence of acquisition noise on image quality is radically reduced; the reconstruction module is used for further optimizing and improving the imaging quality of the reconstructed image by embedding the denoising module into the iterative reconstruction process and combining the judgment of the output module.
Compared with the prior art, the method adopts an iterative reconstruction method based on a diffusion model, can effectively reduce noise interference, improves image imaging quality, and has wide application prospect. The invention improves the precision and reliability of PET imaging technology and further promotes the development of clinical diagnosis, treatment and research.
The embodiment of the PET iterative reconstruction system based on the diffusion model can be applied to any device with data processing capability, such as a computer or a device.
The system embodiment may be implemented in software, or in hardware or a combination of hardware and software. Taking software implementation as an example, as a device in a logic sense, a processor of any device with data processing capability reads corresponding computer program instructions in a nonvolatile memory to a memory to run to form an electronic device diagram from a hardware level, as shown in fig. 6, which is an electronic device diagram for implementing the PET iterative reconstruction method based on a diffusion model according to the embodiment of the present invention, and the electronic device diagram includes a memory, a processor and a computer program stored in the nonvolatile memory and capable of running on the processor, where the processor implements the PET iterative reconstruction method based on the diffusion model when executing the computer program. In addition to the processor, the memory, the network interface, and the nonvolatile memory shown in fig. 6, any device with data processing capability in the system in the embodiment generally includes other hardware according to the actual function of the any device with data processing capability, which will not be described herein.
The implementation process of the functions and roles of each unit in the above system is specifically shown in the implementation process of the corresponding steps in the above method, and will not be described herein again.
For system embodiments, reference is made to the description of method embodiments for the relevant points, since they essentially correspond to the method embodiments. The system embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purposes of the present invention. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
The embodiment of the invention also provides a computer readable storage medium, on which a program is stored, which when executed by a processor, implements a PET iterative reconstruction method based on a diffusion model in the above embodiment.
The computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: portable computer disks, hard disks, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static Random Access Memory (SRAM), portable compact disk read-only memory (CD-ROM), digital Versatile Disks (DVD), memory sticks, floppy disks, mechanical coding devices, punch cards or in-groove structures such as punch cards or grooves having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media, as used herein, are not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical pulses through fiber optic cables), or electrical signals transmitted through wires.
The computer readable program instructions described herein may be downloaded from a computer readable storage medium to a respective computing/processing device or to an external computer or external storage device over a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmissions, wireless transmissions, routers, firewalls, switches, gateway computers and/or edge servers. The network interface card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium in the respective computing/processing device.
It is apparent that the above examples are given by way of illustration only and are not limiting of the embodiments. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary or exhaustive of all embodiments. And obvious variations or modifications thereof are contemplated as falling within the scope of the present invention.
Claims (10)
1. A diffusion model-based PET iterative reconstruction method, comprising, in each PET iterative reconstruction process:
splicing and inputting the input image and the corresponding forward projection image into a trained diffusion model to perform gradual denoising, so as to obtain a denoised input image;
reconstructing an output image based on the denoised input image;
judging whether a preset iteration stopping condition is met, and if so, taking the output image after the iteration as a final PET iteration reconstruction image; otherwise, taking the output image after the current iteration as the input image of the next iteration, and executing the next iteration;
the initial value of the input image is PET original chord graph data; the corresponding forward projection image is obtained by reconstructing PET original chord graph data once and then projecting forward.
2. The method of claim 1, wherein the trained diffusion model is obtained by training based on noise diffusion learning using a training dataset; wherein each sample of the training dataset comprises PET raw chordal image data and a corresponding forward projection image.
3. The method of claim 2, wherein the size of the input image, the forward projection image, and the size of the trained diffusion model network structure are matched.
4. The method according to claim 2, wherein training based on noise diffusion learning is specifically:
carrying out diffusion on PET original chord image data in each sample of a training data set by utilizing random Gaussian noise to obtain a diffusion image, splicing the diffusion image with a corresponding forward projection image to serve as input of a diffusion model, carrying out network back propagation by a minimized loss function to complete gradient updating, and cycling until the model converges or reaches training times to obtain a trained diffusion model; the loss function comprises the deviation loss of the prediction noise and the random Gaussian noise output by the diffusion model.
5. The method according to claim 4, characterized in that the loss function is in particular:
；
wherein,prediction noise indicative of output +.>Representing true random gaussian noise, Σ represents the cumulative summation over all pixels; n represents the number of all pixels in the image and i represents the pixel number.
6. The method of claim 1, wherein the diffusion model employs a network structure of U-net, FCN, deepLab or PSPNet.
7. A diffusion model-based PET iterative reconstruction system, comprising:
the denoising module is used for splicing and inputting the input image and the corresponding forward projection image into the trained diffusion model to perform gradual denoising, so as to obtain a denoised input image; the initial value of the input image is PET original chord graph data; the corresponding forward projection image is obtained by carrying out primary reconstruction on PET original chord graph data and then forward projection;
the reconstruction module is used for reconstructing and obtaining an output image based on the denoised input image;
the output module is used for judging whether a preset iteration stop condition is met, and if so, taking the output image after the iteration as a final PET iteration reconstruction image; otherwise, the output image after the current iteration is used as the input image of the next iteration, and the next iteration is executed.
8. The system of claim 7, wherein the system further comprises:
and the training module is used for training based on noise diffusion learning by utilizing the training data set to obtain a diffusion model.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements a diffusion model based PET iterative reconstruction method as claimed in any one of claims 1-6 when executing the computer program.
10. A storage medium containing computer executable instructions which when executed by a computer processor implement a diffusion model based PET iterative reconstruction method as claimed in any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311764922.5A CN117437152B (en) | 2023-12-21 | 2023-12-21 | PET iterative reconstruction method and system based on diffusion model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311764922.5A CN117437152B (en) | 2023-12-21 | 2023-12-21 | PET iterative reconstruction method and system based on diffusion model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117437152A CN117437152A (en) | 2024-01-23 |
| CN117437152B true CN117437152B (en) | 2024-04-02 |
Family
ID=89558634
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311764922.5A Active CN117437152B (en) | 2023-12-21 | 2023-12-21 | PET iterative reconstruction method and system based on diffusion model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117437152B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117951605B (en) * | 2024-03-26 | 2024-06-07 | 苏州元脑智能科技有限公司 | Quantization method and device for diffusion model, computer equipment and storage medium |
| CN118297836B (en) * | 2024-04-22 | 2024-09-27 | 上海大学 | Rapid image generation method and device based on space sparse diffusion model |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023279316A1 (en) * | 2021-07-08 | 2023-01-12 | 深圳高性能医疗器械国家研究院有限公司 | Pet reconstruction method based on denoising score matching network |
| CN115880187A (en) * | 2023-01-10 | 2023-03-31 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Single-image reflection removing method based on denoising diffusion probability model and related equipment |
| CN115984401A (en) * | 2023-01-28 | 2023-04-18 | 浙江大学 | Dynamic PET image reconstruction method based on model-driven deep learning |
| CN116524048A (en) * | 2023-04-28 | 2023-08-01 | 南开大学 | Natural image compressed sensing method based on potential diffusion model |
| CN116993610A (en) * | 2023-07-25 | 2023-11-03 | 湖北珞珈实验室 | Method and equipment for reconstructing restricted CT image based on denoising diffusion model |
| CN117011673A (en) * | 2023-10-07 | 2023-11-07 | 之江实验室 | Electrical impedance tomography image reconstruction method and device based on noise diffusion learning |
| CN117218013A (en) * | 2023-08-01 | 2023-12-12 | 鹏城实验室 | Event camera image processing method, training method, system, equipment and medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7251306B2 (en) * | 2004-11-17 | 2007-07-31 | General Electric Company | Methods, apparatus, and software to facilitate iterative reconstruction of images |
-
2023
- 2023-12-21 CN CN202311764922.5A patent/CN117437152B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023279316A1 (en) * | 2021-07-08 | 2023-01-12 | 深圳高性能医疗器械国家研究院有限公司 | Pet reconstruction method based on denoising score matching network |
| CN115880187A (en) * | 2023-01-10 | 2023-03-31 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Single-image reflection removing method based on denoising diffusion probability model and related equipment |
| CN115984401A (en) * | 2023-01-28 | 2023-04-18 | 浙江大学 | Dynamic PET image reconstruction method based on model-driven deep learning |
| CN116524048A (en) * | 2023-04-28 | 2023-08-01 | 南开大学 | Natural image compressed sensing method based on potential diffusion model |
| CN116993610A (en) * | 2023-07-25 | 2023-11-03 | 湖北珞珈实验室 | Method and equipment for reconstructing restricted CT image based on denoising diffusion model |
| CN117218013A (en) * | 2023-08-01 | 2023-12-12 | 鹏城实验室 | Event camera image processing method, training method, system, equipment and medium |
| CN117011673A (en) * | 2023-10-07 | 2023-11-07 | 之江实验室 | Electrical impedance tomography image reconstruction method and device based on noise diffusion learning |
Non-Patent Citations (3)
| Title |
|---|
| Direct PET Image Reconstruction Incorporating Deep Image Prior and a Forward Projection Model;Fumio Hashimoto等;《IEEE Transactions on Radiation and Plasma Medical Sciences》;20221130;第6卷(第8期);摘要，正文第I-II节 * |
| Kuang Gong PhD等.The Evolution of Image Reconstruction in PET: From Filtered Back-Projection to Artificial Intelligence.《PET Clinics》.2021,第16卷(第4期),全文. * |
| PET成像的高分辨率快速局域重建算法的建立;朱闻韬等;《中国医学装备》;20151031;第12卷(第10期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117437152A (en) | 2024-01-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117437152B (en) | PET iterative reconstruction method and system based on diffusion model | |
| CN109035169B (en) | Unsupervised/semi-supervised CT image reconstruction depth network training method | |
| Xie et al. | Artifact removal using improved GoogLeNet for sparse-view CT reconstruction | |
| CN110809782B (en) | Attenuation correction system and method | |
| Dong et al. | A deep learning reconstruction framework for X-ray computed tomography with incomplete data | |
| US11540798B2 (en) | Dilated convolutional neural network system and method for positron emission tomography (PET) image denoising | |
| Cui et al. | Deep reconstruction model for dynamic PET images | |
| WO2024066049A1 (en) | Pet image denoising method, terminal device, and readable storage medium | |
| Hutton | Recent advances in iterative reconstruction for clinical SPECT/PET and CT | |
| CN102184559B (en) | Particle filtering-based method of reconstructing static PET (Positron Emission Tomograph) images | |
| Majee et al. | 4D X-ray CT reconstruction using multi-slice fusion | |
| Ote et al. | Deep-learning-based fast TOF-PET image reconstruction using direction information | |
| CN109523605A (en) | A kind of method, apparatus, equipment and the medium of CT image reconstruction | |
| Guo et al. | Deep learning kidney segmentation with very limited training data using a cascaded convolution neural network | |
| Corda-D’Incan et al. | Memory-efficient training for fully unrolled deep learned PET image reconstruction with iteration-dependent targets | |
| Chan et al. | An attention-based deep convolutional neural network for ultra-sparse-view CT reconstruction | |
| US11200669B2 (en) | Systems and methods for determining plasma input function used in positron emission tomography imaging | |
| CN110874855B (en) | Collaborative imaging method and device, storage medium and collaborative imaging equipment | |
| US8989462B2 (en) | Systems, methods and computer readable storage mediums storing instructions for applying multiscale bilateral filtering to magnetic resonance (RI) images | |
| CN111243052A (en) | Image reconstruction method and device, computer equipment and storage medium | |
| CN114187235A (en) | Artifact insensitive medical image deformation field extraction method and registration method and device | |
| Zhou et al. | High-resolution hierarchical adversarial learning for OCT speckle noise reduction | |
| EP4343680A1 (en) | De-noising data | |
| CN111968192A (en) | Construction method of CT image, CT device and storage medium | |
| Luo et al. | Research on several key problems of medical image segmentation and virtual surgery |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |