Bone microscopic CT image reconstruction method and equipment

Through the bone microscopic CT image reconstruction method in sparse sampling mode, the basic visual model of time and frequency domains, super-resolution networks and self-supervised loss function are used to solve the problems of slow bone microscopic CT scanning speed and high radiation dose, and achieve the goal of achieving the High resolution and rapid imaging reduce radiation damage in patients.

CN119048629AActive Publication Date: 2024-11-29SUZHOU INST OF BIOMEDICAL ENG & TECH CHINESE ACADEMY OF SCI
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411235687.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-11-29
Estimated Expiration
2044-09-04

AI Technical Summary

Technical Problem

When scanning large-size and high-density bone structures, bone microscopy is limited by the X-ray source power and focal size, resulting in slow scanning speed, increasing the possibility of radiation dose and motion artifacts in patients, making it difficult to achieve high resolution and Rapid imaging.

Method used

The bone microscopic CT image reconstruction method in sparse sampling mode is adopted to achieve efficient reconstruction of sparse projected images through pre-training of the basic visual model of time and frequency domain, high-dimensional feature coding, super-resolution network architecture and self-supervised loss function, combined with Sine position coding and bilinear interpolation methods improve imaging quality.

Benefits of technology

High resolution and rapid imaging of bone microscopy CT under sparse sampling conditions, reducing the patient's radiation dose damage by more than 80%, improving imaging quality and reducing the appearance of motion artifacts.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048629A_ABST
    Figure CN119048629A_ABST
Patent Text Reader

Abstract

The invention discloses a bone microscopic CT image reconstruction method and equipment, and belongs to the field of microscopic imaging. Through the steps of pre-training of a time domain visual basic model and a frequency domain visual basic model, coordinate sampling of high-dimensional feature coding, training of a continuity representation function, super-resolution reasoning, image reconstruction and the like on a plurality of bone sparse projection images, high resolution and rapid imaging of bone microscopic CT can be realized at the same time in a sparse sampling mode; and more than 80% of radiation dose damage to the patient is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of microscopic imaging, and in particular to a bone microscopic CT image reconstruction method and equipment. Background Art

[0002] Bone micro-CT is an in vivo microscopic imaging system specifically used for joints, bones and their internal microstructures. It can accurately diagnose osteoporosis and assess fracture risk, as well as evaluate osteoarthritis and sports injuries.

[0003] Since the scanned object is a large, high-density bone structure, it is limited by the power and focal size of the X-ray source, and requires an extremely high number of samples, which greatly limits the scanning speed. Excessive scanning time will increase the radiation dose received by the patient or subject, and on the other hand, it may increase the possibility of patient movement, resulting in motion artifacts in the reconstructed image. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, one of the objectives of the present invention is to provide a bone micro-CT image reconstruction method capable of simultaneously achieving high resolution and fast imaging of bone micro-CT in a sparse sampling mode.

[0005] In order to overcome the deficiencies of the prior art, a second object of the present invention is to provide a bone micro-CT image reconstruction device that can simultaneously achieve high resolution and fast imaging of bone micro-CT in a sparse sampling mode.

[0006] One of the purposes of the present invention is achieved by the following technical solution:

[0007] A bone micro-CT image reconstruction method comprises the following steps:

[0008] Pre-training of the time-domain visual basic model and the frequency-domain visual basic model: obtaining a plurality of bone sparse projection images, and using the plurality of bone sparse projection images to perform migration training on the time-domain visual basic model; performing Fourier transform on the plurality of bone sparse projection images to obtain a plurality of bone sparse frequency-domain projection images, and using the plurality of bone sparse frequency-domain projection images to perform migration training on the frequency-domain visual basic model;

[0009] Coordinate sampling for high-dimensional feature encoding: Obtain the encoders of the pre-trained temporal and frequency-domain visual base models, freeze the network parameters, and extract temporal representations from sparse bone CT data. S and frequency domain characterization t F , using the fusion module to accept t S and t F , to obtain complementary feature representation t C , for t C Perform grid processing on high-dimensional features, with a sampling grid of cS , get t G ;

[0010] Fusion of feature grid and coordinate grid: Use sinusoidal position encoding function to map the defined coordinate domain to a high-dimensional function space, γ k (c) = [cos(2 k πu),sin(2 k πu),cos(2 k πv),sin(2 k πv)], where γ k (·) represents the SPE function encoding K th Frequency component, c = (u, v) is the defined coordinate domain;

[0011] Training continuity representation function: Using the continuity representation of high-dimensional functions, constructing a super-resolution network architecture, G and γ k (c S ) are connected in series and input into the multi-layer perceptron MLP of the super-resolution network architecture for predicting the density value

[0012] Super-resolution reasoning and image reconstruction: Using the 2D spatial coordinates c of the full-view sinusoidal graph F =(u F ,v F ) replaces the 2D spatial coordinate c of the sparse sinusoidal graph S =(u S ,v S ), c F =(u F ,v F ) is converted to γ ​​through the SPE function k (c F ) and then γ k (c F ) and t G Combined and fed into a super-resolution network to predict the full-view sinusoidal map right The FBP reconstruction algorithm is applied to obtain the full-view CT image x, where FBP represents the reconstruction operation from the bone CT projection map to the image domain.

[0013] Furthermore, in the pre-training steps of the time-domain visual basic model and the frequency-domain visual basic model, the acquired multiple bone sparse projection images have a viewing angle with a sparsity of 15%-55%.

[0014] Furthermore, in the pre-training steps of the time-domain visual base model and the frequency-domain visual base model, the migration training of the time-domain visual base model and the frequency-domain visual base model is optimized by a self-supervised method, specifically:

[0015] L S =||SDVFM(y),y||1

[0016] L F =||FDVFM(y),y FFT ||1

[0017] Among them, y represents the projection observation data obtained in the sparse sampling mode of bone micro-CT, FFT represents the fast Fourier transform of the projection observation data, SDVFM is the time domain visual basic model, and FDVFM is the frequency domain visual basic model.

[0018] Furthermore, in the coordinate sampling step of the high-dimensional feature encoding, a fusion module is used to accept t S and t F , to obtain complementary feature representation t C Specifically: C =FM(Cat(t S ,t F )), where FM is the feature fusion module, which contains a three-layer convolutional network.

[0019] Furthermore, in the coordinate sampling step of the high-dimensional feature encoding, t C The gridding process of high-dimensional features is as follows: using the cell decoding operation G, which is based on t C and the sampling grid c S As input, the position information of each pixel in the coordinates is then used to obtain t C Extract the corresponding eigenvalues ​​and apply the bilinear interpolation method to calculate the output matrix at t G The eigenvalues ​​at the corresponding positions.

[0020] Furthermore, in the step of fusing the feature grid and the coordinate grid, in order to make the trained super-resolution network applicable to different complex imaging scenes, the generalized feature information learned by the visual basic model from multi-source imaging scenes is coordinate-processed and fused with the SPE function to serve as the input of the super-resolution network.

[0021] Furthermore, in the step of training the continuity representation function, a residual connection is introduced between the input of the MLP and the output of the 10th Linear+ReLU layer to accelerate gradient back propagation and suppress training instability.

[0022] Furthermore, in the step of training the continuity characterization function, the performance of the super-resolution network is improved by a composite self-supervisory loss function, and the composite self-supervisory loss function is:

[0023]

[0024] Among them, in the super-resolution network training, the hyperparameter γ is set to 1e-3.

[0025] Furthermore, in the super-resolution reasoning and image reconstruction steps,

[0026] The second object of the present invention is achieved by adopting the following technical solution:

[0027] A bone micro-CT image reconstruction device, used to implement any one of the above bone micro-CT image reconstruction methods, comprising:

[0028] CT machine, to obtain multiple bone sparse projection images;

[0029] The data processing device performs pre-training of a time domain visual basic model and a frequency domain visual basic model, coordinate sampling of high-dimensional feature encoding, training of continuity representation functions, and super-resolution reasoning and image reconstruction on the multiple bone sparse projection images.

[0030] Compared with the prior art, the bone micro-CT image reconstruction method of the present invention can simultaneously achieve high-resolution and fast imaging of bone micro-CT in a sparse sampling mode, thereby reducing more than 80% of radiation dose damage to patients, by pre-training the time-domain visual basic model and the frequency-domain visual basic model of the multiple bone sparse projection images, sampling the coordinates of high-dimensional feature encoding, training the continuity characterization function, and super-resolution reasoning and image reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a flow chart of the bone micro-CT image reconstruction method of the present invention;

[0032] Figure 2 A network architecture diagram of the bone micro-CT image reconstruction method of the present invention;

[0033] Figure 3 Schematic diagram of high-dimensional feature gridding processing;

[0034] Figure 4 The full-view reconstruction result image in the prior art;

[0035] Figure 5 This is the reconstruction result image of 15% viewing angle;

[0036] Figure 6 This is the reconstruction result image of 20% viewing angle;

[0037] Figure 7 This is the super-resolution output result with a sparsity of 15%;

[0038] Figure 8 This is the super-resolution output result with a sparsity of 20%. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.

[0040] It should be noted that when a component is referred to as being "fixed to" another component, it may be directly on the other component or there may be another intermediate component through which it is fixed. When a component is considered to be "connected to" another component, it may be directly connected to the other component or there may be another intermediate component at the same time. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be another intermediate component at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0042] See also Figure 1 The present invention provides a bone micro-CT image reconstruction method, comprising the following steps:

[0043] Pre-training of the time-domain visual basic model and the frequency-domain visual basic model: obtaining a plurality of bone sparse projection images, and using the plurality of bone sparse projection images to perform migration training on the time-domain visual basic model; performing Fourier transform on the plurality of bone sparse projection images to obtain a plurality of bone sparse frequency-domain projection images, and using the plurality of bone sparse frequency-domain projection images to perform migration training on the frequency-domain visual basic model;

[0044] Coordinate sampling for high-dimensional feature encoding: Obtain the encoders of the pre-trained temporal and frequency-domain visual base models, freeze the network parameters, and extract temporal representations from sparse bone CT data. S and frequency domain characterization t F , using the fusion module to accept t S and t F , to obtain complementary feature representation t C , for t C Perform grid processing on high-dimensional features, with a sampling grid of c S , get tG ;

[0045] Fusion of feature grid and coordinate grid: Use sinusoidal position encoding function to map the defined coordinate domain to a high-dimensional function space, γ k (c) = [cos(2 k πu),sin(2 k πu),cos(2 k πv),sin(2 k πv)], where γ k (·) represents the SPE function encoding K th Frequency component, c = (u, v) is the defined coordinate domain;

[0046] Training continuity representation function: Using the continuity representation of high-dimensional functions, constructing a super-resolution network architecture, G and γ k (c S ) are connected in series and input into the multi-layer perceptron MLP of the super-resolution network architecture for predicting the density value

[0047] Super-resolution reasoning and image reconstruction: Using the 2D spatial coordinates c of the full-view sinusoidal graph F =(u F ,v F ) replaces the 2D spatial coordinate c of the sparse sinusoidal graph S =(u S ,v S ), c F =(u F ,v F ) is converted to γ ​​through the SPE function k (c F ) and then γ k (c F ) and t G Combined and fed into a super-resolution network to predict the full-view sinusoidal map right The FBP reconstruction algorithm is applied to obtain the full-view CT image x, where FBP represents the reconstruction operation from the bone CT projection map to the image domain.

[0048] The present invention provides a bone micro-CT image reconstruction method that combines the powerful data representation capability of the visual basic model with the continuity representation modeling of bone projection data, effectively solving the high-resolution and fast imaging problems of bone micro-CT. Specifically, the present invention uses a pre-trained visual basic model to extract rich feature representations from the time domain and frequency domain of the projection data, and after coordinate processing, includes them in a super-resolution framework based on implicit representation learning, thereby improving the imaging quality of bone micro-CT under extremely sparse acquisition conditions.

[0049] Please continue reading Figure 2 ,The pre-training steps of the time domain visual basic model and the frequency domain visual basic model are as follows:

[0050] Sinusoidal Visual Fundamentals Model (SDVFM) and Frequency Domain Visual Fundamentals Model (FDVFM). The features extracted from different domains by pre-trained visual fundamentals models (VFMs) are used to optimize the downstream image super-resolution tasks. On the one hand, the rich features extracted by VFMs are used to enhance the generalization of the downstream super-resolution model; on the other hand, the imaging performance of bone micro-CT is enhanced by its rich feature characterization. In the present invention, multi-bone sparse projection images are selected, which are 20,000 in this embodiment, with a sparsity of 15%-55% viewing angle, and corresponding multiple frequency domain projection images are used for migration training of SDVFM and FDVFM. The training is optimized using a self-supervised method, and the specific expression is as follows:

[0051] L S =||SDVFM(y),y||1,

[0052] L F =||FDVFM(y),y FFT ||1,

[0053] Where y represents the projection observation data obtained in the sparse sampling mode of bone micro-CT, and FFT represents the fast Fourier transform of the projection observation data. For the pre-trained SDVFM and FDVFM, only their encoders are retained and used for the extraction of time domain and frequency domain features of downstream imaging tasks.

[0054] The coordinate sampling steps for high-dimensional feature encoding are as follows:

[0055] Using the encoders pre-trained with SDVFM and FDVFM, the network parameters are frozen to extract the time domain and frequency domain feature information from the sparsely acquired bone CT data. In order to obtain complementary feature representations, C , using the feature fusion module FM to accept the sinusoidal graph (time domain) representation t S and frequency domain characterization t F Here, the FM module contains three layers of convolutional networks, and the fusion process is specifically defined as:

[0056] t C =FM(Cat(t S ,t F ))

[0057] Using the above formula, we can obtain complementary feature information in the time domain and frequency domain. But we are faced with a new problem: t CThe mismatch between the coordinate grid information of the downstream super-resolution input and the downstream super-resolution process will hinder the execution of the downstream super-resolution process. Therefore, the present invention proposes a gridding operation for high-dimensional features, specifically Figure 3 As shown. Using the cell decoding operation G, the operation takes t C and the sampling grid c S As input, it then uses the position information of each pixel in the coordinates to obtain t C Extract the corresponding eigenvalues ​​and apply the bilinear interpolation method to calculate the output matrix (feature grid) at t G The eigenvalues ​​at the corresponding positions.

[0058] The specific steps of merging the feature grid and the coordinate grid are as follows:

[0059] Considering that the super-resolution network tends to learn from low-frequency functions, we can achieve continuous modeling of high-dimensional functions. For the defined coordinate domain c = (u, v), we use the sinusoidal position encoding (SPE) function to map c = (u, v) to a high-dimensional function space, which is specifically defined as follows:

[0060] γ k (c) = [cos(2 k πu),sin(2 k πu),cos(2 k πv),sin(2 k πv)]

[0061] In the above formula, γ k (·) represents the SPE function encoding K th In addition, in order to make the trained super-resolution network applicable to different complex imaging scenes, the super-resolution input cannot rely solely on the high-dimensional coordinate function input in the above formula. Therefore, we coordinate the generalized feature information learned by the visual basic model from multi-source imaging scenes, and fuse it with the SPE function as the input of the super-resolution network, such as Figure 2 shown.

[0062] The training continuity representation function is specifically:

[0063] This application uses the continuity representation of high-dimensional functions to construct a super-resolution network architecture. Specifically, it includes a 20-layer multi-layer perceptron (MLP), such as Figure 2 Each layer has 320 channels. In order to accelerate gradient back propagation and suppress training instability, a residual connection is introduced between the input of the MLP and the output of the 10th Linear+ReLU layer. G and γ k (c S ) are concatenated and fed into the MLP to predict the density value

[0064]

[0065] In order to improve the performance of the super-resolution network, we designed a carefully constructed composite self-supervised loss function, which consists of L1 loss and frequency loss based on the L1 paradigm, aiming to improve the fidelity of the reconstructed image content and structure. The specific composite self-supervised loss function is expressed as follows:

[0066]

[0067] Among them, in the super-resolution network training, the hyperparameter γ is set to 1e-3.

[0068] The specific steps of super-resolution reasoning and image reconstruction are:

[0069] The network proposed in this application is trained in a self-supervised manner, so during the training phase its predictions are sparse (SV) sinusoidal graphs In order to reconstruct the full-view CT (FVCT) image x, the re-projection technique is used in the inference stage as shown below Figure 2 (B) Workflow where complementary representations are learned from SV sinograms and FFT images. C and the 2D spatial coordinates c of the FV sinogram F =(u F ,v F ) is combined to predict FVCT. Specifically, first use c F =(u F ,v F ) replaces the 2D spatial coordinate c of the SV sinogram S =(u S ,v S ). At the same time, c F =(u F ,v F ) is converted to γ ​​through the SPE function k (c F ). After that, γ k (c F ) and t G Combined and fed into a super-resolution network to predict the FV sinogram Finally, we Apply the FBP reconstruction algorithm to obtain the FVCT image x:

[0070]

[0071] Among them, FBP represents the reconstruction operation from bone CT projection image to image domain.

[0072] The image reconstructed by the bone micro-CT image reconstruction method of the present application is shown in the attached figure. Figure 7 as well as Figure 8 As shown, Figure 7 This is the super-resolution output result when the sparsity is 15%. Figure 8 is the super-resolution output result when the sparsity is 20%. In the prior art, the full-view reconstruction result is as follows Figure 4 As shown, a very high sampling number is required, and the radiation dose is large. When the bone micro-CT image reconstruction method of the present application is not used, and the image reconstruction is performed directly using the 15% viewing angle and the 20% viewing angle, the reconstructed image is as follows Figure 5 as well as Figure 6 As shown, the reconstruction result has significant stripe artifacts and the microstructure inside the bone is severely blurred. In contrast, the super-resolution reconstruction result of the present invention has visual consistency with the full-view reconstruction result of the prior art.

[0073] The present application also discloses a bone micro-CT image reconstruction device for implementing any one of the above bone micro-CT image reconstruction methods, comprising:

[0074] CT machine, to obtain multiple bone sparse projection images;

[0075] The data processing device performs pre-training of a time domain visual basic model and a frequency domain visual basic model, coordinate sampling of high-dimensional feature encoding, training of continuity representation functions, and super-resolution reasoning and image reconstruction on the multiple bone sparse projection images.

[0076] Compared with the prior art, the bone micro-CT image reconstruction method of the present invention can simultaneously achieve high-resolution and fast imaging of bone micro-CT in a sparse sampling mode, thereby reducing more than 80% of radiation dose damage to patients, by pre-training the time-domain visual basic model and the frequency-domain visual basic model of the multiple bone sparse projection images, sampling the coordinates of high-dimensional feature encoding, training the continuity characterization function, and super-resolution reasoning and image reconstruction.

[0077] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which are equivalent modifications and improvements made to the above embodiments based on the essential technology of the present invention, and all of them belong to the protection scope of the present invention.

Claims

1. A bone micro-CT image reconstruction method, characterized in that: The following steps are involved: Pre-training of a time-domain visual basic model and a frequency-domain visual basic model: obtaining a plurality of bone sparse projection images, and using the plurality of bone sparse projection images to perform migration training on the time-domain visual basic model; Performing Fourier transform on the multiple bone sparse projection images to obtain multiple bone sparse frequency domain projection images, and using the multiple bone sparse frequency domain projection images to perform migration training on the frequency domain visual basic model; Coordinate sampling for high-dimensional feature encoding: Obtain the encoders of the pre-trained temporal and frequency-domain visual base models, freeze the network parameters, and extract temporal representations from sparse bone CT data. S and frequency domain characterization t F , using the fusion module to accept t S and t F , to obtain complementary feature representation t C , for t C Perform grid processing on high-dimensional features, with a sampling grid of c S , get t G ; Fusion of feature grid and coordinate grid: Use sinusoidal position encoding function to map the defined coordinate domain to a high-dimensional function space, γ k (c) = [cos(2 k πu),sin(2 k πu),cos(2 k πv),sin(2 k πv)], where γ k (·) represents the SPE function encoding K th Frequency component, c = (u, v) is the defined coordinate domain; Training continuity representation function: Using the continuity representation of high-dimensional functions, constructing a super-resolution network architecture, G and γ k (c S ) are connected in series and input into the multi-layer perceptron MLP of the super-resolution network architecture for predicting the density value Super-resolution reasoning and image reconstruction: Using the 2D spatial coordinates c of the full-view sinusoidal graph F =(u F ,v F ) replaces the 2D spatial coordinate c of the sparse sinusoidal graph S =(u S ,v S ), c F =(u F ,v F ) is converted to γ ​​through the SPE function k (c F ) and then γ k (c F ) and t G Combined and fed into a super-resolution network to predict the full-view sinusoidal map right The FBP reconstruction algorithm is applied to obtain the full-view CT image x, where FBP represents the reconstruction operation from the bone CT projection map to the image domain.

2. The bone micro-CT image reconstruction method according to claim 1, characterized in that: In the pre-training steps of the time-domain visual basic model and the frequency-domain visual basic model, the acquired multiple bone sparse projection images have a viewing angle with a sparsity of 15%-55%.

3. The bone micro-CT image reconstruction method according to claim 1, characterized in that: In the pre-training steps of the time domain visual base model and the frequency domain visual base model, the migration training of the time domain visual base model and the frequency domain visual base model is optimized by a self-supervised method, specifically: L S =||SDVFM(y),y||1 L F =||FDVFM(y),y FFT ||1 Among them, y represents the projection observation data obtained in the sparse sampling mode of bone micro-CT, FFT represents the fast Fourier transform of the projection observation data, SDVFM is the time domain visual basic model, and FDVFM is the frequency domain visual basic model.

4. The bone micro-CT image reconstruction method according to claim 1, characterized in that: In the coordinate sampling step of the high-dimensional feature encoding, a fusion module is used to accept t S and t F , to obtain complementary feature representation t C Specifically: C =FM(Cat(t S ,t F )), where FM is the feature fusion module, which contains a three-layer convolutional network.

5. The bone micro-CT image reconstruction method according to claim 1, characterized in that: In the coordinate sampling step of the high-dimensional feature encoding, t C The gridding process of high-dimensional features is as follows: using the cell decoding operation G, which is based on t C and the sampling grid c S As input, the position information of each pixel in the coordinates is then used to obtain t C Extract the corresponding eigenvalues ​​and apply the bilinear interpolation method to calculate the output matrix at t G The eigenvalues ​​at the corresponding positions.

6. The bone micro-CT image reconstruction method according to claim 1, characterized in that: In the step of fusing the feature grid and the coordinate grid, in order to make the trained super-resolution network applicable to different complex imaging scenes, the generalized feature information learned by the visual basic model from multi-source imaging scenes is coordinate-processed and fused with the SPE function to serve as the input of the super-resolution network.

7. The bone micro-CT image reconstruction method according to claim 1, characterized in that: In the step of training the continuity representation function, a residual connection is introduced between the input of the MLP and the output of the 10th Linear+ReLU layer to accelerate gradient back propagation and suppress training instability.

8. The bone micro-CT image reconstruction method according to claim 1, characterized in that: In the step of training the continuity characterization function, the performance of the super-resolution network is improved by a composite self-supervisory loss function, wherein the composite self-supervisory loss function is: Among them, in the super-resolution network training, the hyperparameter γ is set to 1e-3.

9. The bone micro-CT image reconstruction method according to claim 1, characterized in that: In the super-resolution reasoning and image reconstruction steps, 10. A bone micro-CT image reconstruction device, used to implement the bone micro-CT image reconstruction method according to any one of claims 1 to 9, characterized in that: include CT machine, to obtain multiple bone sparse projection images; The data processing device performs pre-training of a time domain visual basic model and a frequency domain visual basic model, coordinate sampling of high-dimensional feature encoding, training of continuity representation functions, and super-resolution reasoning and image reconstruction on the multiple bone sparse projection images.

Citation Information

Patent Citations

  • Deep learning sparse angle CT reconstruction method based on frequency domain and image domain degradation perception

    CN114140442A

  • Method for detecting and positioning abnormal behavior of target in key area based on multi-domain information fusion

    CN115147921A

  • Morse-based undertiming invariant working mode parameter identification method and device

    CN118468093A