WO2022109928A1

WO2022109928A1 - Image reconstruction method and application

Info

Publication number: WO2022109928A1
Application number: PCT/CN2020/131792
Authority: WO
Inventors: 郑海荣; 梁栋; 胡战利; 刘新; 黄英
Original assignee: 深圳先进技术研究院
Priority date: 2020-11-26
Filing date: 2020-11-26
Publication date: 2022-06-02

Abstract

The present invention belongs to the technical field of computed tomography, and in particular relates to an image reconstruction method and the application thereof. In a conventional filtered back-projection reconstructed image, the method may result in excessive noise and streak artifacts, which further leads to diagnostic and quantitative errors. Provided is an image reconstruction method. The method comprises the following steps: step 1: performing reconstruction on corrected projection data, so as to obtain reconstructed data; step 2: performing regularization on the reconstructed data, so as to obtain a regularized image; step 3: performing reconstruction on the regularized image, so as to obtain a reconstructed image; and step 4: taking the reconstructed image as priori information, performing guided kernel filter under the guidance of the reconstructed image, so as to obtain a recovered image. The recovered image has good structure information.

Description

An image reconstruction method and application

technical field

The present application belongs to the technical field of computed tomography, and in particular relates to an image reconstruction method and application.

Background technique

Computed tomography (referred to as X-CT or CT) is to use x-rays to scan the human body, and then the analog signal r received by the detector is converted into a digital signal, and the attenuation of each pixel is calculated by an electronic computer. It is a device that can display the tomographic structure of various parts of the human body by reconstructing the image. Image fusion refers to the image data collected by multi-source channels about the same target through image processing and computer technology, etc., to maximize the extraction of favorable information in the respective channels, and finally to synthesize high-quality images. Improve the utilization rate of image information, improve the accuracy and reliability of computer interpretation, improve the spatial resolution and spectral resolution of the original image, and facilitate monitoring. The images to be fused have been registered and have the same pixel bit width.

Radiation risk in computed tomography (CT) and the associated cancer risk have been a major clinical concern. Low-dose CT imaging can be achieved by reducing the number of projections per revolution of the human body by reducing the milliamp-second scanning scheme.

However, in conventional filtered back-projection reconstructed images, this approach may lead to excessive noise and streak artifacts, which further lead to diagnostic and quantitative errors.

SUMMARY OF THE INVENTION

1. Technical problems to be solved

Based on the problem that in traditional filtered back-projection reconstructed images, this method may lead to excessive noise and streak artifacts, which further lead to the problem of diagnostic and quantitative errors, the present application provides an image reconstruction method and application.

2. Technical solutions

In order to achieve the above purpose, the present application provides an image reconstruction method, which includes the following steps: Step 1: reconstruct the corrected projection data; obtain reconstructed data; Step 2: regularize the reconstructed data , obtain a regularized image; Step 3: Reconstruct the regularized image to obtain a reconstructed image; Step 4: Use the reconstructed image as prior information, and conduct a guided kernel under the guidance of the reconstructed image Filter (Guided Kernel Filter) to get the restored image.

Another implementation manner provided by the present application is as follows: in the step 1, the corrected projection data is reconstructed by a penalized weighted least squares algorithm to obtain an ideal projection vector to be estimated.

Another embodiment provided by the present application is: the step 1 uses a sinogram recovery method based on a penalized weighted least squares algorithm to recover sinogram data using a combination of low milliampere seconds and sparse view protocols.

Another embodiment provided by the present application is: in step 2, the total variation method based on penalized weighted least squares is used to reconstruct the image using the recovered sinogram data, and then the total variation method based on penalized weighted least squares is used to reconstruct the image. The non-quadratic penalty function performs neighborhood patches instead of single-pixel regularization.

Another implementation manner provided by the present application is: the step 3 uses a regularization algorithm to reconstruct the regularized image.

Another embodiment provided by the present application is: in step 4, a sparse matrix is constructed, and the matrix is normalized to obtain a normalized kernel matrix; then guided kernel filtering is performed under the guidance of the reconstructed image to obtain Restore the image.

Another implementation manner provided by the present application is: the matrix is constructed by using the k-nearest neighbor algorithm.

Another embodiment provided by the present application is: when calculating the image roughness between adjacent pixels in the penalized weighted least squares total variation method, a patch related to each pixel is used.

Another embodiment provided by this application is: the roughness function of the patch is expressed as:

where U(μ) is the patch-based roughness function, f _j (μ), f _k (μ) are feature vectors consisting of the intensity values of all pixels in the center of the patch at pixel j and pixel k, respectively, and h is A positive weighting factor equal to the normalized inverse spatial distance between pixel j and pixel k, ψ(||f _j (μ)-f _k (μ)|| _h ) is a non-quadratic penalty function.

The present application also provides an application of the image reconstruction method, where the image reconstruction method is applied to image noise reduction and artifact removal or image fusion.

3. Beneficial effects

Compared with the prior art, the beneficial effects of the image reconstruction method provided by the present application are:

The image reconstruction method provided in this application is an image reconstruction method of sparse viewing angle and image fusion.

The image reconstruction method provided in this application solves the problem of improving the quality of low-millisecond or sparse reconstructed images.

The image reconstruction method provided in this application, iterative image reconstruction, is a promising option for obtaining high-quality images under low-dose scanning. Therefore, research and development of new low-dose CT reconstruction methods can not only ensure CT imaging quality but also reduce harmful radiation doses, which has important scientific significance and application prospects in the field of medical diagnosis.

With the image reconstruction method provided in this application, the finally obtained restored image not only suppresses noise, but also retains image details.

The image reconstruction method provided in this application includes complete reconstruction and restoration of CT images.

In the image reconstruction method provided in this application, the reconstructed image of PWLS-TV has good structural information.

The image reconstruction method provided in this application is based on image update regularized by a non-quadratic penalty function of patches. The method utilizes spatial regularization to penalize image intensity differences between adjacent pixels, thereby improving image quality. Non-quadratic penalty functions can preserve edges.

Description of drawings

1 is a schematic diagram of the overall framework of the image reconstruction method of the present application;

FIG. 2 is a schematic diagram of an image reconstruction result of the present application.

Detailed ways

First, this study simulates low-dose CT images with different views (ray intensity I0=1*10^5). For simulation studies on low-mas projection data, noise measurements at each i are generated from a statistical model of the pre-log projection data after computing the noise-free line integral y _i in the model-based direct projection operation

In the formula, I0 is the incident X-ray intensity, y _i is the sinogram data,

is the variance of the background electronic noise, set to 0.1. Calculate the noise measurement _yi using the log transform of _bi . When used for simulation of projected data, the original 360 views were undersampled to 4 levels, namely: 60, 90, 120 and 180 views.

Hereinafter, specific embodiments of the present application will be described in detail with reference to the accompanying drawings, from which those skilled in the art can clearly understand the present application and be able to implement the present application. Without departing from the principles of the present application, the features of the various embodiments may be combined to obtain new embodiments, or instead of certain features of certain embodiments, to obtain other preferred embodiments.

Due to the large dose in the CT scanning process, the existing method has potential harm to the human body due to radiation exposure; due to the large amount of data collected, the image reconstruction speed is slow; (3) due to the long scanning time, it leads to the occurrence of patient movement problems. induced artifacts.

Regularization means that in linear algebra theory, an ill-posed problem is usually defined by a set of linear algebraic equations, and this set of equations is usually derived from an ill-posed inverse problem with a large condition number. A large condition number means that rounding errors or other errors can seriously affect the outcome of the problem.

1-2, the present application provides an image reconstruction method, the method includes the following steps:

Step 1: reconstruct the corrected projection data; obtain reconstructed data; Step 2: perform regularization on the reconstructed data to obtain a regularized image; Step 3: reconstruct the regularized image to obtain a reconstructed image ; Step 4: take the reconstructed image as prior information, and carry out guided kernel filtering (Guided Kernel Filter) under the guidance of the reconstructed image to obtain a restored image.

Further, in the step 1, the corrected projection data is reconstructed by penalized weighted least squares (PWLS) algorithm to obtain the ideal projection vector to be estimated.

Further, the step 1 uses a sinogram recovery method based on a penalized weighted least squares algorithm to recover sinogram data using a combination of low milliamps and sparse view protocols.

The corrected projection data is reconstructed by the following PWLS algorithm to obtain P. Perform PWLS-TV reconstruction to obtain μ, although the result of this reconstruction step contains considerable noise, it also contains most of the structural information of the CT image.

① The PWLS method updates the image from the sinogram y

②PWLS-TV method image is updated from P

y represents the obtained sinogram data (system calibration and log-transformed projection), namely: y=(y ₁ , y ₂ ,...,y _M ) ^T ; P represents the ideal projection vector to be estimated; β is A hyperparameter used to balance fidelity; where H represents the projection matrix system; n is the total number of pixels in the image; R(P) is the penalty for image roughness. Traditional image roughness measurements are based on the difference in brightness between adjacent pixels.

Further, described step 2 uses the total variation (PWLS-TV) method based on penalized weighted least squares to use the recovered sinogram data for reconstructing the image, and then the non-two-dimensional variation of the penalized weighted least squares total variation method A secondary penalty function performs neighborhood patches instead of single-pixel regularization (PWLS-PR).

Patch-based regularization is performed on the non-quadratic penalty function of the reconstructed data through PWLS-TV.

The calculation formula of ω _jk (μ ⁿ ) is as follows

in,

is the updated function based on patch regularization, ω _jk (μ ⁿ ) is a weighting factor related to the distance between pixel j and pixel k in the neighborhood N _j , μ ⁿ is the current objective function,

Pixels j and k of the current objective function.

Further, the step 3 uses a regularization algorithm to reconstruct the regularized image.

Further, in step 4, a sparse matrix is constructed, and the matrix is normalized to obtain a normalized kernel matrix; then guided kernel filtering is performed under the guidance of the reconstructed image to obtain a restored image. Although smoothing can effectively remove noise, it also loses some details, resulting in the loss of details in the reconstructed image. Therefore, the guided kernel filtering method is used to restore the lost structural information, so that the denoised image retains the image details well and the structure is clearer.

Further, the matrix is constructed using the k-nearest neighbor algorithm.

Further, when calculating the image roughness between adjacent pixels in the penalized weighted least squares total variation method, a patch associated with each pixel is used.

The k-nearest neighbor algorithm, which is widely used in machine learning, is used to construct sparse matrices. The kNN side finds k similar neighbors for each pixel, and normalizes the matrix to get the normalized kernel matrix K. Then GK filtering (guided kernel filtering) is performed under the guidance of the reconstructed image _μTV to obtain the final image x

x=K· _μTV

where K is the kernel matrix, μTV is the PWLS- _TV method image update from P

Further, the roughness function of the patch is expressed as:

A PWLS (penalized weighted least squares) based sinogram recovery method is used to recover sinogram data using a combination of low milliamps and sparse view protocols. The recovered sinogram data is then used to reconstruct the image using a (penalized weighted least squares-based total variation) PWLS-TV method. PWLS sinogram restoration as a preprocessing step is effective for image reconstruction with a combination of low mas and sparse views.

A patch-based regularization method for iterative image reconstruction that uses neighborhood patches instead of individual pixels when computing its non-quadratic penalty function. Compared with traditional pixel-based regularization methods, this regularization method is more robust in distinguishing between random fluctuations and sharp edges caused by noise.

Figure 2 shows the XCAT Phantom reconstructed by different algorithms; it can be seen from Figure 2 that the method of the present application can well denoise and remove artifacts and preserve the detailed structure of the image, which can effectively improve the peak signal-to-noise ratio and structural similarity of the image At the same time, the image detail information can be recovered to a certain extent.

Although the present application has been described above with reference to specific embodiments, it will be understood by those skilled in the art that many modifications may be made in the configuration and details disclosed herein within the spirit and scope of the present disclosure. The scope of protection of the present application is determined by the appended claims, and the claims are intended to cover all modifications encompassed by the equivalent literal meaning or scope of the technical features in the claims.

Claims

An image reconstruction method, characterized in that: the method comprises the following steps:

Step 1: reconstruct the corrected projection data; obtain reconstructed data;

Step 2: Regularize the reconstructed data to obtain a regularized image;

Step 3: reconstruct the regularized image to obtain a reconstructed image;

Step 4: Take the reconstructed image as prior information, and perform guided kernel filtering under the guidance of the reconstructed image to obtain a restored image.
The image reconstruction method according to claim 1, characterized in that: in said step 1, the corrected projection data is reconstructed by a penalized weighted least squares algorithm to obtain an ideal projection vector to be estimated.
2. The image reconstruction method of claim 2, wherein said step 1 uses a sinogram recovery method based on penalized weighted least squares algorithm to recover sinogram data using a combination of low milliamp second and sparse view protocols.
The image reconstruction method according to claim 1, wherein in said step 2, the recovered sinogram data is used to reconstruct the image using a total variation method based on penalized weighted least squares, and then the penalized weighted least squares is used to reconstruct the image. The non-quadratic penalty function of the total variation method performs regularization of neighborhood patches instead of individual pixels.
The image reconstruction method according to claim 1, wherein in step 3, a regularization algorithm is used to reconstruct the regularized image.
The image reconstruction method according to claim 1, wherein: the step 4 constructs a sparse matrix, normalizes the matrix, and obtains a normalized kernel matrix; and then conducts guidance under the guidance of the reconstructed image Kernel filtering to get the restored image.
The image reconstruction method according to claim 6, wherein the matrix is constructed by a k-nearest neighbor algorithm.
The image reconstruction method according to claim 4, wherein: when calculating the image roughness between adjacent pixels in the penalized weighted least squares total variation method, a patch related to each pixel is used .
The image reconstruction method according to claim 8, wherein the roughness function of the patch is expressed as:

where U(μ) is the patch-based roughness function, f j (μ), f k (μ) are feature vectors consisting of the intensity values of all pixels in the center of the patch at pixel j and pixel k, respectively, and h is A positive weighting factor equal to the normalized inverse spatial distance between pixel j and pixel k, ψ(||f j (μ)-f k (μ)|| h ) is a non-quadratic penalty function.
An application of an image reconstruction method, characterized in that the image reconstruction method according to any one of claims 1 to 9 is applied to image noise reduction and artifact removal or image fusion.