Image reconstruction method, system, electronic device, and medium

Abstract

This invention discloses an image reconstruction method, system, electronic device, and medium. The image reconstruction method includes: acquiring a first image and a second image; the first image and the second image contain at least one identical tissue structure; and at least a portion of the second image has better sharpness than at least a portion of the first image; and acquiring a target image based on the first image and the second image using an image fusion method. This invention overcomes the defects of high image noise and strong graininess in soft tissue areas, enabling the reconstructed image to have a clear display of detailed structures, while also meeting the display requirements of soft tissue areas for clinical diagnosis.

Description

Technical Field

[0001] This invention relates to the field of image technology, and in particular to an image reconstruction method, system, electronic device, and medium. Background Technology

[0002] When reconstructing images of soft tissue commonly used in medicine, the edges of some high-attenuation materials are often blurred, such as the edges of stents and calcified areas, resulting in halo artifacts, also known as non-metallic artifacts, which can affect doctors' judgment on whether the blood vessel lumen is narrowed.

[0003] To address these issues, sharp, enhancing convolutional kernels are used to reconstruct soft tissue images. While this reconstruction method produces clear edges of highly attenuated materials and virtually eliminates halo artifacts, it often results in increased image noise and enhanced graininess in soft tissue areas, which does not meet the diagnostic needs of doctors. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the defects of high image noise and strong graininess in soft tissue areas in the prior art, and to provide an image reconstruction method, system, electronic device and medium.

[0005] The present invention solves the above-mentioned technical problems through the following technical solution:

[0006] This invention provides an image reconstruction method, the image reconstruction method comprising:

[0007] Obtain the first image and the second image;

[0008] The first image and the second image contain at least one identical tissue structure;

[0009] Furthermore, at least a portion of the second image has better sharpness compared to at least a portion of the first image;

[0010] The target image is obtained by image fusion based on the first image and the second image.

[0011] Preferably, the steps for acquiring the second image include:

[0012] The region of interest is segmented from the first image, and the region of interest is sharpened to generate the second image.

[0013] Preferably, the step of acquiring the second image further includes:

[0014] Obtain the third image;

[0015] The third image is smoothed to obtain the first image;

[0016] Segment the second image from the third image.

[0017] Preferably, the step of obtaining the target image by image fusion further includes:

[0018] Obtain the smoothed region of interest image corresponding to the second image;

[0019] The second image and the smoothed region of interest image corresponding to the second image are fused to generate the target sub-image;

[0020] A target image is generated based on the target sub-image and the first image.

[0021] Preferably, the step of sharpening the region of interest image includes:

[0022] The region of interest is deconvolved using a point spread function.

[0023] Preferably, the step of segmenting the region of interest image from the first image includes:

[0024] A deep learning network is used to extract the region of interest from the first image;

[0025] Alternatively, image segmentation can be used to extract the region of interest from the first image.

[0026] Preferably, the image reconstruction method includes:

[0027] Obtain the third image;

[0028] The third image is input into the network model to generate a detail-preserving denoised image;

[0029] The network model is generated by training a detail-preserving smoothing network using target images generated by any of the aforementioned image reconstruction methods as training data.

[0030] The present invention also provides an image reconstruction system, the image reconstruction system comprising:

[0031] The first acquisition module is used to acquire the first image and the second image;

[0032] The first image and the second image contain at least one identical tissue structure;

[0033] Furthermore, at least a portion of the second image has better sharpness compared to at least a portion of the first image;

[0034] The fusion module is used to obtain a target image based on the first image and the second image by means of image fusion.

[0035] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the image reconstruction method as described above.

[0036] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned image reconstruction method.

[0037] The positive and progressive effects of this invention are as follows:

[0038] This invention provides an image reconstruction method, system, electronic device, and medium. The image reconstruction method acquires a first image and a second image, wherein the first and second images contain at least one identical tissue structure; and at least a portion of the second image has better sharpness than at least a portion of the first image. Based on the first and second images, an image fusion method is used to acquire a target image, thereby completing image reconstruction. This overcomes the defects of high image noise and strong graininess in soft tissue areas, enabling the reconstructed image to have a clear display effect of detailed structure, while also meeting the display effect of soft tissue areas to meet the needs of clinical diagnosis. Attached Figure Description

[0039] Figure 1 This is a flowchart of the image reconstruction method of Embodiment 1 of the present invention;

[0040] Figure 2 This is a schematic diagram of the images before and after smoothing in Embodiment 1 of the present invention;

[0041] Figure 3 This is a first schematic diagram of cardiac image reconstruction in Embodiment 1 of the present invention;

[0042] Figure 4 This is a second schematic diagram of cardiac image reconstruction in Embodiment 1 of the present invention;

[0043] Figure 5 This is a flowchart of step S104 in embodiment 1 of the present invention;

[0044] Figure 6 This is a comparison image of the bilateral filtering process before and after implementation 1 of the present invention;

[0045] Figure 7 This is a comparison image of the image before and after reconstruction in Embodiment 1 of the present invention;

[0046] Figure 8 This is a schematic diagram of the mask image of the region of interest in Embodiment 1 of the present invention;

[0047] Figure 9 This is a schematic diagram of network model training in Embodiment 1 of the present invention;

[0048] Figure 10 This is a schematic diagram of the image reconstruction system of Embodiment 2 of the present invention;

[0049] Figure 11 This is a schematic diagram of the structure of the electronic device in Embodiment 3 of the present invention. Detailed Implementation

[0050] The present invention will be further illustrated by way of embodiments below, but the present invention is not limited to the scope of the embodiments described herein.

[0051] Example 1

[0052] like Figure 1 As shown, this embodiment discloses an image reconstruction method, which includes:

[0053] Step S101: Obtain the first image and the second image;

[0054] The first image and the second image contain at least one identical tissue structure;

[0055] Furthermore, at least a portion of the second image has better sharpness compared to at least a portion of the first image;

[0056] In this scheme, the second image is a part of the first image or the third image, that is, the second image is a subset of the first image or the third image.

[0057] Specifically, the first image can be acquired in the following two ways:

[0058] The first method: First, acquire a third image; then, smooth the third image to obtain the second image.

[0059] In this solution, either a deep learning network can be used to smooth the third image to obtain the first image, or a common noise reduction algorithm can be used to smooth the third image to obtain the first image; the specific method used is not limited here. The deep learning network can be a CNN (Convolutional Neural Network), which is a network that learns from low-dose images to obtain high-dose images. In this invention, the network input is the image reconstructed by sharp convolutional kernels, and the network output is the first image.

[0060] In one specific embodiment, such as Figure 2 As shown, the heart image reconstructed by the sharp convolution kernel is smoothed by the deep learning network. The left image is the image under a soft tissue window width and level of [100 / 700], and the right image is the image under a window width and level of [300 / 1500]. It can be seen that the overall image noise level is reduced and the image is very smooth.

[0061] The second method is to obtain the first image through direct parsing and reconstruction.

[0062] Specifically, the first image can be obtained by using a smooth convolution kernel reconstruction method.

[0063] In one specific embodiment, for example, such as Figure 3 As shown, the heart images reconstructed by the smooth convolution kernel are shown in the left image, which is the image under a soft tissue window width and level of [100 / 700], and the right image, which is the image under a window width and level of [300 / 1500]. The detailed structure of the stent on the coronary artery is not clearly displayed.

[0064] In this scheme, the region of interest can be a cardiac stent, bone, etc.

[0065] In this solution, there are two ways to obtain the second image:

[0066] In one approach, a region of interest (ROI) image is segmented from a first image, and the ROI image is then sharpened to generate the second image.

[0067] In one specific embodiment, for example, such as Figure 4 As shown, the heart images reconstructed by sharp convolution kernels are shown in the left image, which is the image under a soft tissue window width and level of [100 / 700], and the right image, which is the image under a window width and level of [300 / 1500]. The detailed structure of the coronary stent is clearly displayed, but the soft tissue area has more noise and is not as smooth as the image reconstructed by the smooth convolution kernel.

[0068] In this approach, reconstruction typically uses filtered backprojection. Different convolution kernels can produce medical images with different styles. For cardiac protocols, smooth convolution kernels are generally used to reconstruct images, resulting in medical images used to diagnose cardiac soft tissue. A key characteristic is smooth soft tissue with low noise. However, detailed structures, such as stents, will be relatively blurry. Images reconstructed with sharp convolution kernels contain more information than those reconstructed with smooth convolution kernels, resulting in clearer detailed structures, but with higher noise levels. Therefore, these sharp kernels are not suitable for reconstructing conventional diagnostic images.

[0069] The second method involves first acquiring a third image; then smoothing the third image to obtain the first image; and finally segmenting the second image from the third image.

[0070] Step S102: Based on the first image and the second image, obtain the target image by image fusion.

[0071] This embodiment discloses an image reconstruction method. The image reconstruction method acquires a first image and a second image, wherein the first image and the second image contain at least one identical tissue structure; and at least a portion of the second image has better sharpness than at least a portion of the first image; and a target image is acquired based on the first image and the second image by image fusion, thereby completing image reconstruction. This overcomes the defects of high image noise and strong graininess in soft tissue areas, so that the reconstructed image can have a clear display effect of detailed structure, while ensuring that the display effect of soft tissue areas meets the needs of clinical diagnosis.

[0072] In one implementable manner, step S102 includes:

[0073] The first image and the second image are fused together to obtain the target image.

[0074] In a specific embodiment, for example, taking a cardiac stent as the region of interest, the first image and the second image can be fused to generate a target image according to the following formula:

[0075]

[0076] M Stent This is a mask image of the region of interest; 1 indicates that the pixel is a scaffold region, and 0 indicates it is not. T stent It is the threshold for determining whether a pixel is a support; R cor This refers to the coronary artery region. In this embodiment, it can be first shown in the first image I. Ori The coronary artery region is extracted, and then the stent mask image is extracted by threshold segmentation.

[0077] W stent (i,j)=M Stent (i,j)*G(σ)

[0078] W stent (i,j)=(W stent (i,j)-max(W stent (i,j))) / (max(W stent (i,j))-min(W stent (i,j)))

[0079] Wstent refers to the weight coefficient of the scaffold, which is obtained by filtering the scaffold mask image through a Gaussian filter and then normalizing the weight coefficient to between 0 and 1.

[0080] I Out (i,j)=W stent (i,j)*I f (i,j)+(1-Wstent (i,j))*I Ori (i,j)

[0081] I f (i,j) is the first image after bilateral filtering, I Out The target image is the improved image with enhanced bracket resolution.

[0082] This solution achieves image reconstruction by directly fusing the first and second images to obtain the target image. It enables the acquisition of an image with low noise and clear tissue in a single reconstruction, thus overcoming the defects of high image noise and strong graininess in soft tissue areas. The reconstructed image can display clear details and structures, while meeting the display requirements of soft tissue areas for clinical diagnosis.

[0083] like Figure 5 As shown, in one implementable embodiment, step S102 further includes:

[0084] Step S1021: Obtain the smoothed region of interest image corresponding to the second image;

[0085] This approach uses very sharp convolution kernels to obtain high-resolution detailed boundary structures, but it also significantly increases image noise, which is detrimental to subsequent image fusion. Therefore, a bilateral filter is needed to filter the third image. The bilateral filter's function is to preserve image boundaries while removing background noise. Before filtering, the third image is cropped into smaller sub-images based on the region of interest. For example... Figure 6 As shown, the left image is a sub-image of the third image, and the right image is a sub-image after bilateral filtering. The details in the region of interest are basically preserved, while other regions have obvious smoothing effects.

[0086] Step S1022: Fuse the second image and the smoothed region of interest image corresponding to the second image to generate a target sub-image;

[0087] Step S1023: Generate a target image based on the target sub-image and the first image.

[0088] Specifically, the generated target sub-image can be backfilled into the first image to generate the target image; alternatively, the generated target sub-image can replace the corresponding region image in the first image to generate the target image. The specific method used to generate the target image can be selected based on the actual application and is not limited here.

[0089] In this plan, such as Figure 7As shown, the left image is the image with a smoothed convolution kernel, and the right image is the generated target image. The resolution of the detailed structure in the region of interest is significantly improved, and other regions have better noise reduction.

[0090] This solution involves acquiring a smoothed region of interest image corresponding to a second image, fusing the second image and the smoothed region of interest image corresponding to the second image to generate a target sub-image, and then generating a target image based on the target sub-image and the first image, thereby completing image reconstruction. This overcomes the defects of high image noise and strong graininess in soft tissue areas, enabling the reconstructed image to have a clear display effect of detailed structure, while also meeting the display effect of soft tissue areas to meet the needs of clinical diagnosis.

[0091] In one feasible approach, the step of sharpening the region of interest image includes:

[0092] The region of interest is deconvolved using a point spread function.

[0093] Specifically, the spatial resolution from the first image to the third image is determined by the point spread function. By using the point spread function to deconvolve the first image, the effect of the point spread function can be reduced. Deconvolution is widely used to improve the resolution of CT images.

[0094] f(x,y)=h(x,y)*g(x,y)+n(x,y)

[0095] f(x,y) is the output image of the system, h(x,y) is the impulse response of the system function, i.e. the point spread function, g(x,y) represents the true image, and n(x,y) represents noise.

[0096] The deconvolution method described here will not be described in detail, as it is not part of the innovative content of this invention. It is merely an illustration of one way to obtain a third image from a smoothed (blurred) image.

[0097] This scheme achieves sharpening of the region of interest by using a point spread function to perform deconvolution on the region of interest, thereby ensuring the effectiveness of subsequent image reconstruction.

[0098] In one feasible manner, step S101 specifically includes:

[0099] A deep learning network is used to extract the region of interest from the first image;

[0100] like Figure 8 The image displayed is a mask image of the region of interest extracted using a deep learning network. In this approach, the image with manually marked locations of interest is used as the target image, and then the network is trained to learn the target of interest.

[0101] This solution uses a deep learning network to extract the region of interest (ROI) from the first image, which improves the accuracy of ROI extraction and thus ensures the effectiveness of subsequent image reconstruction.

[0102] In one feasible embodiment, step S101 further includes:

[0103] The region of interest in the first image is extracted using image segmentation.

[0104] Specifically, taking a region of interest that can be a cardiac stent as an example, the first image is segmented to obtain a cardiac stent mask image of the first image. Optionally, the first image can be input into a preset image segmentation model to perform image segmentation to obtain the cardiac stent mask image in the first image. Alternatively, the first image can be segmented according to an image segmentation algorithm, such as a threshold-based segmentation method, a region-based segmentation method, or an edge-based segmentation method, to obtain the cardiac stent in the first image.

[0105] This scheme extracts the region of interest (ROI) of the first image through image segmentation, which improves the accuracy of ROI extraction and thus ensures the effectiveness of subsequent image reconstruction.

[0106] In one implementable manner, the image reconstruction method further includes:

[0107] Obtain the third image;

[0108] The third image is input into the network model to generate a detail-preserving denoised image; the network model is generated by training the detail-preserving smoothing network with the target image generated by the image reconstruction method described above as training data.

[0109] Specifically, such as Figure 9 As shown, the target image generated by the above image reconstruction method is used as training data to train a detail-preserving denoising network, also known as a detail-preserving smoothing network, which can be a convolutional neural network. For example, using the target image as the training set can serve as the gold standard for network training, enabling the training of a denoising network with strong denoising effects on soft tissues while preserving structure in the region of interest, thus greatly improving the efficiency of image processing.

[0110] Example 2

[0111] like Figure 10 As shown, this embodiment discloses an image reconstruction system, which includes:

[0112] The first acquisition module 1 is used to acquire the first image and the second image;

[0113] The first image and the second image contain at least one identical tissue structure;

[0114] Furthermore, at least a portion of the second image has better sharpness compared to at least a portion of the first image;

[0115] In this scheme, the second image is a part of the first image or the third image, that is, the second image is a subset of the first image or the third image.

[0116] Specifically, the first image can be acquired in the following two ways:

[0117] The first method: First, acquire a third image; then, smooth the third image to obtain the second image.

[0118] In this solution, either a deep learning network can be used to smooth the third image to obtain the first image, or a common noise reduction algorithm can be used to smooth the third image to obtain the first image; the specific method used is not limited here. The deep learning network can be a CNN, which is a network that learns from low-dose images to obtain high-dose images. In this invention, the network input is the image reconstructed by sharp convolutional kernels, and the network output is the first image.

[0119] In one specific embodiment, such as Figure 2 As shown, the heart image reconstructed by the sharp convolution kernel is smoothed by the deep learning network. The left image is the image under a soft tissue window width and level of [100 / 700], and the right image is the image under a window width and level of [300 / 1500]. It can be seen that the overall image noise level is reduced and the image is very smooth.

[0120] The second method is to obtain the first image through direct parsing and reconstruction.

[0121] Specifically, the first image can be obtained by using a smooth convolution kernel reconstruction method.

[0122] In one specific embodiment, for example, such as Figure 3 As shown, the heart images reconstructed by the smooth convolution kernel are shown in the left image, which is the image under a soft tissue window width and level of [100 / 700], and the right image, which is the image under a window width and level of [300 / 1500]. The detailed structure of the stent on the coronary artery is not clearly displayed.

[0123] In this scheme, the region of interest can be a cardiac stent, bone, etc.

[0124] In this solution, there are two ways to obtain the second image:

[0125] In one approach, a region of interest (ROI) image is segmented from a first image, and the ROI image is then sharpened to generate the second image.

[0126] In one specific embodiment, for example, such as Figure 4 As shown, the heart images reconstructed by sharp convolution kernels are shown in the left image, which is the image under a soft tissue window width and level of [100 / 700], and the right image, which is the image under a window width and level of [300 / 1500]. The detailed structure of the coronary stent is clearly displayed, but the soft tissue area has more noise and is not as smooth as the image reconstructed by the smooth convolution kernel.

[0127] In this approach, reconstruction typically uses filtered backprojection. Different convolution kernels can produce medical images with different styles. For cardiac protocols, smooth convolution kernels are generally used to reconstruct images, resulting in medical images used to diagnose cardiac soft tissue. A key characteristic is smooth soft tissue with low noise. However, detailed structures, such as stents, will be relatively blurry. Images reconstructed with sharp convolution kernels contain more information than those reconstructed with smooth convolution kernels, resulting in clearer detailed structures, but with higher noise levels. Therefore, these sharp kernels are not suitable for reconstructing conventional diagnostic images.

[0128] The second method involves first acquiring a third image; then smoothing the third image to obtain the first image; and finally segmenting the second image from the third image.

[0129] Fusion module 2 acquires the target image by fusion of the first image and the second image.

[0130] This embodiment discloses an image reconstruction system. The image reconstruction system acquires a first image and a second image, wherein the first image and the second image contain at least one identical tissue structure; and at least a portion of the second image has better sharpness than at least a portion of the first image; based on the first image and the second image, an image fusion method is used to acquire a target image, thereby completing image reconstruction. This overcomes the defects of high image noise and strong graininess in soft tissue areas, enabling the reconstructed image to have a clear display effect of detailed structure, while also meeting the display effect of soft tissue areas to meet the needs of clinical diagnosis.

[0131] In one feasible embodiment, the fusion module 2 is specifically used for:

[0132] The first image and the second image are fused together to obtain the target image.

[0133] In a specific embodiment, for example, taking a cardiac stent as the region of interest, the first image and the second image can be fused to generate a target image according to the following formula:

[0134]

[0135] M Stent This is a mask image of the region of interest; 1 indicates that the pixel is a scaffold region, and 0 indicates it is not. T stent It is the threshold for determining whether a pixel is a support; R cor This refers to the coronary artery region. In this embodiment, it can be first shown in the first image I. Ori The coronary artery region is extracted, and then the stent mask image is extracted by threshold segmentation.

[0136] W stent (i,j)=M Stent (i,j)*G(σ)

[0137] W stent (i,j)=(W stent (i,j)-max(W stent (i,j))) / (max(W stent (i,j))-min(W stent (i,j)))

[0138] Wstent refers to the weight coefficient of the scaffold, which is obtained by filtering the scaffold mask image through a Gaussian filter and then normalizing the weight coefficient to between 0 and 1.

[0139] I Out (i,j)=W stent (i,j)*I f (i,j)+(1-W stent (i,j))*I Ori (i,j)

[0140] I f (i,j) is the first image after bilateral filtering, I Out The target image is the improved image with enhanced bracket resolution.

[0141] This solution achieves image reconstruction by directly fusing the first and second images to obtain the target image. It enables the acquisition of an image with low noise and clear tissue in a single reconstruction, thus overcoming the defects of high image noise and strong graininess in soft tissue areas. The reconstructed image can display clear details and structures, while meeting the display requirements of soft tissue areas for clinical diagnosis.

[0142] In one feasible embodiment, the fusion module 2 is further used for:

[0143] Obtain the smoothed region of interest image corresponding to the second image;

[0144] This approach uses very sharp convolution kernels to obtain high-resolution detailed boundary structures, but it also significantly increases image noise, which is detrimental to subsequent image fusion. Therefore, a bilateral filter is needed to filter the third image. The bilateral filter's function is to preserve image boundaries while removing background noise. Before filtering, the third image is cropped into smaller sub-images based on the region of interest. For example... Figure 6 As shown, the left image is a sub-image of the third image, and the right image is a sub-image after bilateral filtering. The details in the region of interest are basically preserved, while other regions have obvious smoothing effects.

[0145] The second image and the smoothed region of interest image corresponding to the second image are fused to generate the target sub-image;

[0146] A target image is generated based on the target sub-image and the first image.

[0147] Specifically, the generated target sub-image can be backfilled into the first image to generate the target image; alternatively, the generated target sub-image can replace the corresponding region image in the first image to generate the target image. The specific method used to generate the target image can be selected based on the actual application and is not limited here.

[0148] In this plan, such as Figure 7 As shown, the left image is the image with a smoothed convolution kernel, and the right image is the generated target image. The resolution of the detailed structure in the region of interest is significantly improved, and other regions have better noise reduction.

[0149] This solution involves acquiring a smoothed region of interest image corresponding to a second image, fusing the second image and the smoothed region of interest image corresponding to the second image to generate a target sub-image, and then generating a target image based on the target sub-image and the first image, thereby completing image reconstruction. This overcomes the defects of high image noise and strong graininess in soft tissue areas, enabling the reconstructed image to have a clear display effect of detailed structure, while also meeting the display effect of soft tissue areas to meet the needs of clinical diagnosis.

[0150] In one feasible embodiment, the first acquisition module 1 is further configured to:

[0151] The region of interest is deconvolved using a point spread function.

[0152] Specifically, the spatial resolution from the first image to the third image is determined by the point spread function. By deconvolving the first image with the point spread function, the effect of the point spread function can be reduced. Deconvolution is widely used to improve the resolution of CT images.

[0153] f(x,y)=h(x,y)*g(→,y)+n(x,y)

[0154] f(x,y) is the system's output image, h(x,y) is the impulse response of the system function, i.e., the point spread function, g(x,y) represents the true image, and n(x,y) represents noise. The deconvolution method used here is not described in detail and is not part of the invention's innovation; it merely illustrates one method of obtaining a third image from a smoothed (blurred) image.

[0155] This scheme achieves sharpening of the region of interest by using a point spread function to perform deconvolution on the region of interest, thereby ensuring the effectiveness of subsequent image reconstruction.

[0156] In one feasible approach, the first acquisition module 1 is specifically used for:

[0157] A deep learning network is used to extract the region of interest from the first image;

[0158] like Figure 8 The image displayed is a mask image of the region of interest extracted using a deep learning network. In this approach, the image with manually marked locations of interest is used as the target image, and then the network is trained to learn the target of interest.

[0159] This solution uses a deep learning network to extract the region of interest (ROI) from the first image, which improves the accuracy of ROI extraction and thus ensures the effectiveness of subsequent image reconstruction.

[0160] In one feasible embodiment, the first acquisition module 1 is further configured to:

[0161] The region of interest in the first image is extracted using image segmentation.

[0162] Specifically, taking a region of interest that can be a cardiac stent as an example, the first image is segmented to obtain a cardiac stent mask image of the first image. Optionally, the first image can be input into a preset image segmentation model to perform image segmentation to obtain the cardiac stent mask image in the first image. Alternatively, the first image can be segmented according to an image segmentation algorithm, such as a threshold-based segmentation method, a region-based segmentation method, or an edge-based segmentation method, to obtain the cardiac stent in the first image.

[0163] This scheme extracts the region of interest (ROI) of the first image through image segmentation, which improves the accuracy of ROI extraction and thus ensures the effectiveness of subsequent image reconstruction.

[0164] In one implementable manner, the image reconstruction system further includes:

[0165] The second acquisition module 3 is used to acquire the third image;

[0166] The target image generation module 4 is used to input the third image into the network model to generate a detail-preserving and noise-reduced image; the network model is generated by training the detail-preserving and smoothing network with the target image generated by the image reconstruction method described above as training data.

[0167] Specifically, the target image generated by the above image reconstruction method is used as training data to train a detail-preserving denoising network, also known as a detail-preserving smoothing network. The detail-preserving smoothing network can be a convolutional neural network. For example, using the target image as the training set can serve as the gold standard for network training, training a network that has a strong denoising effect on soft tissues and preserves the structure of the region of interest, thus greatly improving the efficiency of image processing.

[0168] Example 3

[0169] Figure 11 This is a schematic diagram of an electronic device provided in Embodiment 3 of the present invention. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the image reconstruction method provided in Embodiment 1. Figure 11 The electronic device 40 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0170] like Figure 11 As shown, the electronic device 40 can be manifested as a general-purpose computing device, such as a server device. The components of the electronic device 40 may include, but are not limited to: at least one processor 41, at least one memory 42, and a bus 43 connecting different system components (including memory 42 and processor 41).

[0171] Bus 43 includes a data bus, an address bus, and a control bus.

[0172] The memory 42 may include volatile memory, such as random access memory (RAM) 421 and / or cache memory 422, and may further include read-only memory (ROM) 423.

[0173] The memory 42 may also include a program / utility 425 having a set (at least one) of program modules 424, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0174] The processor 41 executes various functional applications and data processing by running computer programs stored in the memory 42, such as the image reconstruction method provided in Embodiment 1 of the present invention.

[0175] Electronic device 40 can also communicate with one or more external devices 44 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 45. Furthermore, the model-generated device 40 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 46. As shown, network adapter 46 communicates with other modules of the model-generated device 40 via bus 43. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated device 40, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0176] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0177] Example 4

[0178] This embodiment provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the image reconstruction method provided in Embodiment 1.

[0179] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0180] In a possible implementation, the present invention can also be implemented as a program product comprising program code, which, when the program product is run on a terminal device, causes the terminal device to execute the image reconstruction method provided in Embodiment 1.

[0181] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0182] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but all such changes and modifications fall within the scope of protection of the present invention.

Claims

1. A medical image reconstruction method characterized by, The image reconstruction method, applied to a computed tomography imaging system, includes: Obtain the first image and the second image; Wherein, the first image is a computed tomography (CT) medical image, and the second image is obtained by segmenting the region of interest (ROI) from the first image and sharpening the ROI; the ROI includes high-attenuation material, which includes a cardiac stent or bone. The first image and the second image contain at least one identical tissue structure; Furthermore, at least a portion of the second image has better sharpness compared to at least a portion of the first image; Based on the first image and the second image, an image fusion method is used to obtain the target image, including: Extracting the coronary artery region from the first image and extracting the mask image of the high-attenuation material through threshold segmentation includes: If a pixel in the first image is located within the coronary artery region and its pixel value is greater than or equal to the threshold for determining the high-attenuation substance, then the value of the corresponding pixel in the mask image is 1; otherwise, it is 0. The initial weight values ​​are obtained by Gaussian filtering based on the mask image. The first difference is obtained by subtracting the maximum value among all the initial weight values ​​from the initial weight values. The second difference is obtained by subtracting the minimum value from the maximum value among all the initial weight values. The quotient is used as the weight coefficient of the high-attenuation substance. The target image is calculated based on the first image after bilateral filtering and the weighting coefficients of the high-attenuation substance, including: The value of each pixel in the target image is the sum of the first product value and the second product value: The first product value is obtained by multiplying the weight coefficient of the high-attenuation substance with the corresponding pixel value in the first image after bilateral filtering. The third difference obtained by subtracting the weighting coefficient from the value 1 is multiplied by the corresponding pixel value in the first image to obtain the second product value.

2. The medical image reconstruction method of claim 1, wherein, The steps for obtaining the second image also include: Obtain the third image; The third image is smoothed to obtain the first image; Segment the second image from the third image.

3. The medical image reconstruction method of claim 1, wherein, The step of obtaining the target image by image fusion also includes: Obtain the smoothed region of interest image corresponding to the second image; The second image and the smoothed region of interest image corresponding to the second image are fused to generate the target sub-image; A target image is generated based on the target sub-image and the first image.

4. The medical image reconstruction method as described in claim 1, characterized in that, The steps for sharpening the region of interest image include: The region of interest is deconvolved using a point spread function.

5. The medical image reconstruction method as described in claim 1, characterized in that, The steps for segmenting the region of interest from the first image include: A deep learning network is used to extract the region of interest from the first image; Alternatively, image segmentation can be used to extract the region of interest from the first image.

6. The medical image reconstruction method as described in claim 1, characterized in that, The image reconstruction method includes: Obtain the third image; The third image is input into the network model to generate a detail-preserving denoised image; The network model is generated by training a detail-preserving smoothing network using target images generated by the image reconstruction method as described in any one of claims 1 to 5 as training data.

7. A medical image reconstruction system, characterized in that, A medical image reconstruction system applied to computed tomography (CT) imaging systems includes: The first acquisition module is used to acquire the first image and the second image; Wherein, the first image is a computed tomography (CT) medical image, and the second image is generated by segmenting the region of interest (ROI) from the first image and sharpening the ROI; wherein, the ROI includes high-attenuation material, and the high-attenuation material includes a cardiac stent or bone; The first image and the second image contain at least one identical tissue structure; Furthermore, at least a portion of the second image has better sharpness compared to at least a portion of the first image; The fusion module is used to obtain a target image based on the first image and the second image by means of image fusion; The fusion module is also used for: Extracting the coronary artery region from the first image and extracting the mask image of the high-attenuation material through threshold segmentation includes: If a pixel in the first image is located within the coronary artery region and its pixel value is greater than or equal to the threshold for determining the high-attenuation substance, then the value of the corresponding pixel in the mask image is 1; otherwise, it is 0. The initial weight values ​​are obtained by Gaussian filtering based on the mask image. The first difference is obtained by subtracting the maximum value among all the initial weight values ​​from the initial weight values. The second difference is obtained by subtracting the minimum value from the maximum value among all the initial weight values. The quotient is used as the weight coefficient of the high-attenuation substance. The target image is calculated based on the first image after bilateral filtering and the weighting coefficients of the high-attenuation substance, including: The value of each pixel in the target image is the sum of the first product value and the second product value: The first product value is obtained by multiplying the weight coefficient of the high-attenuation substance with the corresponding pixel value in the first image after bilateral filtering. The third difference obtained by subtracting the weighting coefficient from the value 1 is multiplied by the corresponding pixel value in the first image to obtain the second product value.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the image reconstruction method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the image reconstruction method as described in any one of claims 1 to 6.

Images