CN116228915A

CN116228915A - Image reconstruction method, system and equipment based on region judgment

Info

Publication number: CN116228915A
Application number: CN202310518925.4A
Authority: CN
Inventors: 杨浩宇; 刘敏; 刘安琪; 邓美; 许文清; 司超增
Original assignee: China Japan Friendship Hospital
Current assignee: China Japan Friendship Hospital
Priority date: 2023-05-10
Filing date: 2023-05-10
Publication date: 2023-06-06
Anticipated expiration: 2043-05-10
Also published as: CN116228915B

Abstract

The invention relates to an image reconstruction method, an image reconstruction system and image reconstruction equipment based on region judgment. Comprising the following steps: acquiring a metal artifact image, and performing feature mapping and feature reconstruction on the metal artifact image based on the distribution features of the metal artifact-free image to obtain a preliminarily reconstructed metal artifact image; calculating a characteristic difference map of the preliminarily reconstructed metal artifact image and the metal artifact image; obtaining a metal artifact region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference graph; and (5) carrying out enhancement or inhibition treatment on the metal artifact region, and reconstructing to obtain a reconstruction result without metal artifacts. The method aims at removing the artifacts of the metal artifact region based on the distribution characteristics of the metal artifact-free image and the characteristic distribution conditions of the metal artifact-free image so as to discover the specific analysis capability and the potential application value of the characteristic distribution difference in image reconstruction.

Description

Image reconstruction method, system and equipment based on region judgment

Technical Field

The invention relates to the technical field of medical image analysis and reconstruction, in particular to an image reconstruction method, an image reconstruction system, image reconstruction equipment, a computer readable storage medium and application thereof based on region judgment.

Background

In CT examinations, whatever the scanning mode used, metal implants and the like in the patient's body can cause signal loss in the original X-ray projections, resulting in severe metal artifacts in the reconstructed CT images. These metal artifacts can affect the visual quality of the data, adversely affecting subsequent diagnostics. Currently, the methods for metal artifact mitigation (Metal Artifact Reduction, MAR) mainly include three types of physical impact correction, sinogram signal complementation, and iterative reconstruction. Physical impact correction is not useful for artifacts caused by certain types of metals and therefore is not versatile; although the sinogram signal complement can remove a part of metal artifacts, new interpolation artifacts can be introduced at the same time, so that the effect is poor; iterative reconstruction is optimized based on a manually designed canonical optimizer, which has the problem of poor generalization for a wide variety of imaging devices and different artifact types.

In recent years, with the approach of deep learning, metal artifact removal is mainly performed by using an original signal and/or complete consistent calculation is performed on the whole image area, but the original sinusoidal signal that needs to be used is difficult to obtain in actual clinical application, and the complete consistent calculation on the whole image area does not conform to a real MAR target in actual application and affects the overall effect. For example, when a filtered back-projection reconstruction algorithm is employed, artifacts may occur in the final reconstructed result because the fourier transform during filtering is a global treatment of the data. Therefore, how to effectively remove the metal artifact is a problem to be solved.

Disclosure of Invention

The present application provides an image reconstruction method, system, analysis device, computer readable storage medium and application thereof based on region judgment, which aims to solve the problem of metal artifact based on regional characteristic distribution difference, explore feasibility of application in image reconstruction, save labor and time cost of data processing, and improve data processing speed and metal artifact removing effect.

According to a first aspect of the present application, an embodiment of the present application provides an image reconstruction method based on region judgment, including: acquiring a metal artifact image, and performing feature mapping and feature reconstruction on the metal artifact image based on the distribution features of the metal artifact-free image to obtain a preliminarily reconstructed metal artifact image; calculating a characteristic difference map of the preliminarily reconstructed metal artifact image and the metal artifact image; obtaining a metal artifact region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference graph; and carrying out enhancement or inhibition treatment on the metal artifact region, and reconstructing to obtain a reconstruction result without metal artifacts.

Further, the metal-free artifact image distribution feature comprises any one or more of the following features: intensity distribution features, region distribution features, edge features, texture features, color features; preferably, the metal artifact free image distribution feature is an intensity distribution feature of a metal artifact free image.

Still further, the intensity distribution characteristics of the metal artifact free image are obtained by carrying out characteristic extraction and characteristic statistical analysis on the metal artifact free image through an external memory mechanism, the metal artifact free image is divided into a plurality of subareas through a grid division mode by the external memory mechanism, and then an intensity distribution histogram of each subarea is calculated.

Further, the feature map is implemented using one or more of the following algorithms: polynomial feature mapping, kernel method, sparse coding, self-encoder, convolutional neural network, recurrent neural network, generating countermeasure network.

Further, the preliminarily reconstructed metal artifact image further comprises being obtained by inputting the metal artifact image into a MAR reconstruction model, wherein the MAR reconstruction model is constructed based on the intensity distribution characteristics of the metal artifact-free image.

In some alternative embodiments, the process of constructing the MAR reconstruction model includes:

acquiring a metal artifact-free image, and performing feature mapping on the metal artifact-free image to obtain a metal artifact-free feature image;

extracting the intensity characteristic of the image without metal artifact to obtain intensity distribution characteristic;

performing similarity calculation on the intensity distribution characteristics and the metal artifact-free characteristic images, selecting areas with high global similarity to perform characteristic enhancement to obtain optimized characteristic images, and performing characteristic reconstruction on the optimized characteristic images to obtain preliminarily reconstructed metal artifact-free images;

And calculating the error of each pixel point between the preliminarily reconstructed metal artifact free image and the obtained metal artifact free image, performing self-supervision training according to the error, and optimizing to obtain a trained MAR reconstruction model.

Still further, in the process of constructing the MAR reconstruction model, the method further comprises the steps of inputting the intensity distribution characteristics into an external memory module for similarity calculation, and iterating the intensity distribution characteristics in the external memory module according to a preset threshold, wherein the external memory module is a memory dictionary which is obtained by carrying out characteristic calculation based on the metal artifact-free images and contains N representative metal artifact-free region intensity distribution characteristics.

In some alternative embodiments, the MAR reconstruction model comprises a feature mapping module, an intensity distribution calculation module, an external memory module, and a feature reconstruction module, wherein the modules are connected according to functions of the modules.

Further, the feature mapping module is used for performing feature mapping on the metal artifact free image to obtain a metal artifact free feature image; the intensity distribution calculation module is used for carrying out feature extraction on the metal artifact free image to obtain intensity distribution features; the external memory module is used for carrying out correlation calculation on the intensity distribution characteristic and the metal artifact free characteristic map, and carrying out iteration according to a preset threshold value to obtain an intensity distribution characteristic map; the feature reconstruction module is used for carrying out similarity calculation on the intensity distribution feature map and the metal artifact-free feature map, selecting a region with high global similarity to carry out feature enhancement of metal artifact-free feature to obtain an optimized feature map, carrying out feature reconstruction on the optimized feature map to obtain a preliminarily reconstructed metal artifact-free image, calculating errors of pixel points between the preliminarily reconstructed metal artifact-free image and the metal artifact-free image, carrying out self-supervision training according to the errors, and optimizing to obtain a trained MAR reconstruction model.

Further, the characteristic difference map includes any one or several of the following characteristics: intensity distribution features, region distribution features, edge features, texture features, color features; preferably, the characteristic difference map is an intensity distribution characteristic difference map.

Further, the metal artifact region refers to a region to be determined, suspected and having a high probability of containing metal artifacts, for example, the region is distinguished by intensity differences, and the metal artifact region refers to a region with large intensity distribution characteristic differences.

In some embodiments, the method further comprises:

acquiring a metal artifact image, and performing feature mapping and feature reconstruction on the metal artifact image based on the distribution features of the metal artifact-free image to obtain a preliminarily reconstructed metal artifact image;

calculating a characteristic difference map of the preliminarily reconstructed metal artifact image and the metal artifact image;

obtaining a metal artifact region and a normal region in a preliminarily reconstructed metal artifact image by comparing a difference value in a feature difference diagram with a preset feature threshold, wherein the preset feature threshold is a feature threshold preset based on different metal implant types;

according to the differential feature distribution condition of the metal artifact region, the classification judgment of the type of the metal artifact region is realized; and carrying out enhancement or inhibition treatment on the metal artifact region according to the classification result to obtain a reconstructed metal artifact region, and obtaining a reconstruction result without metal artifacts.

Further, the metal artifact region type includes any one or more of the following metal region types: metal implants, metal fillers, metal foreign bodies, metal instruments, metal steel plates or screws, metal catheters or stents, metal markers, metal fragments.

Still further, the method further comprises: obtaining a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference map, and obtaining a corresponding anatomical structure region image according to the characteristic distribution condition of the normal region; and carrying out enhancement or inhibition treatment on the metal artifact region to obtain a reconstructed metal artifact region, and reconstructing the reconstructed metal artifact region based on the anatomical structure region image to obtain a metal artifact-free reconstruction result.

Further, the corresponding anatomical structure region image is obtained according to the feature distribution condition of the normal region, the corresponding anatomical structure region image is generated by performing similarity calculation on the intensity distribution feature of the normal region and the intensity distribution feature in an external memory module, and the external memory module is a memory dictionary containing the intensity distribution features of the metal artifact free regions of N representative anatomical structure regions, which is obtained by performing feature calculation on the metal artifact free images of different anatomical structures.

Further, the enhancement or suppression process employs any one or more of the following algorithms: MAR suppression model, interpolation method, back projection method, least square method, least mean square error method, iterative optimization algorithm, support vector machine, deep learning, compressed sensing reconstruction algorithm, sparse representation reconstruction algorithm, markov random field model algorithm, wavelet transformation. The MAR suppression model is used for carrying out enhancement or suppression treatment on the metal artifact region to obtain a reconstructed metal artifact region, and then a normal region is integrated based on the reconstructed metal artifact region to obtain a reconstruction result without metal artifacts.

In some alternative embodiments, the MAR suppression model is obtained by calculating the pixel-by-pixel error between the preliminarily reconstructed metal artifact image and the original metal artifact image, and/or calculating the pixel-by-pixel error between the preliminarily reconstructed metal artifact image and the metal artifact-free image as a loss value, and then performing enhancement or suppression processing on the metal artifact region according to the loss value, and performing iterative training.

Still further, the MAR suppression model further includes classifying the metal artifact region types, inputting the metal artifact region into the multi-stream MAR suppression sub-model according to the classification result to perform enhancement or suppression treatment to obtain a reconstructed metal artifact region, and obtaining a metal artifact-free reconstruction result based on the reconstructed metal artifact region; the multi-stream MAR inhibitor sub-model is a plurality of MAR inhibitor models with different metal artifact region type differentiation based on a plurality of different metal artifact region type reconstructions.

According to a second aspect of the present application, an embodiment of the present application provides an image reconstruction system based on region judgment, which implements the above image reconstruction method based on region judgment when executed.

Further, the modular structure of the system includes: the device comprises an acquisition module, a characteristic calculation module, a region judgment module and an image reconstruction module.

Still further, an obtaining module is configured to obtain a metal artifact image, and perform feature mapping and feature reconstruction on the metal artifact image based on the distribution feature of the metal artifact-free image, so as to obtain a preliminarily reconstructed metal artifact image.

Wherein the metal-free artifact image distribution feature comprises any one or more of the following features: intensity distribution features, region distribution features, edge features, texture features, color features; preferably, the metal artifact free image distribution feature is an intensity distribution feature of a metal artifact free image. Specifically, the intensity distribution characteristics of the metal artifact-free image are obtained by carrying out characteristic extraction and characteristic statistical analysis on the metal artifact-free image through an external memory mechanism, the metal artifact-free image is divided into a plurality of subareas through a grid division mode by the external memory mechanism, and then an intensity distribution histogram of each subarea is calculated.

In some embodiments, the feature map is implemented using one or more of the following algorithms: polynomial feature mapping, kernel method, sparse coding, self-encoder, convolutional neural network, recurrent neural network, generating countermeasure network.

In a specific embodiment, the preliminarily reconstructed metal artifact image further comprises being obtained by inputting the metal artifact image into a MAR reconstruction model, which is constructed based on the intensity distribution characteristics of the metal artifact-free image. In the MAR reconstruction model construction process, the method further comprises the steps of inputting the intensity distribution characteristics into an external memory module for similarity calculation, and iterating the intensity distribution characteristics in the external memory module according to a preset threshold, wherein the external memory module is a memory dictionary which is obtained by carrying out characteristic calculation on the basis of the metal artifact-free images and contains N representative metal artifact-free region intensity distribution characteristics.

Still further, a feature calculation module is configured to calculate a feature difference map between the reconstructed metal artifact image and the metal artifact image.

Wherein the characteristic difference map comprises any one or more of the following characteristics: intensity distribution features, region distribution features, edge features, texture features, color features; preferably, the characteristic difference map is an intensity distribution characteristic difference map.

Still further, the region judgment module is configured to obtain a metal artifact region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the feature difference map.

The region judging module further comprises a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image by comparing a difference value in the feature difference graph with a preset feature threshold, wherein the preset feature threshold is a feature threshold preset based on different metal implant types; and then, according to the difference characteristic distribution condition of the metal artifact region, realizing classification judgment of the type of the metal artifact region.

In the specific implementation process, the classification result obtained by the classification judgment of the region judgment module is input to the image reconstruction module, and the image reconstruction module performs the enhancement or inhibition treatment of the metal artifact region according to the classification result to obtain a reconstructed metal artifact region, so as to obtain a reconstruction result without metal artifacts.

And the image reconstruction module is used for carrying out enhancement or inhibition treatment on the metal artifact region, and reconstructing to obtain a reconstruction result without metal artifacts.

Still further, the region judgment module and the image reconstruction module in the system further include: the region judging module obtains a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference map, and obtains a corresponding anatomical structure region image according to the characteristic distribution condition of the normal region; the region judgment module is used for carrying out enhancement or inhibition treatment on the metal artifact region to obtain a reconstructed metal artifact region, and carrying out fusion reconstruction on the reconstructed metal artifact region based on the anatomical structure region image to obtain a metal artifact-free reconstruction result.

According to a third aspect of the present application, an embodiment of the present application provides a computer apparatus, comprising: a memory and a processor; the memory is used for storing program instructions; the processor is used for calling program instructions, and when the program instructions are executed, the image reconstruction method based on the region judgment is realized.

According to a fourth aspect of the present application, an embodiment of the present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the above-described image reconstruction method based on region judgment.

According to a fifth aspect of the present application, an embodiment of the present application provides related applications thereof, mainly including:

the application of the device or the system in the detection and classification task of the metal artifact area; optionally, the region detection and classification includes classification prediction of normal regions, metal artifact regions, and regions of the relief structure.

The device or the system is beneficial to realizing more and larger-scale reconstruction without metal artifacts of the CT image with the metal artifacts by carrying out prediction and analysis on the metal artifact region based on the image with the metal artifacts and carrying out automatic correction and image reconstruction on the metal artifact region through inhibiting or enhancing treatment.

The invention provides an image reconstruction method, an image reconstruction system and image reconstruction equipment based on region judgment, which aim at carrying out feature mapping and feature reconstruction on an image with metal artifacts based on metal artifact-free image distribution features to obtain a preliminarily reconstructed metal artifact image; calculating a characteristic difference map of the preliminarily reconstructed metal artifact image and the original metal artifact image; according to the difference value distribution in the characteristic difference graph, a metal artifact region is obtained, and the metal artifact region is automatically inhibited or enhanced by the MAR inhibition model to remove the metal artifact, so that the image reconstruction of the image with the metal artifact is realized, the potential application value of the difference value distribution characteristic in the characteristic difference graph in removing the metal artifact is discovered, and the method has strong innovation.

The application has the advantages that:

1. the application creatively discloses a method for carrying out feature mapping and feature reconstruction on a metal artifact image based on a metal artifact-free image distribution feature to obtain a reconstructed metal artifact image, then carrying out difference value distribution calculation to obtain a metal artifact region in the reconstructed metal artifact image, and carrying out enhancement or inhibition on the metal artifact region so as to discover the specific analysis capability and potential application value of the difference value distribution feature in a metal artifact removal task, thereby objectively improving the precision and depth of data analysis;

2. The method comprises the steps of creatively constructing a MAR reconstruction model, realizing preliminary reconstruction of a metal artifact image based on image priori knowledge of distribution characteristics of a metal artifact-free image, then calculating a characteristic difference image of the preliminarily reconstructed metal artifact image and an original metal artifact image, simultaneously comparing intensity distribution characteristics of the preliminarily reconstructed metal artifact image and the metal artifact-free image to obtain intensity distribution differences, combining the characteristic difference image and the intensity distribution differences to obtain a metal artifact region, carrying out artifact removal processing on the metal artifact region through cross-correlation calculation in a MAR inhibition model to obtain a metal artifact-free reconstruction result, so as to increase characteristic differences between the metal artifact image and the original metal artifact image, reduce the intensity distribution differences between the metal artifact image and the metal artifact-free image, and providing an objective, rapid, convenient and sensitive intelligent detection and automatic artifact removal processing method for the metal artifact region;

3. the metal artifact region-based and/or the dig structure region-based metal artifact region prediction, judgment and metal artifact removal processing are creatively realized, and meanwhile, the metal artifact regions are finely classified into different metal implants, so that more sufficient support and potential application value are provided in the metal artifact removal scheme selection.

Related noun interpretation: the MAR reconstruction model represents a metal artifact reduction reconstruction model, the MAR suppression model represents a metal artifact suppression model, and the multi-stream MAR suppression sub-model represents a plurality of metal artifact suppression models constructed based on different metal artifact region types.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of image reconstruction based on region judgment according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a construction flow of a MAR reconstruction model according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of an image reconstruction process without metal artifacts according to an embodiment of the present invention;

fig. 4 is a schematic diagram of an image reconstruction device based on region judgment according to an embodiment of the present invention.

Detailed Description

In order to enable those skilled in the art to better understand the present invention, the following description will make clear and complete descriptions of the technical solutions according to the embodiments of the present invention with reference to the accompanying drawings.

In some of the flows described in the specification and claims of the present invention and in the above figures, a plurality of operations appearing in a particular order are included, but it should be clearly understood that the operations may be performed in other than the order in which they appear herein or in parallel, the sequence numbers of the operations such as S101, S102, etc. are merely used to distinguish between the various operations, and the sequence numbers themselves do not represent any order of execution. In addition, the flows may include more or fewer operations, and the operations may be performed sequentially or in parallel.

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments according to the invention without any creative effort, are within the protection scope of the invention.

The embodiment of the application provides an image reconstruction method, an image reconstruction system, image reconstruction equipment, a computer readable storage medium and application thereof based on region judgment. The corresponding training device based on the image reconstruction method based on the region judgment can be integrated in computer equipment, and the computer equipment can be a terminal or a server and other equipment. The terminal can be a smart phone, a tablet computer, a notebook computer, a personal computer and the like. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud storage, network services, cloud communication, middleware services, domain name services, security services, a content delivery network (Content Delivery Network, abbreviated as CDN), basic cloud computing services such as big data and an artificial intelligent platform. The terminal and the server may be directly or indirectly connected through wired or wireless communication, which is not limited herein.

Referring to fig. 1, fig. 1 is a flowchart of image reconstruction based on region determination according to an embodiment of the present invention, and specifically includes:

s101: and acquiring a metal artifact image, and performing feature mapping and feature reconstruction on the metal artifact image based on the distribution features of the metal artifact-free image to obtain a preliminarily reconstructed metal artifact image.

In one embodiment, the acquired metal artifact image comprises a CT image, an MRI image.

Further, in particular embodiments, the metal artifact free image distribution features comprise any one or more of the following features: intensity distribution features, region distribution features, edge features, texture features, color features. Preferably, the metal artifact free image distribution feature is an intensity distribution feature of a metal artifact free image.

Still further, the intensity distribution characteristics of the metal artifact free image are obtained by performing characteristic extraction and characteristic statistical analysis on the metal artifact free image through an external memory mechanism. The external memory mechanism divides the image without metal artifacts into a plurality of subareas in a grid division mode, and then calculates an intensity distribution histogram of each subarea.

Further, the feature mapping is implemented using one or more of the following algorithms: polynomial feature mapping, kernel method, sparse coding, self-encoder, convolutional neural network, recurrent neural network, generating countermeasure network.

And (3) polynomial feature mapping, namely mapping the input feature vector into a higher-dimensional polynomial space so as to better fit a nonlinear model, and mapping the features from low latitude to high latitude by adding some power-level features on the basis of original features or adding interaction features and polynomial features of original data.

The kernel method utilizes kernel functions to map, namely, a kernel function is found to improve the linear separability of data.

Sparse coding is an unsupervised learning method to find a set of "overcomplete" basis vectors that more efficiently represent sample data, and can represent input vectors as linear combinations of these basis vectors.

The self-encoder is an unsupervised learning algorithm that uses the encoder to map the input data to a low-dimensional space, and uses the decoder to map the vectors of the low-dimensional space back to the original space.

The convolutional neural network is a special feedforward neural network, is mainly used for processing data with a grid structure, mainly uses a convolutional layer and a pooling layer to extract the characteristics of an image, and then uses a full-connection layer to classify.

The cyclic neural network is a recurrent neural network, and is mainly used for processing sequence data, and the cyclic layer is used for processing the sequence data, combining the previous state and the current input information, and extracting the characteristics of the sequence.

Generating an antagonism network is a model that can generate specific distribution data.

Still further, feature reconstruction may be performed by an encoder to reconstruct the mapped features to obtain reconstructed metal artifact images. The distribution characteristics of the reconstructed metal artifact image are adaptively adjusted based on the distribution characteristics of the metal artifact-free image, and on the adjustment distribution, the distribution characteristics of the reconstructed metal artifact image can be adaptively adjusted according to the peak distribution conditions in the distribution characteristics of the metal artifact-free image. The adjustment method comprises an interpolation method, an adaptive thresholding method, an adaptive histogram equalization method and other image processing methods.

Further, the preliminarily reconstructed metal artifact image further comprises being obtained by inputting the metal artifact image into a MAR reconstruction model, and the MAR reconstruction model is constructed based on the intensity distribution characteristics of the metal artifact-free image.

In some alternative embodiments, the process of constructing the MAR reconstruction model includes: acquiring a metal artifact-free image, and performing feature mapping on the metal artifact-free image to obtain a feature image of the metal artifact-free image; extracting the intensity characteristic of the image without metal artifact to obtain intensity distribution characteristic; performing similarity calculation on the intensity distribution characteristics and the characteristic images, selecting a region with high global similarity for characteristic enhancement to obtain an optimized characteristic image, and performing characteristic reconstruction on the optimized characteristic image to obtain a preliminarily reconstructed metal artifact-free image; and calculating the error of each pixel point between the preliminarily reconstructed metal artifact free image and the obtained metal artifact free image, performing self-supervision training according to the error, and optimizing to obtain a trained MAR reconstruction model. The characteristic diagrams comprise an intensity distribution characteristic diagram, an edge intensity distribution characteristic diagram and an edge definition distribution characteristic diagram. The optimization method comprises data preprocessing, data augmentation, random weight attenuation, learning rate, countermeasure training, regularization, adamW optimizer and self-knowledge distillation.

Still further, in the process of constructing the MAR reconstruction model, the method further comprises the steps of inputting the intensity distribution characteristics into an external memory module for similarity calculation, and iterating the intensity distribution characteristics in the external memory module according to a preset threshold value, wherein the external memory module is a memory dictionary which is obtained by carrying out characteristic calculation based on the metal artifact-free images and contains N representative metal artifact-free region intensity distribution characteristics.

In a specific embodiment, as shown in fig. 2, a specific construction process of a MAR reconstruction model provided by an embodiment of the present invention includes: inputting the metal artifact free images into two modules respectively, wherein one module is an encoder, and removing redundant information of the metal artifact free images through an encoding process to obtain a feature map; the other is an intensity distribution calculation module, which divides the image without metal artifacts into a plurality of subareas by a grid division mode and calculates the intensity distribution of each subarea as the intensity distribution characteristic of the current area. And then inputting the intensity distribution characteristics into the external memory module, performing similarity calculation with the intensity distribution characteristics stored in the external memory module, and updating the content in the external memory module. Specifically, the external memory module is configured to maintain a memory dictionary of length K (including K intensity distribution patterns that are most representative). And then, carrying out correlation calculation on the intensity distribution characteristics and the characteristic map output by the encoder, selecting the region with the highest global similarity in the characteristic map, carrying out characteristic enhancement (to optimize the expression capability of the characteristics) on each region to obtain an optimized characteristic map, inputting the optimized characteristic map into a decoder to obtain a preliminarily reconstructed metal artifact free image, and optimizing by taking pixel-by-pixel errors between the preliminarily reconstructed metal artifact free image and the original image as loss values to obtain the MAR reconstruction model. Finally, after the MAR reconstruction model is trained, a MAR reconstruction model with the capability of reconstructing an image without metal artifacts and a memory dictionary containing K representative intensity distributions without metal artifacts are obtained as an external memory module.

In some alternative embodiments, the MAR reconstruction model comprises a feature mapping module, an intensity distribution calculation module, an external memory module, and a feature reconstruction module, which are connected according to the functions of the modules.

Further, the feature mapping module is used for carrying out feature mapping on the metal artifact-free image to obtain a metal artifact-free feature image; the intensity distribution calculation module is used for extracting the characteristics of the image without metal artifacts to obtain intensity distribution characteristics; the external memory module is used for carrying out correlation calculation on the intensity distribution characteristic and the metal artifact free characteristic map, and carrying out iteration according to a preset threshold value to obtain the intensity distribution characteristic map; the feature reconstruction module is used for carrying out similarity calculation on the intensity distribution feature map and the metal artifact-free feature map, selecting a region with high global similarity to carry out feature enhancement of metal artifact-free feature to obtain an optimized feature map, carrying out feature reconstruction on the optimized feature map to obtain a preliminarily reconstructed metal artifact-free image, calculating errors between the preliminarily reconstructed metal artifact-free image and the metal artifact-free image, carrying out self-supervision training according to the errors, and optimizing to obtain a trained MAR reconstruction model.

S102: and calculating a characteristic difference map of the preliminarily reconstructed metal artifact image and the metal artifact image.

Further, the feature difference map includes distribution conditions of each feature peak and each feature histogram.

In some specific embodiments, the feature difference map includes any one or more of the following features: intensity distribution features, region distribution features, edge features, texture features, color features.

In a preferred embodiment, the characteristic difference map is an intensity distribution characteristic difference map.

S103: and obtaining a metal artifact region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference graph.

Further, the variance value distribution includes individual variance value distribution cases of different metal artifact region types. According to the distribution condition of the difference values in the characteristic difference graph, the metal artifact region with the highest similarity in different metal artifact region types is matched, and the metal artifact region in the preliminarily reconstructed metal artifact image is obtained through judgment of a preset threshold value. The preset threshold is a characteristic threshold preset based on different metal artifact region types.

Still further, the different metal artifact region types include any one or more of the following metal types: metal implants, metal fillers, metal foreign bodies, metal instruments, metal steel plates or screws, metal catheters or stents, metal markers, metal fragments. The metal artifact region refers to a region to be determined, suspected and having a high probability of containing metal artifacts, for example, the region is distinguished by intensity differences, and the metal artifact region refers to a region with large intensity distribution characteristic differences.

In some specific embodiments, step S103 further includes obtaining a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image by comparing the difference value in the feature difference map with a preset feature threshold, then implementing classification judgment of the metal artifact region type according to the difference feature distribution condition of the metal artifact region, and then executing the step shown in S104, where the specifically executed operation is that the metal artifact region is enhanced or suppressed according to the classification result to obtain a reconstructed metal artifact region, so as to obtain a non-metal artifact reconstruction result. Wherein the preset feature threshold is a feature threshold preset based on different metal implant types.

Still further, in some embodiments, step S103 further comprises: obtaining a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference map, and obtaining a corresponding anatomical structure region image according to the characteristic distribution condition of the normal region; then, the step shown in S104 is executed, and specifically, the operation of performing enhancement or suppression processing on the metal artifact region to obtain a reconstructed metal artifact region, and reconstructing the reconstructed metal artifact region based on the anatomical region image to obtain a reconstruction result without metal artifacts.

The corresponding anatomical structure region image is obtained according to the characteristic distribution condition of the normal region by carrying out similarity calculation on the intensity distribution characteristics of the normal region and the intensity distribution characteristics in an external memory module, the corresponding anatomical structure region image is generated, and the external memory module is a memory dictionary containing the intensity distribution characteristics of the metal artifact free regions of N representative anatomical structure regions, which is obtained by carrying out characteristic calculation on the metal artifact free images of different anatomical structures.

S104: and (5) carrying out enhancement or inhibition treatment on the metal artifact region, and reconstructing to obtain a reconstruction result without metal artifacts.

In some embodiments, the enhancement or suppression of the metal artifact region employs any one or more of the following algorithms: MAR suppression model, interpolation method, back projection method, least square method, least mean square error method, iterative optimization algorithm, support vector machine, deep learning, compressed sensing reconstruction algorithm, sparse representation reconstruction algorithm, markov random field model algorithm, wavelet transformation.

The MAR suppression model is obtained by calculating the pixel-by-pixel error between the preliminarily reconstructed metal artifact image and the obtained metal artifact image and/or calculating the pixel-by-pixel error between the preliminarily reconstructed metal artifact image and the metal artifact-free image as a loss value, then carrying out enhancement or suppression treatment on the metal artifact region according to the loss value, and carrying out iterative training.

Interpolation is performed to interpolate a missing pixel by existing pixel point information. Common interpolation algorithms are nearest neighbor interpolation, bilinear interpolation, bicubic interpolation, lagrangian interpolation, etc.

Back projection methods, back projection-based reconstruction algorithms such as direct back projection methods, filtered back projection methods, and the like.

The least square method, the least mean square error method and the iterative optimization algorithm are all reconstruction algorithms based on the iterative algorithm. The image reconstruction algorithm based on the weighted least square method also comprises a polynomial weighted least square method, an exponential weighted least square method and the like.

Both support vector machine and deep learning are a reconstruction algorithm based on statistical learning.

The compressed sensing reconstruction algorithm and the sparse representation reconstruction algorithm are both reconstruction algorithms based on sparse representation. The compressed sensing algorithm compresses the acquired sparse signals and then restores the signals to the original complete state by using the reconstruction algorithm.

The Markov random field model algorithm models the image into a Markov random field model, and the method such as maximum likelihood estimation or maximum posterior probability estimation is utilized to estimate model parameters, so that image reconstruction is realized. The algorithm is applicable to images with spatial correlation.

The wavelet transformation decomposes the image into a plurality of wavelet coefficients with different scales and different frequencies, and the operations of denoising, compressing, enhancing and the like of the image are realized through the processing of the wavelet coefficients.

In a specific embodiment, the training process of the MAR inhibition model comprises: inputting the obtained metal artifact image into a MAR reconstruction model shown in fig. 2 to reconstruct to obtain a preliminarily reconstructed metal artifact image, and calculating a characteristic difference image of the preliminarily reconstructed metal artifact image and the obtained metal artifact image; and/or calculating regional intensity distribution based on the obtained metal artifact image, comparing the regional intensity distribution with a representative intensity distribution set in an external memory module to obtain intensity distribution difference, and judging to obtain the metal artifact region. Then, inputting the metal artifact area into an encoder to obtain an abnormal feature map; and inputting the internal related normal region into an encoder to obtain a normal feature map, performing cross-correlation calculation on the normal feature map and the abnormal feature map to realize feature enhancement, obtaining an enhanced feature map, inputting the enhanced feature map into a decoder to obtain a precisely predicted metal artifact region, performing enhancement or suppression treatment on the metal artifact region according to a cross-correlation calculation result, and iterating the metal artifact region to obtain a trained MAR suppression model.

Further, in the implementation process, the metal artifact region in the metal artifact image can be determined by combining the results of the feature difference graph and the intensity distribution difference, and the other regions are defined as normal regions.

In some alternative embodiments, the MAR suppression model further includes classifying the metal artifact region by metal artifact region type, inputting the metal artifact region into a multi-stream MAR suppression sub-model according to the classification result to perform enhancement or suppression processing to obtain a reconstructed metal artifact region, and then obtaining a metal artifact-free reconstruction result based on the reconstructed metal artifact region; the multi-stream MAR inhibitor sub-model is a plurality of MAR inhibitor models with metal artifact region type differentiation reconstructed based on a plurality of different metal artifact region types.

Still further, the construction process of the multi-stream MAR inhibitor sub-model includes two phases: the first stage, inputting a metal artifact region into an encoder of a MAR suppression model to obtain an abnormal feature map, inputting the abnormal feature map into a decoder to directly optimize to obtain coarse predictions (further, classification judgment of specific metals can be realized according to the abnormal feature map); and in the second stage, inputting the internal relevant normal region into an encoder to obtain a normal feature map (further, the normal feature map can be obtained based on the classification of the planning structural region, specifically, according to the feature distribution condition of the normal region, the normal feature map is analyzed to obtain a corresponding anatomical structure region image), performing cross-correlation calculation on the normal feature map and the abnormal feature map to realize feature enhancement, obtaining an enhanced feature map, inputting the enhanced feature map into a decoder to obtain fine prediction, and obtaining a built multi-stream MAR suppression sub-model according to the fine prediction result and loss optimization of the metal artifact-free image. Wherein the multi-stream MAR inhibitor sub-model further comprises different sub-models constructed based on different sub-region type classifications of the normal region (i.e. its corresponding region of the relief structure) and/or the metal artifact region.

In some embodiments, the cross-correlation calculation uses one or more of the following algorithms: correlation matrix, mutual information, chi-square test, lasso regression, ridge regression, principal component analysis, independent component analysis.

The correlation matrix is used to measure the correlation between a plurality of variables, and is generally calculated by using methods such as pearson correlation coefficient, kendall Tau correlation coefficient or Spearman class correlation coefficient.

Mutual information: the method is used for measuring the nonlinear relation between two variables, is applicable to any type of data, and can be used for feature selection or feature extraction.

Chi-square test is used to measure the correlation between two discrete variables, enabling feature selection or feature extraction.

Lasso regression is used for feature selection, the most relevant features being selected by sparsifying the target variable.

Ridge regression is used for feature selection, the most relevant feature being selected by smoothing the target variable.

Principal component analysis is used to extract the principal feature components of the data and the dimensionality reduction of the high-dimensional data, the most relevant features being extracted by projecting the high-dimensional data into a low-dimensional space.

Independent component analysis is used for dimension reduction and feature extraction, the most relevant features being extracted by decomposing the high-dimensional data into independent components.

Further, the specific calculations of the cross-correlation calculation include calculations of an internal style consistency loss and an external style consistency loss.

Still further, the internal style consistency loss calculation process includes: and for the normal region, carrying out sub-region division and sub-region type classification according to the normal region belonging to the same input image, selecting the sub-region with the same category as the metal artifact region, and carrying out consistency calculation on the sub-region with the corresponding region in the predicted image to obtain the internal style consistency loss.

Still further, the external style consistency loss calculation process includes: and for the metal artifact region, carrying out subregion division and subregion type classification on the metal artifact region, selecting subregions of the same category as the metal artifact region from the metal artifact-free image obtained from the outside, and carrying out consistency calculation on the subregions and the corresponding regions in the predicted image to obtain the consistency loss of the external style.

Still further, region division divides the image into a plurality of sub-regions by way of meshing.

Wherein, the classification of the subarea types is realized by any one or more of the following models: decision trees, logistic regression, k nearest neighbor, support vector machines, naive bayes, convolutional neural networks, recurrent neural networks, fully connected neural networks, attention models, long-term short-term memory networks, deepLab, YOLO, SSD, R-CNN, hopfield networks, and boltzmann machines.

In some embodiments, the above image reconstruction method based on region judgment further includes: acquiring a metal artifact image, and performing feature mapping and feature reconstruction on the metal artifact image based on the distribution features of the metal artifact-free image to obtain a preliminarily reconstructed metal artifact image; calculating a characteristic difference map of the preliminarily reconstructed metal artifact image and the metal artifact image; obtaining a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image by comparing a difference value in the feature difference map with a preset feature threshold value, wherein the preset feature threshold value is a feature threshold value preset based on different metal implant types; according to the differential feature distribution condition of the metal artifact region, the classification judgment of the type of the metal artifact region is realized; and carrying out enhancement or inhibition treatment on the metal artifact region according to the classification result to obtain a reconstructed metal artifact region, and obtaining a reconstruction result without metal artifacts.

In a specific embodiment, the above image reconstruction method based on region judgment further includes:

obtaining a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference map, and obtaining a corresponding anatomical structure region image according to the characteristic distribution condition of the normal region;

and carrying out enhancement or inhibition treatment on the metal artifact region to obtain a reconstructed metal artifact region, and reconstructing the reconstructed metal artifact region based on the anatomical structure region image to obtain a reconstruction result without metal artifacts.

Further, according to the characteristic distribution condition of the normal region, a corresponding anatomical structure region image is obtained, and the corresponding anatomical structure region image is generated by carrying out similarity calculation on the intensity distribution characteristics of the normal region and the intensity distribution characteristics in the external memory module. The external memory module is a memory dictionary which is obtained by carrying out feature calculation based on metal artifact free images of different anatomical structures and contains the intensity distribution features of metal artifact free areas of N representative anatomical structure areas.

In a more specific embodiment, a chest CT image is taken as an example, and can be roughly divided into three parts, i.e., an in vitro region, an in vivo lung-out region, and an in vivo lung region. In one aspect, the calculation of in vitro regions is redundant to MAR results; on the other hand, there is a significant difference in intensity distribution between the in vivo and in vitro regions and the in vivo region, which if the same MAR algorithm is performed will generally result in an average distribution under both intensity distributions, resulting in sub-optimal MAR results in both types of regions.

In the embodiment of the invention, a schematic image reconstruction flow without metal artifacts is provided as shown in fig. 3, a normal area and a metal artifact area are obtained after reconstruction and intensity distribution calculation are performed on an obtained image with metal artifacts, and then a corresponding multi-stream MAR inhibitor model is selected by an area judging module to perform metal artifact reduction, so that reconstruction and prediction without metal artifacts are realized. Specifically, the image reconstruction process without metal artifacts is as follows: firstly, acquiring an image with metal artifact, inputting the acquired image with metal artifact into a trained MAR reconstruction model to obtain a preliminarily reconstructed metal artifact image, and calculating a characteristic difference image of the reconstructed metal artifact image and an original image through an external memory module; meanwhile, the acquired image with the metal artifact is input into an intensity distribution calculation module to calculate the intensity distribution of the image with the metal artifact, and the intensity distribution is compared with a representative intensity distribution set in an external memory module to obtain intensity distribution differences. Then, by combining the results of both the feature difference map and the intensity distribution difference, a metal artifact region and a normal region in the metal artifact image are judged, wherein a region other than the metal artifact region is defined as a normal region. Then, inputting the metal artifact region, the normal region and the metal artifact-free image into a region judging module to obtain an internal related normal region and an external related normal region, inputting the obtained regions (including the metal artifact region) into a multi-stream MAR inhibitor model, reconstructing the regions together through an external memory module to obtain a reconstructed metal artifact region, and calculating the external style consistency loss and the internal style consistency loss respectively by combining the external related normal region and the internal related normal region to obtain a metal artifact-free reconstruction result, namely a reconstructed image.

The method is feasible for automatic metal artifact removal processing of the metal artifact image, shows that the characteristic mapping and characteristic reconstruction are carried out on the metal artifact image based on the distribution characteristic of the metal artifact-free image to obtain a reconstructed metal artifact image, the distribution characteristic of the metal artifact-free image is automatically learned to a certain extent, then a metal artifact area is obtained according to the characteristic difference image between the metal artifact-free image and/or the intensity distribution difference between the metal artifact-free image and the metal artifact-free image, the specific analysis capability and the potential application value of the difference value distribution in the artifact removal are discovered, the accuracy and the depth of data analysis are objectively improved, and the metal artifact removal processing of the metal artifact image is more rapidly, effectively and conveniently automatically realized.

The image reconstruction system based on region judgment provided by the embodiment of the invention comprises a computer program which, when executed by a processor, realizes the image reconstruction method based on region judgment, and comprises the following steps:

obtaining a metal artifact region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference graph;

the metal artifact area is enhanced or inhibited, and the reconstruction is carried out to obtain a reconstruction result without metal artifact

In some embodiments, the system performs the steps of further comprising: acquiring a metal artifact image, and performing feature mapping and feature reconstruction on the metal artifact image based on the distribution features of the metal artifact-free image to obtain a preliminarily reconstructed metal artifact image; calculating a characteristic difference map of the preliminarily reconstructed metal artifact image and the metal artifact image; obtaining a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image by comparing a difference value in the feature difference map with a preset feature threshold value, wherein the preset feature threshold value is a feature threshold value preset based on different metal implant types; according to the differential feature distribution condition of the metal artifact region, the classification judgment of the type of the metal artifact region is realized; and carrying out enhancement or inhibition treatment on the metal artifact region according to the classification result to obtain a reconstructed metal artifact region, and obtaining a reconstruction result without metal artifacts.

Further, the metal artifact region type includes any one or several of the following metal region types: metal implants, metal fillers, metal foreign bodies, metal instruments, metal steel plates or screws, metal catheters or stents, metal markers, metal fragments.

Still further, the system further comprises: obtaining a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference map, and obtaining a corresponding anatomical structure region image according to the characteristic distribution condition of the normal region; and carrying out enhancement or inhibition treatment on the metal artifact region to obtain a reconstructed metal artifact region, and reconstructing the reconstructed metal artifact region based on the anatomical structure region image to obtain a reconstruction result without metal artifacts.

Furthermore, the corresponding anatomical structure region image is obtained according to the characteristic distribution condition of the normal region, the corresponding anatomical structure region image is generated by carrying out similarity calculation on the intensity distribution characteristics of the normal region and the intensity distribution characteristics in an external memory module, and the external memory module is a memory dictionary containing the intensity distribution characteristics of the metal artifact free regions of N representative anatomical structure regions, which is obtained by carrying out characteristic calculation on the metal artifact free images of different anatomical structures.

Wherein the metal artifact free image distribution features comprise any one or more of the following features: intensity distribution features, region distribution features, edge features, texture features, color features; preferably, the metal artifact free image distribution feature is an intensity distribution feature of a metal artifact free image.

Specifically, the intensity distribution characteristics of the image without metal artifacts are obtained by carrying out characteristic extraction and characteristic statistical analysis on the image without metal artifacts through an external memory mechanism, the image without metal artifacts is divided into a plurality of subareas through a grid division mode through the external memory mechanism, and then an intensity distribution histogram of each subarea is calculated.

In some embodiments, the feature mapping is implemented using one or more of the following algorithms: polynomial feature mapping, kernel method, sparse coding, self-encoder, convolutional neural network, recurrent neural network, generating countermeasure network.

In a specific embodiment, the preliminarily reconstructed metal artifact image further comprises being obtained by inputting the metal artifact image into a MAR reconstruction model, the MAR reconstruction model being constructed based on the intensity distribution characteristics of the metal artifact-free image. In the MAR reconstruction model construction process, the method further comprises the steps of inputting the intensity distribution characteristics into an external memory module for similarity calculation, iterating the intensity distribution characteristics in the external memory module according to a preset threshold value, and obtaining a memory dictionary containing N representative intensity distribution characteristics of the metal artifact-free region by the external memory module based on the metal artifact-free image for characteristic calculation.

The characteristic difference graph comprises distribution conditions of each characteristic peak and each characteristic histogram. Specifically, the features include an intensity distribution feature, an edge intensity distribution feature, and an edge sharpness distribution feature.

The region judging module further comprises a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image by comparing the difference value in the feature difference graph with a preset feature threshold value, wherein the preset feature threshold value is a feature threshold value preset based on different metal implant types; and then, according to the difference characteristic distribution condition of the metal artifact region, the classification judgment of the type of the metal artifact region is realized.

In the specific implementation process, the classification result obtained by the classification judgment of the region judgment module is input to the image reconstruction module, and the image reconstruction module carries out the enhancement or inhibition treatment of the metal artifact region according to the classification result to obtain a reconstructed metal artifact region, so as to obtain a reconstruction result without metal artifacts.

Still further, the image reconstruction module is used for carrying out enhancement or inhibition treatment on the metal artifact region, and the reconstruction is carried out to obtain a reconstruction result without metal artifacts.

Further, the enhancement or suppression of the metal artifact region may be performed using any one or more of the following algorithms: MAR suppression model, interpolation method, back projection method, least square method, least mean square error method, iterative optimization algorithm, support vector machine, deep learning, compressed sensing reconstruction algorithm, sparse representation reconstruction algorithm, markov random field model algorithm, wavelet transformation. The MAR inhibition model is used for carrying out enhancement or inhibition treatment on the metal artifact region to obtain a reconstructed metal artifact region, and then a normal region is integrated based on the reconstructed metal artifact region to obtain a reconstruction result without metal artifacts.

In one embodiment, the MAR suppression model is trained by enhancing or suppressing metal artifact regions with pixel-by-pixel errors between reconstructed metal artifact images and original metal artifact images as loss values. Specifically, the MAR inhibition model construction process includes two stages: the first stage, inputting a metal artifact region into an encoder of a MAR suppression model to obtain an abnormal feature map, wherein the abnormal feature map can be directly optimized by a decoder to obtain a coarse prediction; and in the second stage, inputting the internal relevant normal region into an encoder, performing cross-correlation calculation on the normal feature map and the abnormal feature map to realize feature enhancement, obtaining an enhanced feature map, inputting the enhanced feature map into a decoder to obtain fine prediction, and obtaining a constructed MAR suppression model according to a fine prediction result and loss optimization of the metal artifact-free image. Wherein the cross-correlation calculation includes an internal style consistency loss and an external style consistency loss. When the cross-correlation calculation is performed, any one or more of the following algorithms are adopted: correlation matrix, mutual information, chi-square test, lasso regression, ridge regression, principal component analysis, independent component analysis.

Still further, the region judging module and the image reconstructing module in the system further include: the region judging module is used for obtaining a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image in a matching mode according to the difference value distribution in the feature difference graph obtained by the feature calculating module, and obtaining a corresponding anatomical structure region image according to the feature distribution condition of the normal region; the region judgment module is used for carrying out enhancement or inhibition treatment on the metal artifact region to obtain a reconstructed metal artifact region, and carrying out fusion reconstruction on the reconstructed metal artifact region based on the anatomical structure region image to obtain a reconstruction result without metal artifact.

In a specific embodiment, the system further comprises a MAR reconstruction model, a MAR suppression model, a multi-stream MAR suppression sub-model, and a plurality of calculation, region judgment modules, memory module implementations, such as intensity distribution calculation, feature mapping modules, and the like.

Fig. 4 is an image reconstruction apparatus based on region judgment according to an embodiment of the present invention, including: a memory and a processor; the apparatus may further include: input means and output means.

The memory, processor, input device, and output device may be connected by a bus or other means, as illustrated by way of example in FIG. 4; wherein the memory is used for storing program instructions; the processor is configured to invoke the program instructions, which when executed, are configured to perform or implement the above-described region-based image reconstruction method.

The invention provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the above-mentioned image reconstruction method based on region judgment.

It will be clearly understood by those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described system, apparatus and module may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.

In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the device embodiments described above are merely illustrative; for another example, the division of the modules is just one logic function division, and other division modes can be adopted in actual implementation; as another example, multiple modules or components may be combined or may be integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices, or modules, which may be in electrical, mechanical, or other forms.

The modules illustrated as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., may be located in one place, or may be distributed over a plurality of network modules. Specifically, some or all modules in the system may be selected according to actual needs to achieve the purpose of the embodiment.

In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module. The integrated module can be realized in a form of hardware or a form of a software functional module.

Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be implemented by a program to instruct related hardware, the program may be stored in a computer readable storage medium, and the storage medium may include: read Only Memory (ROM), random access Memory (RAM, random Access Memory), magnetic or optical disk, and the like.

Those of ordinary skill in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by a program, and the program may be stored in a computer readable storage medium, where the storage medium may be a read only memory, a magnetic disk, or an optical disk.

While the invention has been described in detail with respect to a computer device, those skilled in the art will appreciate that they can readily use the disclosed embodiments as a basis for the teaching of the present invention. In summary, the present description should not be construed as limiting the invention.

Claims

1. An image reconstruction method based on region judgment, the method comprising:

and carrying out enhancement or inhibition treatment on the metal artifact region, and reconstructing to obtain a reconstruction result without metal artifacts.

2. The image reconstruction method based on region judgment according to claim 1, wherein the preliminarily reconstructed metal artifact image further comprises a step of inputting the metal artifact image into a MAR reconstruction model, wherein the MAR reconstruction model is constructed based on intensity distribution characteristics of the metal artifact-free image;

specifically, the construction process of the MAR reconstruction model comprises the following steps:

3. The image reconstruction method based on region judgment according to claim 1, wherein the MAR reconstruction model further comprises inputting the intensity distribution characteristics to an external memory module for similarity calculation, and iterating the intensity distribution characteristics in the external memory module according to a preset threshold, wherein the external memory module is a memory dictionary containing N representative metal artifact-free region intensity distribution characteristics obtained by performing characteristic calculation based on metal artifact-free images.

4. The image reconstruction method based on region judgment according to claim 1, further comprising: acquiring a metal artifact image, and performing feature mapping and feature reconstruction on the metal artifact image based on the distribution features of the metal artifact-free image to obtain a preliminarily reconstructed metal artifact image;

5. The image reconstruction method based on region judgment according to claim 1, further comprising: obtaining a metal artifact region and a normal region in the preliminarily reconstructed metal artifact image according to the difference value distribution in the characteristic difference map, and obtaining a corresponding anatomical structure region image according to the characteristic distribution condition of the normal region; and carrying out enhancement or inhibition treatment on the metal artifact region to obtain a reconstructed metal artifact region, and reconstructing the reconstructed metal artifact region based on the anatomical structure region image to obtain a metal artifact-free reconstruction result.

6. The image reconstruction method based on region judgment according to claim 5, wherein the corresponding anatomical region image is obtained according to the feature distribution condition of the normal region, the corresponding anatomical region image is generated by performing similarity calculation on the intensity distribution feature of the normal region and the intensity distribution feature in an external memory module, and the external memory module is a memory dictionary containing the intensity distribution features of the metal-free regions of N representative anatomical regions obtained by performing feature calculation based on the metal-free images of different anatomical structures.

7. The image reconstruction method based on region judgment according to claim 1, wherein the enhancement or suppression process adopts any one or several of the following algorithms: the method comprises the steps of a MAR suppression model, an interpolation method, a back projection method, a least square method, a minimum mean square error method, an iterative optimization algorithm, a support vector machine, deep learning, a compressed sensing reconstruction algorithm, a sparse representation reconstruction algorithm, a Markov random field model algorithm and wavelet transformation, wherein the MAR suppression model is obtained by calculating pixel-by-pixel errors between a preliminarily reconstructed metal artifact image and a metal artifact image and/or calculating pixel-by-pixel errors between a preliminarily reconstructed metal artifact image and a metal artifact-free image as loss values, then carrying out enhancement or suppression processing on a metal artifact region according to the loss values to obtain a reconstructed metal artifact image, and carrying out iterative training.

8. An area judgment based image reconstruction system, characterized in that the system comprises a computer program which, when executed by a processor, implements the area judgment based image reconstruction method of any one of claims 1 to 7.

9. An image reconstruction apparatus based on region judgment, the apparatus comprising: a memory and a processor; the memory is used for storing program instructions; the processor is configured to invoke program instructions, which when executed, are configured to perform the region-based determined image reconstruction method of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that the region judgment based image reconstruction method according to any one of claims 1 to 7 is implemented when the computer program is executed by a processor.