Metal artifact removal method based on multi-energy-level broad-spectrum X-ray data

Abstract

The invention relates to a metal artifact removal method based on multi-energy-level broad-spectrum X-ray data. The method comprises the following steps: acquiring double-energy-level CT image data and multi-energy-level CT image data of a plurality of patients containing metal artifacts, and standard image data after removing the metal artifacts, and constructing a training sample set; based on the CT image data of different energy levels corresponding to the same patient in the training sample set, performing metal region segmentation to obtain a metal region mask image; utilizing the training sample set and the metal area mask image training model to obtain a convergent metal artifact removal model; and inputting the to-be-processed CT image data of any energy level into the metal artifact removal model to obtain the CT image data after metal artifacts are removed. The metal artifact removing method solves the problems that in the prior art, a metal artifact removing method depends on a high-end CT scanner, or analog data are adopted, the data energy level is single, and consequently the existing method is poor in generalization ability and poor in low-energy-level CT scanning image processing effect.

Description

Technical Field

[0001] The present invention belongs to the technical field of CT imaging, and particularly relates to a method for removing metal artifacts based on multi-level broad-spectrum X-ray data. Background Art

[0002] With the progress of materials science, the application of medical metal implants in the human body is becoming more and more extensive, especially in the treatment of spinal degenerative diseases. Among patients with lumbar metal internal fixation implants, orthopedic surgeons usually choose computed tomography (CT) as the imaging examination method for postoperative evaluation. However, due to the existence of metal artifacts, the image quality of CT examinations will be significantly affected, thus greatly reducing the diagnostic efficiency of CT images.

[0003] In order to reduce the interference of metal artifacts in CT scans, researchers have proposed various methods for removing metal artifacts, including increasing the tube voltage of the scan, using high-energy virtual monoenergetic images, and adopting metal artifact reduction (MAR) algorithms, etc. However, all existing commercial MAR technologies rely on high-end CT scanners and can only perform metal artifact removal based on the original projection domain data of each manufacturer, lacking generality among different manufacturers. In addition, the large number of mid- and low-end CT scanners currently existing in grass-roots hospitals in China basically cannot be upgraded to use MAR technology. In current medical image diagnosis, the general image data format standard is the DICOM standard. Therefore, it is very necessary to develop a new MAR technology with broad generality that does not depend on certain specific types of CT scanners and is applicable to post-processing of all DICOM images.

[0004] In the context of the rapid development and application of big data and artificial intelligence, the rapid development of deep learning technology provides a new option for this demand. In recent years, the deep learning-based MAR technology (deep-MAR) has shown good performance in reducing metal artifacts. However, during the experiment, since it is difficult to obtain CT data pairs with the same anatomical structure that are affected by metal artifacts and not affected by metal artifacts, all deep-MAR technology studies have used simulated data sets. However, there are inevitably differences between simulated data and clinical data, and the deviation from the real data distribution undermines the effectiveness and generalization of the method. To avoid these drawbacks, real clinical data sets should be used throughout the model training process. In addition, in actual clinical applications, depending on the scanned part, common tube voltages for CT scans are, for example, 100Kvp, 120Kvp, or 140Kvp. If it is a scan examination related to children, a tube voltage of 80Kvp is generally recommended. In the examination of pulmonary artery vessels or head and neck artery vessels, a tube voltage of 70Kvp is currently recommended to reduce the scanning radiation dose and increase the image contrast. As people's understanding of radiation-induced carcinogenesis and teratogenesis deepens, the clinical requirement is to use a lower tube voltage for pediatric CT scans. To sum up, in actual clinical applications, patients affected by metal artifacts may face numerous different CT scan protocols with different tube voltages according to different clinical needs. However, in existing deep-MAR technology studies, many studies only focus on using a single simulated discrete x-ray spectrum for research, such as only studying 120Kvp or 100Kvp, and the included range is too small to reflect the complexity of the real scenario. These problems greatly reduce the practicality of the technology and limit the promotion from laboratory to clinical application. Summary of the Invention

[0005] In view of the above analysis, the present invention aims to disclose a method for removing metal artifacts based on multi-level broad-spectrum X-ray data, which solves the problems that the existing methods for removing metal artifacts rely on high-end CT scanners, or all use simulated data with a single data energy level, resulting in poor generalization ability of the existing methods and poor processing effect on low-energy CT scan images.

[0006] The object of the present invention is mainly achieved by the following technical solutions:

[0007] The present invention discloses a method for removing metal artifacts based on multi-level broad-spectrum X-ray data, including:

[0008] Obtain dual-level CT image data and multi-level CT image data of multiple patients with metal artifacts, as well as the standard image data after metal artifact removal corresponding to each CT image data, perform normalization processing on each image data, and construct a training sample set;

[0009] Based on the CT image data of different energy levels corresponding to the same patient in the training sample set, perform metal region segmentation to obtain a metal region mask image;

[0010] Using the standard image data and the metal region mask image as labels, successively perform iterative training on the neural network model using the normalized dual-energy CT image data and multi-energy CT image data in the training sample set to obtain a converged metal artifact removal model;

[0011] After normalizing the CT image data of any energy level to be processed, input it into the metal artifact removal model to obtain the CT image data with metal artifacts removed.

[0012] Further, the normalization processing of each image data includes:

[0013] Set the CT value threshold, and set the CT value of the pixels in the CT image data that are less than the threshold to the value corresponding to the threshold;

[0014] Based on the CT values of each pixel point of the CT image data, normalize the CT image data through the following formula:

[0015] U = CTvalue / 1000 * Uwater(E) + Uwater(E);

[0016] Where, U is the energy function of each pixel point after normalization, Uwater(E) is the radiation absorption energy function of water, E is the average energy of different spectral CTs, and CTvalue is the CT value of each pixel point.

[0017] Further, perform metal region segmentation through the following method to obtain a metal region mask image:

[0018] Take the difference of the characteristic values of the corresponding pixel points in the CT image data of two energy levels corresponding to the same patient after normalization to obtain a difference map;

[0019] Extract the pixel points in the difference map that are greater than the preset threshold and set them to 1, and set other pixel points to 0;

[0020] Take the regional image of the CT image data containing metal artifacts that is greater than the metal CT threshold and is set to 0 in the difference map as the metal region mask image.

[0021] Further, the metal artifact removal model is an encoder-decoder structure; the iterative training of the neural network model using the dual-energy CT image data and multi-energy CT image data includes:

[0022] Use the dual - energy CT image data to perform preliminary iterative training on the neural network model to obtain a primary metal artifact removal model;

[0023] Freeze the encoding layer of the primary metal artifact removal model, and use multi - energy CT image data to perform transfer training on the metal artifact removal model to obtain a final metal artifact removal model;

[0024] During preliminary iterative training and transfer training, for images with metal artifacts and corresponding standard image data, use mean squared error and structural similarity weighted loss for iterative training; for CT image data with metal artifacts and metal region masks, perform iterative training through mask loss.

[0025] Further, the encoder includes two convolutional modules, a feature fusion extraction module, and two TokMLP modules arranged in sequence; the encoder is used to sequentially perform multiple size compressions and feature extractions on the input CT image data to obtain feature maps of multiple different sizes;

[0026] The decoder includes two TokMLP modules, a feature fusion extraction module, and two convolutional modules arranged in sequence corresponding to each module in the encoder; it is used to perform upsampling and feature extraction on the feature map output by the last layer of the encoder in sequence, and fuse the feature maps of corresponding sizes in the encoder through skip connections to decode and obtain the final image after removing metal artifacts.

[0027] Further, the feature fusion extraction module includes an improved Transformer module, a parallel convolutional module, and a fusion module;

[0028] The improved Transformer module and the parallel convolutional module are arranged in parallel, and are used to perform feature extractions on the received feature maps respectively to obtain the global feature map and the local feature map corresponding to the CT image data;

[0029] The fusion module is used to perform an in - channel addition operation on the global feature map and the local feature map; and perform feature extraction on the feature map after the addition operation through a convolutional layer to obtain a feature map with local and global features.

[0030] Further, the improved Transformer module includes a one - dimensional sequence conversion layer, a window attention layer, an MLP layer, and a three - dimensional sequence restoration layer arranged in sequence;

[0031] The one - dimensional sequence conversion layer is used to perform a three - dimensional to sequence conversion on the received feature map to obtain one - dimensional sequence data;

[0032] The window attention layer is used to enhance the global feature data of the one - dimensional sequence data through window attention operations;

[0033] Perform the first addition operation on the feature data after data augmentation and the one-dimensional sequence data output by the one-dimensional sequence conversion layer, and then input it into the MLP layer for feature extraction to obtain one-dimensional sequence data containing global features;

[0034] Perform the second addition operation on the feature data obtained after the first addition operation and the feature data output by the MLP layer, and perform sequence-to-3D conversion through the three-dimensional sequence restoration layer to obtain a three-dimensional feature map containing global features.

[0035] Furthermore, the mean square error and structural similarity weighted loss are expressed as;

[0036] Loss Total =α·MSE + β·(1 - SSIM);

[0037]

[0038]

[0039] where Loss Total is the mean square error and structural similarity weighted loss, α and β are weight coefficients, MSE is the mean square error, y i is the true value of the i-th pixel point of the standard image data, is the predicted value, n is the number of pixel points, SSIM is the structural similarity index, μ x and μ y are the means of x and y respectively, and are the variances of x and y respectively, σ xy is the covariance of x and y, and C1 and C1 are constants.

[0040] Furthermore, the mask loss is expressed as:

[0041] L metal =L + γ·filter(metal_mask)·L;

[0042] where L is the pixel difference loss; γ is the weight of the metal area part, and filter is the filtering operation.

[0043] Furthermore, each dual-energy CT image data is image data including 70keV and 100keV of the same patient at the same part, and each multi-energy CT image data is image data including 70keV, 80keV, 100keV, 120keV, and 140keV of the same patient at the same part.

[0044] The present invention can at least achieve the following beneficial effects:

[0045] 1. The present invention proposes a method for removing metal artifacts based on multi - energy - level broad - spectrum X - ray data. A real - world clinical data set containing multiple X - ray spectra is used for model training. The data at each energy level is normalized based on the radiation absorption energy coefficient of water, and the metal region is segmented by a metal region extraction method that matches multi - energy - level data, improving the segmentation effect of the metal region. The subjective evaluation of reducing metal artifacts on virtual monochromatic images at different energy levels is realized, better reducing the metal artifacts in the CT image data of patients with metal implants, and solving the problem of CT image diagnosis during the postoperative follow - up of patients.

[0046] 2. The present invention performs global feature extraction through an improved Transformer module, local feature extraction through parallel convolution, and after feature fusion, realizes the connection of deep - layer and shallow - layer information of the image, captures the global information of the image, significantly enhances the representation of complex spatial transformation and long - distance feature dependencies, and improves the effect of removing metal artifacts. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings are only for the purpose of showing specific embodiments and are not considered as a limitation of the present invention. Throughout the drawings, the same reference signs denote the same components;

[0048] Figure 1 is a flowchart of the method for removing metal artifacts based on multi - energy - level broad - spectrum X - ray data in an embodiment of the present invention;

[0049] Figure 2 is a schematic diagram of a CT image with metal artifacts in an embodiment of the present invention;

[0050] Figure 3 is a schematic diagram of metal region segmentation in an embodiment of the present invention;

[0051] Figure 4 is a schematic diagram of the effect of removing metal artifacts by different models in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The following will specifically describe the preferred embodiments of the present invention with reference to the drawings, where the drawings form a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention.

[0053] An embodiment of the present invention discloses a method for removing metal artifacts based on multi - energy - level broad - spectrum X - ray data, as Figure 1 shown, including the following steps:

[0054] Step S1: Obtain the dual-energy CT image data and multi-energy CT image data of multiple patients with metal artifacts, as well as the standard image data after metal artifact removal corresponding to each CT image data, perform normalization processing on each image data, and construct a training sample set;

[0055] Specifically, when constructing the training sample set, existing metal artifact removal algorithms all adopt simulation methods, that is, adding a metal mask to data without metal artifacts and generating simulated metal artifacts through algorithms; while in this embodiment, different parts of CT under a clinical multi-device environment are used as the original samples of the training data set. (In order to improve the robustness of the model, the original samples can include CT images with and without metal artifacts); the standard images obtained after processing with traditional metal artifact removal algorithms are used as labels. This brings two advantages. First, existing traditional algorithms can directly call the raw data signals collected by CT images. Compared with threshold segmentation, they have better judgment ability for the existence and shape of real metal artifacts; in addition, in existing methods, the CT values of metals for simulating metal artifacts often use fixed values for processing, but there are a large number of small changes and fluctuations in the density and structure of actual metals. Using clinical metal artifact data in this embodiment can have a better perception of real metal artifacts.

[0056] Particularly, in the training sample set of this embodiment, each dual-energy CT image data is the image data including 70 keV and 100 keV of the same patient at the same part, and each multi-energy CT image data is the image data including 70 keV, 80 keV, 100 keV, 120 keV, and 140 keV of the same patient at the same part. Since multi-energy CT image data is difficult to obtain and has a large amount of data and high storage difficulty, when collecting data in this embodiment, a large number of dual-energy CT image data of the metal implantation parts of patients with metal implants are collected for the preliminary training of the model. And a small amount of multi-energy CT image data of patients is collected for transfer training; the metal implantation parts of each patient are not limited, and the patients with multi-energy CT image data and the patients corresponding to the dual-energy CT image data are preferably different patients to increase the generalization ability of the model.

[0057] This embodiment uses real clinical CT image data of multiple energy levels to achieve high-quality metal artifact removal for metal artifact images at any energy level and any CT scanning device, improving the applicable range and processing speed of the metal artifact removal method.

[0058] Furthermore, in order to make the model applicable to data at each energy level, it is necessary to preprocess the image data at each energy level. In this embodiment, normalization processing is performed on each image data, including:

[0059] Set the CT scan image threshold, and set the CT value of the pixels in the CT image data that are less than the CT scan image threshold to the value corresponding to the CT scan image threshold;

[0060] Based on the CT values of each pixel point of the CT image data, normalize the CT image data through the following formula:

[0061] U = CTvalue / 1000*Uwater(E)+Uwater(E);

[0062] Wherein, U is the energy function of each pixel point after normalization, Uwater(E) is the radiation absorption energy function of water, E is the average energy of different spectral CTs, and CTvalue is the CT value of each pixel point.

[0063] Exemplarily, since the CT values around the metal artifacts change greatly, before inputting the CT image data into the neural network, in this embodiment, the threshold of the CT value is set to -1000, and the CT values of the pixel points less than -1000 are set to -1000; then, the CT image (in Housefield Unit) is converted into a radiation absorption attenuation image for input. The specific method is: replace the CT values of the background area part (out of FOV) with a very small value. According to the definition of CT, the minimum CT value is -1000, so the CT values of the background area part are set to -1000. In addition, in the conventional CT image metal artifact removal work, Uwater adopts a fixed value of a single energy. In this embodiment, different spectral characteristic values of Uwater are processed; for lights of different energies, Uwater has different values. In this embodiment, the average energy E of the spectral CT is used as a coefficient, that is, Uwater = Uwater(E); wherein, the value of Uwater can be obtained by looking up a table. Due to the data storage precision and the fact that the background area is not within the CT value calculation range, the pixel values of some pixels may be less than the theoretical minimum value of the CT value (i.e., CT). The processing schemes for this part of the data by each manufacturer are different. Through the above normalization processing, this method can be applied to the devices of all manufacturers, improving the applicable range of the metal artifact removal method of the present invention.

[0064] Step S2: Based on the CT image data of different energy levels corresponding to the same patient in the training sample set, perform metal area segmentation to obtain a metal area mask image;

[0065] Specifically, perform metal area segmentation through the following method to obtain a metal area mask image:

[0066] Take the difference between the characteristic values of the corresponding pixel points in the CT image data of two energy levels corresponding to the same patient after normalization to obtain a difference image;

[0067] Extract the pixel points in the difference image that are greater than a preset threshold and set them to 1, and set other pixel points to 0;

[0068] Take the region image in the CT image data with metal artifacts that is greater than the metal CT threshold and is set to 0 in the difference image, which is the metal region mask image.

[0069] An important step in the metal artifact removal method is to extract the metal region. In the existing deep learning metal artifact removal framework, the metal region can only be determined by the method of metal threshold. However, in practical applications, the metal threshold is affected by factors such as the energy level of the scanning ray and the reconstruction method, resulting in inaccurate extraction of the metal region. This embodiment proposes a metal region extraction method based on multi-level data matching. The metal artifact morphology differences between different energy level data can be used to better separate the metal region, that is, subtract the two normalized energy level images u_1 and u_2 corresponding to each other to obtain a difference image du. When the maximum CT value of du is less than a certain threshold, it is considered that there are no metal artifacts in the two images. When the maximum value of du is greater than a certain threshold, it is considered that there are metal artifacts in the image. At this time, by extracting the mask image greater than this threshold, the region where the ray absorption rate changes greatly under different energy level rays can be obtained. And the region in the original metal artifact image that is greater than the metal CT threshold and does not show a large change in absorption rate in the mask image is the real metal region metal_mask; as Figure 2 and Figure 3 shown, Figure 2 is the CT image data with metal artifacts, Figure 3 is the image after segmentation of the corresponding metal region. It can be seen from the figure that this method can better exclude the influence of high-density bones and extract the metal region more accurately.

[0070] Step S3: Using the standard image data and the metal region mask image as labels, and using the normalized dual-energy CT image data and multi-energy CT image data in the training sample set to iteratively train the neural network model in turn to obtain a converged metal artifact removal model;

[0071] Specifically, the metal artifact removal model is an encoder-decoder structure; the iterative training of the neural network model using the dual-energy CT image data and the multi-energy CT image data in turn includes:

[0072] Using a large amount of dual-energy CT image data in the training sample set to perform preliminary iterative training on the neural network model to obtain a primary metal artifact removal model;

[0073] Freeze the encoding layer of the primary metal artifact removal model, and use a small amount of multi-energy CT image data to perform transfer training on the metal artifact removal model to obtain the final metal artifact removal model;

[0074] During the initial iterative training and transfer training, for the images with metal artifacts and the corresponding standard image data, use the mean squared error and structural similarity weighted loss for iterative training; for the CT image data with metal artifacts and the metal region mask, perform iterative training through the mask loss.

[0075] Furthermore, the encoder includes two convolutional modules, a feature fusion extraction module, and two TokMLP modules arranged in sequence; the encoder is used to sequentially perform multiple size compressions and feature extractions on the input CT image data to obtain feature maps of multiple different sizes;

[0076] The decoder includes two TokMLP modules, a feature fusion extraction module, and two convolutional modules arranged in sequence corresponding to each module in the encoder; it is used to perform upsampling and feature extraction on the feature map output by the last layer of the encoder in sequence, and fuse the feature maps of the corresponding sizes in the encoder through skip connections to decode and obtain the final image after removing metal artifacts.

[0077] More specifically, the feature fusion extraction module includes an improved Transformer module, a parallel convolutional module, and a fusion module;

[0078] The improved Transformer module and the parallel convolutional module are arranged in parallel, and are used to perform feature extractions on the received feature maps respectively to obtain the global feature map and the local feature map corresponding to the CT image data;

[0079] The fusion module is used to perform an in-channel addition operation on the global feature map and the local feature map; and perform feature extraction on the feature map after the addition operation through a convolutional layer to obtain a feature map with local and global features.

[0080] The improved Transformer module includes a one-dimensional sequence conversion layer, a window attention layer, an MLP layer, and a three-dimensional sequence restoration layer arranged in sequence;

[0081] The one-dimensional sequence conversion layer is used to perform a three-dimensional to sequence conversion on the received feature map to obtain one-dimensional sequence data; that is, slice the three-dimensional feature map along any one of the depth, height, or width dimensions, and regard each slice obtained as an element in the sequence, and reshape all the slice data into a one-dimensional sequence;

[0082] The window attention layer is used to enhance the global feature data of the one-dimensional sequence data through window attention operations;

[0083] Perform a first addition operation on the feature data after data augmentation and the one-dimensional sequence data output by the one-dimensional sequence conversion layer, and then input it into the MLP layer for feature extraction to obtain one-dimensional sequence data containing global features;

[0084] Perform a second addition operation on the feature data obtained after the first addition operation and the feature data output by the MLP layer, and splice each element in the one-dimensional sequence data obtained after the addition operation along the slice dimension through a three-dimensional sequence restoration layer to obtain a three-dimensional feature map containing global features.

[0085] The metal artifact removal model of this embodiment performs global feature extraction through an improved Transformer module, performs local feature extraction through parallel convolution, and through feature fusion, realizes the connection of top-layer and bottom-layer information, captures the global information of the image to be processed, significantly enhances the representation of complex spatial transformations and long-range feature dependencies, and improves the effect of metal artifact removal.

[0086] Furthermore, during initial iterative training and transfer training, for images with metal artifacts and corresponding standard image data, use mean squared error and structural similarity weighted loss for iterative training, expressed as;

[0087] Loss Total = α·MSE + β·(1 - SSIM);

[0088]

[0089]

[0090] where Loss Total is the mean squared error and structural similarity weighted loss, α and β are weight coefficients, MSE is the mean squared error, y i is the true value of the i-th pixel point of the standard image data, is the predicted value, n is the number of pixel points, SSIM is the structural similarity index, μ x and μ y are the means of x and y respectively, x and y are the normalized feature values of the original CT image data and the standard image data respectively, and are the variances of x and y respectively, σ xy is the covariance of x and y, and C1 and C1 are constants. In this embodiment, α is set to 1.0, and β is set to 0.1 - 0.5 to enhance the image structure recovery ability.

[0091] For CT image data with metal artifacts and a metal region mask, perform iterative training through a mask loss, expressed as:

[0092] L metal = L + γ·filter(metal_mask)·L;

[0093] Among them, L is the pixel difference loss; γ is the weight of the metal area part, filter is the filtering operation, and in this embodiment, Gaussian filtering is used as filter, so that the area closer to the metal area mask has a greater weight, and the area farther from the metal area mask has a smaller weight.

[0094] During model training, the Adam optimizer is adopted, and its learning rate is set between 0.0001 and 0.001, and the momentum parameters are Beta1 = 0.9 and Beta2 = 0.999. The learning rate is dynamically adjusted in combination with cosine annealing scheduling (cycle 5 - 10) to balance the convergence speed and stability. During the training process, the batch size is 16 - 64, the number of training epochs is 50 - 400, and an early stopping strategy (patience value 10 - 20) is adopted to prevent overfitting.

[0095] In the model evaluation stage, the quality of the artifact - removed image is measured by quantitative metrics such as peak signal - to - noise ratio (PSNR) and structural similarity (SSIM), and at the same time, combined with qualitative visual analysis, the performance of the model in detail restoration and artifact removal is verified. The generalization ability of the model is further ensured by reasonably setting the Dropout rate (0.2 - 0.5). Through the optimization of the above parameter ranges, the model can achieve efficient training and excellent performance in the CT metal artifact removal task.

[0096] In addition, in terms of image quality evaluation, this embodiment adopts a cross - validation method for the metal artifact removal efficiency of CT image data at different keV energy levels. In order to comprehensively evaluate the effectiveness of the proposed method, this embodiment adopts a method combining objective indicators and subjective analysis. Cross - energy evaluation is used to evaluate the generalization ability of cross - domain learning methods. The model trained under one keV setting is evaluated on the dataset obtained under different keV settings, that is, the model trained on 70 keV images is tested on 100 keV data.

[0097] In this embodiment, three metal artifact removal models, namely model70, model100, and modelmix (i.e., the metal artifact removal model of this embodiment), are trained using 70KeV energy level data, 100KeV energy level data, and the multi - energy level data of this embodiment respectively; Figure 4Axial images reconstructed using a soft tissue window for a patient with lumbar spinal metal internal fixation, and the effect images after metal artifact removal from these images using three metal artifact removal models and commercial modelMAR technology respectively. Among them, 70keV NoMAR is the original metal artifact image under the condition of 70keV, 70keV modelMAR is the image after metal artifact removal using commercial modelMAR technology under the condition of 70keV, 70keV model70 is the image after metal artifact removal using model70 technology under the condition of 70keV, 70keV model100 is the image after metal artifact removal using model100 technology under the condition of 70keV, 70keV modelmix is the image after metal artifact removal using modelmix technology under the condition of 70keV, 100keV NoMAR is the original metal artifact image under the condition of 100keV, 100keV modelMAR is the image after metal artifact removal using commercial modelMAR technology under the condition of 100keV, 100keV model70 is the image after metal artifact removal using model70 technology under the condition of 100keV, 100keV model100 is the image after metal artifact removal using model100 technology under the condition of 100keV, 100keV modelmix is the image after metal artifact removal using modelmix technology under the condition of 100keV. From the comparison of these images, it can be seen that for the images obtained at either the 70keV or 100keV energy level, the artifacts around the metal fixator in the CT images processed by the metal artifact removal model based on this embodiment are fewer than those in other images, and the spinal canal structure is shown more clearly. This demonstrates the effectiveness of the metal artifact removal method in this embodiment.

[0098] Step S4: Normalize the CT image data of any energy level to be processed and then input it into the metal artifact removal model to obtain the CT image data after metal artifact removal.

[0099] After training the metal artifact removal model using dual-energy and multi-energy CT image data, the metal artifact removal operation can be performed on the CT image data of any energy level, improving the efficiency and auxiliary diagnosis effect of computed tomography technology in clinical applications.

[0100] In summary, the metal artifact removal method based on multi - energy - level broad - spectrum X - ray data of the present invention uses real clinical CT image data containing multiple energy levels for model training, normalizes the data of each energy level based on the radiation absorption energy coefficient of water, and segments the metal region through a metal region extraction method matching multi - energy - level data, improving the segmentation effect of the metal region, achieving a subjective evaluation of reducing metal artifacts on virtual monochromatic images of different energy levels, better reducing metal artifacts in CT image data of patients with metal implants, and solving the problem of CT image diagnosis in postoperative follow - up of patients. Moreover, the present invention extracts global features through an improved Transformer module, extracts local features through parallel convolution, and through feature fusion, realizes the connection of deep - layer and shallow - layer information of the image, captures the global information of the image, significantly enhances the representation of complex spatial transformation and long - distance feature dependence, and improves the effect of metal artifact removal.

[0101] Those skilled in the art can understand that all or part of the processes of implementing the methods in the above - mentioned embodiments can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer - readable storage medium. Among them, the computer - readable storage medium is a magnetic disk, an optical disk, a read - only memory or a random access memory, etc.

[0102] The above - mentioned are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for removing metal artifacts based on multi-level broad-spectrum X-ray data, characterized in that: include: Obtain dual-level CT image data and multi-level CT image data of multiple patients with metal artifacts, as well as standard image data after metal artifacts are removed corresponding to each CT image data, perform normalization processing on each image data, and construct a training sample set; Based on the CT image data of different energy levels corresponding to the same patient in the training sample set, metal area segmentation is performed to obtain a metal area mask image; Using the standard image data and the metal area mask image as labels, and using the normalized dual-level CT image data and multi-level CT image data in the training sample set, the neural network model is iteratively trained in turn to obtain a converged metal artifact removal model; The CT image data of any energy level to be processed is normalized and then input into the metal artifact removal model to obtain the CT image data after the metal artifacts are removed.

2. The metal artifact removal method based on multi-energy level broad spectrum X-ray data according to claim 1, characterized in that: The normalization process of each image data includes: Setting a CT value threshold, and setting the CT value of pixels in the CT image data that are smaller than the threshold to a value corresponding to the threshold; Based on the CT value of each pixel point of the CT image data, the CT image data is normalized by the following formula: U=CTvalue / 1000*Uwater(E)+Uwater(E); Among them, U is the normalized energy function of each pixel, Uwater(E) is the radiation absorption energy function of water, E is the average energy of different energy spectra CT, and CTvalue is the CT value of each pixel.

3. The metal artifact removal method based on multi-energy level broad spectrum X-ray data according to claim 2, characterized in that: The metal area is segmented by the following method to obtain the metal area mask image: Subtract the characteristic values ​​of corresponding pixel points in the normalized CT image data of two energy levels corresponding to the same patient to obtain a difference map; Extract the pixel points whose values ​​are greater than a preset threshold in the difference map and set them to 1, and set the other pixel points to 0; The area image of the CT image data containing metal artifacts that is greater than the metal CT threshold and is set to 0 in the difference map is taken as the metal area mask image.

4. The method for removing metal artifacts based on multi-energy-level broad-spectrum X-ray data according to claim 2, characterized in that: The metal artifact removal model is an encoder-decoder structure; The step of iteratively training the neural network model in sequence using the dual-level CT image data and the multi-level CT image data includes: Using the dual-level CT image data to perform preliminary iterative training on the neural network model to obtain a primary metal artifact removal model; Freezing the coding layer of the primary metal artifact removal model, and performing migration training on the metal artifact removal model using multi-level CT image data to obtain a final metal artifact removal model; During the initial iterative training and transfer training, the images containing metal artifacts and the corresponding standard image data are iteratively trained using mean square error and structural similarity weighted loss; the CT image data containing metal artifacts and the metal area mask are iteratively trained using mask loss.

5. The method for removing metal artifacts based on multi-energy-level broad-spectrum X-ray data according to claim 4, characterized in that: The encoder includes two convolution modules, a feature fusion extraction module and two TokMLP modules arranged in sequence; the encoder is used to perform multiple size compression and feature extraction on the input CT image data in sequence to obtain multiple feature maps of different sizes; The decoder includes two TokMLP modules, a feature fusion extraction module and two convolution modules which are arranged in sequence and correspond to the modules in the encoder; the decoder is used to sequentially perform upsampling and feature extraction based on the feature map output by the last layer of the encoder, and fuse the feature maps of corresponding sizes in the encoder through jump connections, so as to decode and obtain the final image after removing metal artifacts.

6. The method for removing metal artifacts based on multi-energy-level broad-spectrum X-ray data according to claim 5, characterized in that: The feature fusion extraction module includes an improved Transformer module, a parallel convolution module and a fusion module; The improved Transformer module and the parallel convolution module are arranged in parallel to perform feature extraction on the received feature maps respectively to obtain a global feature map and a local feature map corresponding to the CT image data; The fusion module is used to perform an intra-channel addition operation on the global feature map and the local feature map; and perform feature extraction on the feature map after the addition operation through a convolution layer to obtain a feature map with local features and global features.

7. The method for removing metal artifacts based on multi-energy-level broad-spectrum X-ray data according to claim 6, characterized in that: The improved Transformer module includes a one-dimensional sequence conversion layer, a window attention layer, an MLP layer and a three-dimensional sequence restoration layer arranged in sequence; The one-dimensional sequence conversion layer is used to convert the received feature map from three dimensions to a sequence to obtain one-dimensional sequence data; The window attention layer is used to perform global feature data enhancement on the one-dimensional sequence data through a window attention operation; Performing a first addition operation on the feature data after data enhancement and the one-dimensional sequence data output by the one-dimensional sequence conversion layer, and then inputting the result into the MLP layer for feature extraction, so as to obtain one-dimensional sequence data containing global features; The feature data obtained after the first addition operation is added to the feature data output by the MLP layer for a second addition operation, and the sequence is converted to three dimensions through a three-dimensional sequence restoration layer to obtain a three-dimensional feature map containing global features.

8. The method for removing metal artifacts based on multi-energy-level broad-spectrum X-ray data according to claim 4, characterized in that: The mean square error and structural similarity weighted loss are expressed as; Loss Total =α·MSE+β·(1-SSIM); Among them, Loss Total is the mean square error and structural similarity weighted loss, α and β are weight coefficients, MSE is the mean square error, y i is the true value of the i-th pixel of the standard image data, is the predicted value, n is the number of pixels, SSIM is the structural similarity index, μ x and μ y are the means of x and y respectively, and are the variances of x and y, σ xy is the covariance of x and y, and C1 and C2 are constants.

9. The method for removing metal artifacts based on multi-energy-level broad-spectrum X-ray data according to claim 4, characterized in that: The mask loss is expressed as: L metal =L+γ·filter(metal_mask)·L; Among them, L is the pixel difference loss; γ is the weight of the metal area, and filter is the filtering operation.

10. The method for removing metal artifacts based on multi-energy level broad spectrum X-ray data according to claim 1, characterized in that: Each dual-energy level CT image data includes 70keV and 100keV image data of the same patient and the same part, and each multi-energy level CT image data includes 70keV, 80keV, 100keV, 120keV and 140keV image data of the same patient and the same part.

Images