Iterative tomographic reconstruction

Abstract

A computer-implemented tomographic reconstruction method is disclosed that includes receiving tomographic projection data of a subject, performing an initial portion of an iterative reconstruction procedure using the tomographic projection data to generate an initial volume estimate, identifying one or more metal regions within the initial volume estimate as a metal mask, forward projecting the metal mask to determine metal projection data, replacing data in the tomographic projection data corresponding to the metal projection data with interpolated data to generate interpolated tomographic projection data, performing a final portion of the iterative reconstruction procedure using the interpolated tomographic projection data to generate a final volume estimate, and inserting the final metal mask into the final volume estimate to generate a corrected final volume estimate.

Description

Technical Field

[0001] This invention relates to iterative tomographic reconstruction techniques. More specifically, this invention relates to methods for tomographic reconstruction that reduce the presence of metal artifacts, as well as related data processing apparatus, computer programs, and computer-readable storage media. Background Technology

[0002] Tomography is a non-invasive imaging technique that allows visualization of the internal structure of an object without the superposition of other structures that often interfere with conventional projection images. For example, in a conventional chest X-ray, ribs, heart, and lungs might be superimposed on the same film, while computed tomography (CT) slices capture the individual organs in their actual three-dimensional locations. Tomography has been applied to a range of disciplines, including medicine, physics, chemistry, and astronomy. While X-ray CT is perhaps the most familiar application of tomography, alternative imaging modalities, including ultrasound, magnetic resonance imaging (MRI), nuclear medicine, and microwave techniques, can be used to perform tomography.

[0003] As an example, in a conventional X-ray CT scan, an X-ray source projects an X-ray beam through the patient, and a detector measures the attenuation of the X-rays. Simultaneously, the device rotates about an axis running longitudinally through the patient. Therefore, the detector acquires data indicating the beam attenuation in various directions within the plane of rotation. Based on this data (typically in the form of a sine curve), a computing device can calculate the patient's internal structures on that particular plane (or "slice"). The patient or device can then be indexed along this axis, and other planes can be studied. In this way, a three-dimensional image or volumetric representation of the patient can be constructed from various slices.

[0004] Mathematically, the raw data acquired by a tomographic detector consists of multiple "projections" of the scanned object. These projections are effectively Radon transforms of the object's structure. Reconstruction using tomographic reconstruction essentially involves solving the inverse Radon transform.

[0005] Many practical reconstruction algorithms have been developed to reconstruct three-dimensional objects (or their two-dimensional images) from their projections. These algorithms are primarily designed based on the mathematics of the Radon transform, statistical knowledge of the data acquisition process, and the geometry of the data processing apparatus. Example reconstruction algorithms include filtered backprojection algorithms and Fourier domain reconstruction algorithms.

[0006] Iterative algorithms use multiple iterative steps to approximate the correct solution, allowing for better reconstructions at the cost of higher computation time. In summary, iterative algorithms start with a hypothetical image or volume, calculate a projection from the image or volume, compare the original projection data, and update the image or volume based on the difference between the calculated and actual projections. Advantages of iterative methods include improved insensitivity to noise and the ability to reconstruct the best image even with incomplete data.

[0007] A significant problem in CT reconstruction processes (iterative or otherwise) is the issue of metal artifacts. Metallic objects are a common cause of artifacts in computed tomography (CT) and cone-beam CT (CBCT), used in radiography and radiotherapy. Because metallic structures within the subject cause high attenuation of X-rays, they effectively block a large proportion of incident radiation. This, in turn, results in strong streaks in any reconstructed image, which can obscure the structures of interest.

[0008] Therefore, metal artifacts can obscure or mimic real anatomical structures or pathological conditions. This can lead to misdiagnosis and incorrect diagnosis, potentially resulting in inappropriate or harmful treatment. In medical imaging, particularly in CT scans, accurate visualization of internal tissues and organs is essential for detecting diseases, injuries, and other abnormalities. Therefore, improved tomographic reconstruction techniques that minimize or eliminate metal artifacts are needed. Summary of the Invention

[0009] According to one aspect of the present invention, a computer-implemented tomographic reconstruction method is provided. The method includes the step of receiving tomographic projection data of a subject (e.g., a human patient or an animal). The projection data can be received in real time from an imaging device (e.g., a CT scanner or a CBCT scanner), or it can be received from an external computing device that can store pre-acquired projection data.

[0010] The tomographic reconstruction method includes an initial portion that performs an iterative reconstruction process. The iterative reconstruction process uses tomographic projection data to ultimately generate a volume estimate of the subject. The initial portion of the iterative reconstruction process generates an initial volume estimate. The initial portion includes at least one reconstruction step, and preferably multiple reconstruction steps, such that the process iteratively refines the volume estimate until an initial volume estimate is obtained.

[0011] Tomographic reconstruction methods include the step of identifying one or more metallic regions within an initial volume estimation. These metallic regions will be areas where high attenuation is observed. The identified metallic regions are used to form a metallic mask (i.e., a binary mask in image space). The same techniques described herein can be used to process other high-attenuation non-metallic regions within the subject (e.g., stones, glass, or other foreign objects).

[0012] Tomographic reconstruction methods involve forward-projecting a metal mask into a projection space to determine or compute metal projection data (i.e., determining the region of the detector containing the metal object). For example, the identified metal can be converted into a sine curve using forward projection techniques (mathematically, from image space to projection space). The metal projection data will be zero in regions unaffected by metal artifacts and non-zero in regions of suspected metal.

[0013] The tomographic reconstruction method then includes the step of replacing the data in the tomographic projection data corresponding to the metallic projection data with interpolated data. That is, the projection data corresponding to the region considered metallic is replaced with interpolated data (in the projection space). The interpolated data is based on, for example, surrounding data considered non-metallic. This step of replacing the projection data in the region defined by the metallic mask with interpolated data can be considered a metallic restoration step, and this step can be referred to as metal artifact reduction (MAR).

[0014] The tomographic reconstruction method then includes a final section of performing an iterative reconstruction process. This final section uses interpolated tomographic projection data to generate a final volume estimate of the subject. The final section includes at least one reconstruction step, and preferably multiple reconstruction steps, such that the process iteratively refines the volume estimate until a final volume estimate is obtained.

[0015] The tomographic reconstruction method then includes the step of interpolating the final metal volume into the final volume estimate. In this way, the method generates a corrected final volume estimate (in this context, "corrected" means that metal artifacts are removed or reduced). The final metal volume can be any previously determined metal mask. That is, the final metal volume can correspond to the original pixel or voxel values ​​from the initial volume estimate. Alternatively, the final metal volume interpolated back into image space can be a binary mask scaled to a fixed, suitable intensity value.

[0016] By reducing or eliminating metal artifacts in tomographic images used in medical imaging, the technique presented in this paper ensures that images accurately represent a patient's anatomy, enabling healthcare professionals to make informed decisions and provide optimal patient care. Reducing metal artifacts enhances the clarity and sharpness of reconstructed images, allowing for improved visualization of anatomical structures and pathological features. Metal artifacts can produce abnormal appearances, leading to unnecessary follow-up tests, biopsies, or other invasive procedures. By reducing artifacts, the risk of misinterpretation is minimized, thereby reducing patient anxiety and healthcare costs.

[0017] Reconstructed tomographic images are frequently used for quantitative analyses, such as measuring tissue volume, density, or size. Metal artifacts can distort these measurements, leading to inaccurate results. In research and clinical applications, accurate quantitative data—such as those provided using the techniques described in this article—are essential for monitoring disease progression, evaluating treatment response, and performing other analyses.

[0018] The iterative reconstruction process includes at least an initial part and a final part. Tomographic reconstruction methods can also use an iterative reconstruction process with an intermediate part. The intermediate part similarly includes at least one reconstruction step, and preferably includes multiple reconstruction steps. The method then includes the following steps: performing the intermediate part of the iterative reconstruction process using interpolated tomographic projection data, and again generating an intermediate volume estimate, which can then undergo another MAR process.

[0019] Similar to the processing following the initial portion of an iterative reconstruction process, this method then identifies metallic regions within an intermediate volume and forms another metallic mask (the intermediate metallic mask). The method then forward-projects the metallic mask into the projection space to determine intermediate metallic projection data. The method then replaces the data corresponding to the intermediate metallic projection data in the interpolated tomographic projection data with interpolated data. This process updates the interpolated tomographic projection data.

[0020] Of course, the iterative reconstruction process may include additional intermediate parts, and the MAR process described above may be repeated accordingly. For example, the method may include: for each of the multiple intermediate parts of the iterative reconstruction process, performing that part to generate a volume estimate, identifying metallic regions within the volume estimate as metallic masks, forward projecting the masks to determine metallic projection data, and updating the interpolated tomographic projection data by replacing the data corresponding to the intermediate metallic projection data with interpolated data.

[0021] Any or all instances of the MAR process may involve normalizing the projected data prior to any interpolation step. This process may be referred to as Normalized Metal Artifact Reduction (NMAR) process. For example, in cases where an iterative reconstruction process includes initial and final parts, the method may include a step of normalizing the tomographic projection data before the step of replacing the data in the tomographic projection data with interpolated data. After the step of replacing the data in the tomographic projection data, the method may then include a step of denormalizing the interpolated tomographic projection data (i.e., the opposite of the previous normalization step). In effect, this NMAR process performs a "repair" in metallic regions with normalized projection data. This has been found to significantly reduce the effects of image artifacts, which are typically significant in areas surrounding any metallic object.

[0022] In another example, where the iterative reconstruction process includes an initial, intermediate, and final portion, the method may include a step of normalizing the interpolated tomographic projection data before the step of replacing the data in the interpolated tomographic projection data with interpolated data (after the intermediate portion). After the step of updating the interpolated tomographic projection data, the method may then include a step of denormalizing the updated interpolated tomographic projection data. This example (including a MAR process followed by an NMAR process) is particularly suitable for resolving low-quality identification of soft tissue regions in the presence of strong attenuated metal artifacts, compared to applying a single NMAR process alone. The first reconstruction pass (initial portion) identifies the metal; the next reconstruction pass (intermediate portion) interpolates data and compensates for the metal; and the final reconstruction pass (final portion) performs NMAR on the metal-compensated image from the previous pass.

[0023] For any or all normalization processes, the projected data can be normalized based on a volumetric forward projection of a previous image. This previous image can be obtained, for example, through thresholding of an initial image. The raw data can be segmented using the projected data from the previous image and then denormalized after interpolation. For example, the Hounsfield scale can be used, and the normalization process can assign a Hounsfield unit (HU) value of 0 (corresponding to water) to any tissue identified as having a value below 0 (e.g., fat, which typically has a HU value between -120 and -90). Thresholding may not necessarily require explicit identification of, for example, soft tissue regions, but can be obtained as a multicolor forward projection through the volume, where metallic regions are removed and soft tissue regions are thresholded to water. In practice, the normalization process can be implemented using a multimaterial ray tracer that can simultaneously track the total volume and length of individual materials present in the scanned volume. Thus, thresholding can be implicitly performed in a single step using the total length of the "soft tissue" category (i.e., materials with an intensity above a certain threshold and below bone).

[0024] In computed tomography reconstruction methods involving iterative reconstruction processes, there exists a range for optimizing the number of reconstruction iterations involved in each part. The iterative reconstruction process can be decomposed into an initial part (for iteratively generating an initial volume estimate before metal restoration), an intermediate part (for iteratively generating intermediate volume estimates and normalizing the soft tissue region), and a final part (for iteratively generating a final volume estimate). Empirically, particularly accurate results have been provided where the initial part typically comprises 15% to 25% of the total number of iterations in the iterative reconstruction process, the intermediate part typically comprises 25% to 35% of the total number of iterations, and the final part includes the remaining iterations. For example, in cases where the iterative process comprises a total of 100 update iterations, a 20-iteration initial part, a subsequent 30-iteration intermediate part, and a subsequent 50-iteration final part have been found to be particularly suitable.

[0025] In any tomographic reconstruction method, the step of identifying metallic regions in the initial volume estimation may include a thresholding process targeting metallic regions. For example, a metallic thresholding process may assign metallic labels to pixels or voxels in the initial volume estimation that have intensity higher than (or lower than) a predetermined static threshold.

[0026] Of course, more nuanced values ​​can be used in any thresholding process instead of predetermined static values. For example, the threshold can vary depending on the anatomy being reconstructed. For instance, for metal thresholding, it is desirable to have a threshold as low as possible to capture even lower-density metals, while still being higher than any other biological material to avoid accidentally classifying non-metallic areas as metal. In this respect, compact bone can be particularly problematic. Then, an appropriate choice of metal threshold can be made slightly higher than the strength value of any bone material identified in that volume. This does not have to be a fixed value for all use cases; for example, teeth are particularly dense, so the metal threshold for jaw scans can be higher than, for example, chest scans.

[0027] In any thresholding process, the threshold can vary depending on the presence of additional correction techniques used during reconstruction. For example, if no scattering correction techniques are applied during reconstruction, the result is typically a reduction in pixel or voxel intensity values ​​within the reconstructed volume. The threshold will then change accordingly.

[0028] Alternatively, techniques other than thresholding can be used to perform the identification of soft tissue regions and / or metal regions. Suitable techniques include machine learning-based methods or the use of prior information about a particular implant (where the implant shape and approximate location are already known). This technique may be particularly useful when both very dense bone material and low-density metal are present in the same region, where the intensity of dense bone pixels or voxels is higher than that of low-density metal.

[0029] In tomographic reconstruction methods involving iterative reconstruction processes, there exists a range for optimizing the number of reconstruction iterations involved in each part. The iterative reconstruction process can be decomposed into an initial part (for iteratively generating an initial volume estimate before metal restoration), an intermediate part (for iteratively generating an intermediate volume estimate), and a final part (for iteratively generating a final volume estimate). Empirically, particularly accurate results have been provided where the initial part typically comprises 15% to 25% of the total number of iterations in the iterative reconstruction process, the intermediate part typically comprises 25% to 35% of the total number of iterations, and the final part includes the remaining iterations. For example, in cases where the iterative process comprises a total of 100 update iterations, a 20-iteration initial part, a subsequent 30-iteration intermediate part, and a subsequent 50-iteration final part have been found to be particularly suitable.

[0030] The iterative reconstruction process (and any part thereof) used in tomographic reconstruction methods may include the following steps: (i) back-projecting tomographic projection data to generate candidate volume estimates; (ii) forward-projecting the candidate volume estimates to generate candidate projection data; and (iii) updating the candidate volume estimates based on the differences between the tomographic projection data and the candidate projection data. Steps (i) to (iii) may be repeated until a predetermined stopping criterion (e.g., a predetermined number of iterations or a predetermined convergence threshold) is met. The finally updated candidate volume estimate is then output as a volume estimate (i.e., the initial volume estimate, the final volume estimate, or an intermediate volume estimate).

[0031] In any tomographic reconstruction method according to the invention, the method may further include a step of performing a cleaning operation on the metal mask before determining the metal projection data. The cleaning operation may include any or all of dilation, etching, hole filling, and debris removal. For example, dilation allows for the application of error tolerance on the initial metal segmentation. Hole filling can be used to approximate small areas that are completely surrounded by suspected metallic material but otherwise considered to have a metallic signature. Debris removal may, for example, involve a dilation process following etching and is effective in removing small non-metallic areas within larger metallic areas (and vice versa).

[0032] In any tomographic reconstruction method according to the present invention, linear interpolation can be used to determine any interpolated data. An interpolation process is employed to repair specific regions of the projected data. It has been found that linear interpolation, processing each projection independently, is appropriately accurate while ensuring sufficient computational processing time. Note that the aim here is to remove "metal-damaged" projected values ​​so that they no longer cause metallic artifacts in subsequent reconstruction processes; perfectly accurate values ​​are not necessary. Linear interpolation can further process individual pixels or voxel rows independently. Thus, the interpolation process involves repairing values ​​in a one-dimensional list of values ​​by using linear interpolation between values ​​at either end of the segment to be filled. Of course, interpolation can be performed in two or three dimensions. Alternative interpolation processes can be used, including nonlinear interpolation (e.g., splines), filling with constant values, or predictions of the values ​​to be repaired (e.g., using a trained neural network).

[0033] In any tomographic reconstruction method according to the present invention, additional correction steps may be incorporated, including scattering correction steps and / or beam hardening correction steps.

[0034] While the techniques described in this paper are well-suited for any iterative tomographic reconstruction method (including fan-beam iterative reconstruction), tomographic reconstruction is particularly well-suited for CBCT reconstruction processes for which no effective techniques for reducing metal artifacts exist. For this use case, the tomographic projection data can be CBCT projection data.

[0035] Another embodiment of the present invention includes a data processing apparatus comprising a memory and a processor. The memory stores computer-executable instructions to implement a computer-implemented tomographic reconstruction method. The processing unit is configured to execute the instructions to perform the tomographic reconstruction method.

[0036] Another embodiment of the invention includes a computer program comprising instructions that, when executed by a data processing apparatus, cause the data processing apparatus to implement a computer-implemented tomographic reconstruction method.

[0037] Another embodiment of the present invention includes a non-transient computer-readable storage medium comprising instructions that, when executed by a data processing apparatus, cause the data processing apparatus to implement a computer-implemented tomographic reconstruction method.

[0038] The embodiments may be implemented in digital electronic circuits, or in computer hardware, firmware, software, or a combination thereof. The embodiments may be implemented as computer programs or computer program products, that is, computer programs tangibly embodied in non-transient information carriers (e.g., in machine-readable storage devices or in propagating signals) for execution by or control of the operation of one or more hardware modules.

[0039] Aspects of the embodiments have been described using terms such as “computing device,” “means for accepting input,” and “processing unit.” Those skilled in the art will understand that such functional terms and their equivalents can refer to spatially separated but combined parts of a system to provide the defined functions. Similarly, the same physical part of the system can provide two or more defined functions. For example, the same memory and / or processor can be used to appropriately implement separately defined means. Attached Figure Description

[0040] Refer to the attached diagram for illustrative purposes only. The diagram shows:

[0041] Figure 1 This is a flowchart of a general tomographic reconstruction method according to an embodiment;

[0042] Figure 2 These are example reconstructed CT images using known techniques;

[0043] Figure 3 These are example reconstructed CT images using the technique according to the present invention;

[0044] Figure 4 These are example reconstructed CT images reconstructed using the modified technique according to the present invention;

[0045] Figure 5 It is a set of comparative reconstructed CT images;

[0046] Figure 6 This is another set of reconstructed CT images for comparison;

[0047] Figure 7 This is an example of a tomographic reconstruction method according to the present invention;

[0048] Figure 8 This is an example of a modified tomographic reconstruction method according to the present invention;

[0049] Figure 9 This is a schematic overview of a computing device suitable for implementing the tomographic reconstruction method according to the present invention; and

[0050] Figure 10 This is a schematic overview of an imaging system suitable for implementing the tomographic reconstruction method according to the present invention. Detailed Implementation

[0051] Figure 1 This is an overview of a general tomographic reconstruction method according to aspects of the present invention. This general method for generating synthetic data is implemented using a computing device.

[0052] In S11, the computing device receives the tomographic projection data of the subject.

[0053] In S12, the computing device uses the tomographic projection data to perform the initial part of the iterative reconstruction process to generate an initial volume estimate.

[0054] In S13, the computing device identifies one or more metal regions within the initial volume estimate as the initial metal mask.

[0055] In S14, the computing device projects an initial metal mask forward to determine metal projection data.

[0056] In S15, the computing device replaces the data corresponding to the metal projection data in the tomographic projection data with interpolated data to generate interpolated tomographic projection data.

[0057] In S16, the computing device uses the interpolated tomographic projection data to perform the final part of the iterative reconstruction process to generate a final volume estimate.

[0058] In S17, the computing device inserts the final metal volume into the final volume estimate to generate a corrected final volume estimate.

[0059] In this field, the most common approach to resolving metal artifacts involves completely removing the metal object from the measurements before any reconstruction process is initiated. This can be achieved in two steps: first, identifying the location of the metal in all measurements; and second, replacing these areas to approximate the measurements as if the metal were not present. However, this typically requires two complete reconstruction processes: the first to identify the location of the metal, and the second to create a metal-compensated reconstruction after metal removal. Naturally, this conventional technique is particularly time-consuming.

[0060] However, the tomographic reconstruction method presented in this paper can integrate a metal artifact reduction process in a single iterative reconstruction process. Thus, reconstruction requires only a single reconstruction process.

[0061] Figure 2 This is an example reconstructed CT image of a human patient's pelvis. The reconstructed image has not undergone any metal artifact reduction process. Instead, the image reconstruction process used herein is known in the art. The inset is a magnified view of a smaller area where metal (e.g., a metal-to-metal hip implant) is present. Image reconstruction was performed here using an iterative reconstruction process enabled by Polyquant, which is further described below. As can be seen, the presence of metal causes significant metal artifacts in the form of white radial streaks, areas of pixel intensity oversaturation. These metal artifacts obscure anatomical details in the area surrounding the patient.

[0062] Figure 3 These are example reconstructed CT images of a human patient's pelvis obtained using the technique according to the present invention. Figure 2Conversely, the reconstruction process used for this reconstructed image incorporates projection inpainting during the iterative reconstruction process (i.e., the MAR process; and enabled by Polyquant). Metal artifacts are significantly reduced, and pixels around the metallic areas are now resolvable. The corrected final image estimate (e.g., the corrected final image estimate illustrated here) represents the patient's anatomy significantly more accurately, enabling healthcare professionals to make informed decisions and provide optimal patient care.

[0063] As described above, the technique according to the invention may optionally further include a normalization step, wherein interpolation (NMAR) of the projected data is performed on the normalized measurement results. This NMAR process provides a significant improvement over simple linear repairs. The normalization step may include classifying the projected data into soft tissue and bone categories by projecting through a thresholded (i.e., segmented or otherwise labeled) volume, resulting in normalized measurements with an intensity close to 1. This has been shown to greatly improve resolution, which is otherwise often lost in and around metallic objects.

[0064] Figure 4 These are example reconstructed CT images of a human patient's pelvis obtained using techniques of the present invention, which further include a normalization step (i.e., MAR, followed by NMAR). Figure 3 Conversely, the reconstruction process used for this reconstructed image incorporates additional normalization of the projection data and subsequent replacement of the metal projection data within the normalized projection data. As shown in the figure, the additional normalization process further improves the quality of the reconstructed image—this is particularly significant in reducing metal artifacts and improving resolution within and around the boundaries of metal regions.

[0065] Figure 5 It is the human patient's pelvis (and) Figures 2 to 4 A set of example reconstructed CT images (the same as those in the text). Each column corresponds to a different slice within the patient's reconstructed volume. The top row shows reconstructed images that have not yet undergone any metal artifact reduction process. The middle row shows reconstructed images that have undergone a metal artifact reduction process according to the techniques of this paper (MAR, followed by NMAR), which also includes a normalization process. The bottom row shows reconstructed images that have undergone a metal artifact reduction process according to the techniques of this paper (MAR only).

[0066] Figure 6 This is a set of example reconstructed CT images of the pelvis of another human patient. Each column corresponds to a different slice within the patient's reconstructed volume. The top row shows reconstructed images that have not yet undergone any metal artifact reduction process. The middle row shows reconstructed images that have undergone metal artifact reduction according to the techniques described herein (MAR, followed by NMAR), which also include a normalization process. The bottom row shows reconstructed images that have undergone metal artifact reduction according to the techniques described herein (MAR only).

[0067] In all the reconstructed images presented here, when the tomographic reconstruction technique described herein is applied, an improvement in image quality in and around the metal artifacts can be directly observed. It is also clear that this improvement is particularly noticeable when the anatomical region does not contain any metal (e.g., for...). Figure 5 and Figure 6 (As shown in the rightmost image in the text, this is the case.) As described in this article, the tomographic reconstruction technique has little or no impact on the quality or fidelity of the reconstructed image.

[0068] The following structured outline provides an example implementation of tomographic reconstruction that integrates metal artifact reduction through projection instigation within an iterative reconstruction process (incorporated into MAR, optionally followed by NMAR). The iterative reconstruction process includes... Steps (here) This could be the number of volume / image updates or the number of periods.

[0069] 1) In The steps (of which) After that, perform metal repair:

[0070] a) Identify metallic structures in the current volume estimate (e.g., through thresholding) to create a metallic mask.

[0071] b) Optionally, cleaning operations (e.g., expansion, erosion, hole filling, debris removal) are performed on the mask.

[0072] c) The application uses a forward projection operation through a metal mask to identify all measured elements containing metal that are to be replaced.

[0073] d) For example, use linear interpolation to replace all measurements containing metal.

[0074] e) Update the measurement results with the repaired measurement results and continue iterating.

[0075] 2) Optionally, in After the steps, among which ( ):

[0076] a) Create a normalized projection from the current estimate by thresholding the soft tissue according to a predetermined value (e.g., water, which has a Henle unit (HU) value of 0) and applying a forward projection operation.

[0077] b) Normalize the measurement results and correct the metal-containing areas as identified in the first step.

[0078] c) Denormalize the repaired measurement results and continue iterating.

[0079] 3) In After this step: insert the metal mask into the final reconstructed volume to complete the metal compensation.

[0080] In this example, from arrive The iterative reconstruction process can be referred to as the initial part; from arrive The iterative reconstruction process can be referred to as the intermediate part; and from arrive The iterative reconstruction process can be referred to as the final part. Step 1) above can be called the MAR process. Step 2) above can be called the NMAR process.

[0081] The above , and The value of L is variable, and the optimal value can depend on other specific parameter values. Empirically, smaller values ​​of L and values ​​of N that are essentially around M / 2 are well-suited for sufficient reconstruction quality without unduly high computational costs. This leaves sufficient volume / image updates or periods after the MAR and / or NMAR steps so that the iterative reconstruction process converges to a stable volume estimate. For example, in an iterative reconstruction process that includes MAR, followed by NMAR, and includes 100 steps (… In the case of ), MAR can be performed in 20 steps ( This can be achieved after 30 steps, and NMAR can be implemented in another 30 steps (i.e., the 50 steps of the iterative reconstruction process). This is then implemented. The final part of this example then comprises half the number of iterations of the entire iterative reconstruction process.

[0082] Denormalization of the projected data can be achieved using any conventional means, including, for example, by multiplying pixel-wise (or voxel-wise) based on previous projected data (e.g., based on a sine wave from a step prior to normalization during iterative reconstruction).

[0083] The techniques described in this paper are well-suited for implementation within a series of iterative reconstruction processes. A quantitative iterative reconstruction process known as Polyquant is particularly appropriate (Mason et al., “Quantitative cone-beam CT reconstruction with polyenergetic scatter model fusion,” Physics in Medicine & Biology (2018) and Mason et al., “Polyquant CT: direct electron and mass density reconstruction from a single polyenergetic source,” Physics in Medicine and Biology (2017c)).

[0084] Polyquant offers several key advantages in medical imaging, leading to improved diagnostic and therapeutic outcomes. A primary benefit is its ability to produce enhanced image quality with reduced artifacts. By addressing limitations common in conventional CBCT, such as beam hardening, Polyquant provides a clearer and more accurate representation of patient anatomy, resulting in reconstructed volume estimation and enhanced contrast in images. Experimental evaluations have also demonstrated Polyquant's ability to significantly reduce beam hardening artifacts, which can obscure details and affect the accuracy of quantitative measurements.

[0085] The foundation of the Polyquant technique lies in its piecewise linear model of X-ray attenuation as a function of electron or mass density. This model represents the attenuation of X-rays passing through a material as the sum of several linear segments. Each segment is defined by a slope and a y-intercept, and these parameters vary depending on the energy of the X-ray photons. This energy dependence is crucial for accurately representing how different materials attenuate X-rays across a broad spectrum produced by a clinical CT scanner. The transition points between these linear segments (called “inflection points”) are key features of the model and are intentionally kept constant across all X-ray energies used in the imaging process. This constraint on the transition points is useful for maintaining computational efficiency within the Polyquant algorithm. The mathematical formulation of the model includes specific constraints to ensure that the model behaves in a physically realistic way, such as maintaining continuity between different linear segments and ensuring that a perfect vacuum results in zero X-ray attenuation.

[0086] The iterative reconstruction process in Polyquant typically involves minimizing a penalized negative log-likelihood (NLL) function. This objective function is designed to model the statistical properties of X-ray measurements, usually assuming a Poisson distribution, which is a good approximation of the photon counting process in CT. Minimizing this NLL function aims to find the distribution of electron or mass density within the scanned object that best explains the observed X-ray projection data according to the chosen statistical model. A common approach to achieving this minimization is through gradient descent, which can be enhanced with acceleration techniques such as momentum to improve convergence speed. To further refine the reconstructed image and ensure its quality, regularization techniques, such as total variation (TV) regularization, can be incorporated. These techniques help suppress noise in the image while preserving important features, such as edges and boundaries between different tissues. The entire iterative process typically begins with an initial estimate of the density distribution and then continues through a series of iterations. Each iteration involves a gradient descent step, possibly with an inertia term to accelerate convergence, followed by the application of the proximity operator. The proximity operator is used to impose constraints on the reconstructed density values, such as ensuring that they remain non-negative and within physically reasonable limits, and also incorporates the effects of the chosen regularization method.

[0087] To effectively address the challenges posed by scattered radiation, particularly in the case of cone-beam computed tomography (CBCT), a significant advancement offered by Polyquant techniques is the ability to incorporate, for example, CNN-based scattering estimation processes.

[0088] Continue iterating until a predefined convergence criterion is met (e.g., a small change in density estimation between iterations or a maximum number of iterations). The final output is a 3D image (volume) representing the reconstructed quantitative density (electron density or mass density) of the scanned subject.

[0089] From a practical perspective, computational requirements are a consideration for any iterative reconstruction process, such as Polyquant. Suitable hardware for implementing tomographic reconstruction using the techniques described in this paper includes any general-purpose CPU. The techniques described in this paper can be implemented using many computing platforms and languages, including MATLAB (for which the Polyquant CT Reconstruction Toolbox is available) or C++. Optimized implementations and hardware, such as those using GPUs and GPGPUs, can be leveraged to further reduce tomographic reconstruction time.

[0090] Figure 7This is a schematic flowchart of an example tomographic reconstruction method according to the present invention. The process can be described based on the operations performed and the data structures provided in the image space and projection space. Input tomographic projection data is fed into the initial part of the iterative reconstruction process. The iterative process may, for example, involve repeated back projection (denoted as "B") and forward projection or reconstruction (denoted as "F") between the projection data (in the projection space) and the candidate volume estimate (in the image space). When a predetermined stopping criterion is met, the current candidate volume estimate becomes the initial volume estimate.

[0091] A metal identification process is performed on the initial volume estimate to generate a metal mask. The metal mask is then forward-projected to provide metal projection data. The interpolated projection data (corresponding to the initial volume estimate; i.e., the projection data present at the end of the first part of the iterative reconstruction process) is replaced (via an interpolation process) with the metal projection data to generate interpolated projection data.

[0092] The iterative process then restarts, with the final part of the iterative reconstruction process similarly involving repeated back-projection and forward projection between the projected data and the candidate volume estimates. When a predetermined stopping criterion is met, the current candidate volume estimate becomes the final volume estimate. The final metal mask (e.g., a mask corresponding to an earlier identified metal) is then inserted into the final volume estimate to provide a corrected volume estimate, enabling healthcare professionals to make informed decisions and provide optimal patient care.

[0093] Figure 8 This is another schematic flowchart of an example tomographic reconstruction method according to the present invention. Relative to Figure 7 The process outlined above includes an intermediate portion of an iterative reconstruction process, followed by a normalization process. The above relates to... Figure 7 The descriptions of the operations and data structures in the same way also apply to the operations and data structures described in the previous section. Figure 8 .

[0094] After replacing the projection data with (interpolated) metal projection data, the intermediate portion of the iterative reconstruction process produces an intermediate volume estimate. Similar to the above, a metal identification process is performed on the intermediate volume estimate to produce an intermediate metal mask, which is forward-projected to provide intermediate metal projection data. The projection data (corresponding to the intermediate volume estimate; i.e., the projection data present at the end of the intermediate portion of the iterative reconstruction process) is normalized (e.g., referring to projection data from a previous image) to produce normalized projection data. This normalization process can be implemented using the Polyquant forward projector function, which is essentially a multi-material X-ray tracer and multicolor integrator, and allows for the computation of multicolor X-ray measurements. In the normalizer, the density integral for the soft tissue extent is discarded to perform implicit thresholding for water.

[0095] The projection data corresponding to the metal projection data in the (interpolated) normalized projection data is replaced to update the existing interpolated projection data, thereby providing normalized interpolated projection data. As previously described, the iterative process then restarts, where the final part of the iterative reconstruction process similarly involves repeated back-projection and forward projection between the projection data and the candidate volume estimates. This modified process (involving MAR, followed by NMAR) improves the accuracy and fidelity of the reconstructed volume estimate.

[0096] Further modifications to the general process can be envisioned and implemented, including performing metal identification and soft tissue identification together across different parts of the iterative reconstruction process (as well as subsequent interpolation and normalization processes).

[0097] Figure 9 This is a block diagram of a computing device that can be used to implement the tomographic reconstruction method according to the present invention. The computing device includes a processor 93 and a memory 94. Optionally, the computing device also includes a network interface 97 for communicating with other computing devices. For example, the system may consist of a network of such computing devices. Optionally, the computing device also includes one or more input mechanisms 96 (e.g., a keyboard and mouse) and a display unit 95 (e.g., one or more monitors). These components may be connected to each other via a bus 92.

[0098] Memory 94 may include computer-readable media, the term can refer to a single medium or multiple media (e.g., a centralized or distributed database and / or associated caches and servers) configured to carry computer-executable instructions or have data structures stored thereon. Computer-executable instructions may include, for example, instructions and data accessible and caused to perform one or more functions or operations by a general-purpose computer, a special-purpose computer, or a special-purpose processing device (e.g., one or more processors). Therefore, the term "computer-readable storage medium" may also include any medium capable of storing, encoding, or carrying a set of instructions for execution by a machine and causing the machine to perform any one or more methods of the present invention. The term "computer-readable storage medium" may accordingly be taken as including, but not limited to, solid-state memory, optical media, and magnetic media. By way of example and not limitation, such computer-readable media may include non-transient computer-readable storage media, including random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), optical disc read-only memory (CD-ROM) or other optical disc storage, disk storage or other magnetic storage devices, and flash memory devices (e.g., solid-state storage devices).

[0099] Processor 93 is configured to control computing devices and perform processing operations, such as executing code stored in memory to implement various functions of the tomographic reconstruction techniques described herein. Memory 94 stores data read from and written by processor 93. As mentioned herein, the processor may include one or more general-purpose processing devices, such as microprocessors, central processing units, etc. The processor may include complex instruction set computing (CISC) microprocessors, reduced instruction set computing (RISC) microprocessors, very long instruction word (VLIW) microprocessors, or processors implementing other instruction sets, or processors implementing combinations of instruction sets. The processor may also include one or more special-purpose processing devices, such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), network processors, etc. In one or more embodiments, the processor is configured to execute instructions for performing the operations and steps discussed herein.

[0100] The display unit 95 can display a representation of data stored by the computing device (e.g., a reconstructed volume estimate and / or its associated image), and can also display a cursor as well as dialog boxes and screens that enable the user to interact with the programs and data stored on the computing device.

[0101] The input mechanism 96 enables users to input data (such as tomographic projection data) and instructions into the computing device.

[0102] The network interface (network I / F) 97 can be connected to a network (such as the Internet) and can connect to other such computing devices via the network. The network I / F 97 can control the input / output of data to / from other devices via the network. Other peripheral devices (such as microphones, speakers, printers, power supply units, fans, enclosures, scanners, trackballs, etc.) can be included in the computing device.

[0103] The method described herein can be applied to computing devices (e.g., Figure 9 It executes on the exemplified computing device. Such a computing device does not need to have Figure 9 Each of the components is exemplified, and can be composed of a subset of these components.

[0104] The tomographic reconstruction method described herein can be performed by a single computing device communicating with one or more data storage servers via a network. The computing device may be a data storage device that itself stores volumetric estimates and / or projection data.

[0105] The tomographic reconstruction method described herein can be performed by multiple computing devices operating collaboratively. One or more of these computing devices may be a data storage server storing at least a portion of the volumetric estimation and / or projection data.

[0106] Figure 10 This is a block diagram illustrating an implementation of an imaging system 1000 suitable for performing the tomographic reconstruction method described herein. The example imaging system 1000 includes a computing system 1010 within which a set of instructions can be executed to cause the computing system 1010 to perform the method (or steps thereof) discussed herein. The computing system 1010 can implement tomographic reconstruction. The computing system 1010 may also be referred to as a computer, computing device, or computing apparatus, and may correspond to... Figure 9 An illustrative computing device. In particular, the method described herein can be implemented by the processor or controller circuitry 1011 of the computing system 1010.

[0107] The computing system 1010 should be considered to include any number of machines or collections of machines, such as one or more computing devices, which individually or jointly execute one or more sets of instructions to perform any one or more methods discussed herein. That is, the hardware and / or software may be provided in a single computing device or distributed across multiple computing devices in the computing system. In some implementations, one or more elements of the computing system may be connected (e.g., networked) to other machines in, for example, a local area network (LAN), intranet, extranet, or the Internet. One or more elements of the computing system may operate as a server or client machine in a client-server network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. One or more elements of the computing system may be a personal computer (PC), tablet computer, set-top box (STB), personal digital assistant (PDA), cellular phone, network appliance, server, network router, switch, or bridge, or any machine capable of executing a set of instructions (sequential or otherwise) specifying the actions to be taken by that machine.

[0108] The computing system 1010 includes controller circuitry 1011 and memory 1013 (e.g., read-only memory (ROM), flash memory, dynamic random access memory (DRAM) such as synchronous DRAM (SDRAM) or Rambus DRAM (RDRAM), etc.). Memory 1013 may include static memory (e.g., flash memory, static random access memory (SRAM), etc.) and / or auxiliary memory (e.g., data storage devices), which communicate with each other via a bus (not shown). Memory 1013 can be used to store or buffer tomographic projection data and parameters until needed for image reconstruction, and can be used to store volume estimates throughout the processing.

[0109] Controller circuit 1011 represents one or more general-purpose processors, such as microprocessors, central processing units, accelerated processing units, etc. More specifically, controller circuit 1011 may include complex instruction set computing (CISC) microprocessors, reduced instruction set computing (RISC) microprocessors, very long instruction word (VLIW) microprocessors, processors implementing other instruction sets, or processors implementing combinations of instruction sets. Controller circuit 1011 may also include one or more special-purpose processing devices, such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), network processors, etc. The one or more processors of the controller circuit may have a multi-core design. Controller circuit 1011 is configured to execute processing logic for performing the operations and steps discussed herein.

[0110] The computing system 1010 may also include network interface circuitry 1015. The computing system 1010 may be communicatively coupled to input device 1020 and / or output device 1030 via input / output circuitry 1016. In some implementations, input device 1020 and / or output device 1030 may be elements of the computing system 1010. Input device 1020 may include alphanumeric input devices (e.g., keyboard or touchscreen), cursor control devices (e.g., mouse or touchscreen), audio devices such as microphones, and / or haptic input devices. Output device 1030 may include audio devices such as speakers, video display units (e.g., liquid crystal displays (LCDs) or cathode ray tubes (CRTs)), and / or haptic output devices. In some implementations, input device 1020 and output device 1030 may be provided as a single device or as separate devices.

[0111] In some implementations, the computing system 1010 may include image processing circuitry 1014. Image processing circuitry 1014 may be configured to process measurement data 1070 (e.g., images, imaging data, projections, projection data), such as tomographic projection data obtained from one or more imaging data sources, treatment device 1050, and / or image acquisition device 1040. Image processing circuitry 1014 may be configured to process or preprocess measurement data 1070. For example, image processing circuitry 1014 may convert received data into a specific format, size, resolution, etc. Image processing circuitry 1014 may be configured to perform image reconstruction. In some implementations, image processing circuitry 1014 may be combined with controller circuitry 1011.

[0112] In some implementations, the imaging system 1000 may further include an image acquisition device or apparatus 1040 and / or a treatment device or apparatus 1050. The image acquisition device 1040 and the treatment device 1050 may be provided as a single device. In some implementations, the treatment device 1050 is configured to perform imaging, for example, in addition to providing treatment and / or during treatment.

[0113] The image acquisition device 1040 can be configured to perform positron emission tomography (PET), computed tomography (CT) (including CBCT), magnetic resonance imaging (MRI), single positron emission tomography (SPECT), X-ray, etc.

[0114] Image acquisition device 1040 can be configured to output measurement data 1070 that can be accessed by computing system 1010. Treatment device 1050 can be configured to output treatment data 960 that can be accessed by computing system 1010. Treatment data 960 can be obtained from an internal data source (e.g., from memory 1013) or from an external data source (e.g., treatment device 1050 or an external database).

[0115] The methods described above can be implemented by computer programs. Computer programs may include computer code (e.g., instructions) arranged to instruct a computer to perform one or more of the functions described above. For example, regarding... Figure 1 The steps of the described method can be executed by computer code. The steps of the above method can be executed in any suitable order. A computer program and / or code for performing this method can be provided to an apparatus, such as a computer, on one or more computer-readable media, or more generally, on a computer program product. The computer-readable media can be transient or non-transient. One or more computer-readable media can be, for example, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, or a propagation medium for data transmission (e.g., for downloading code via the Internet). Alternatively, one or more computer-readable media can take the form of one or more physical computer-readable media, such as semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), hard disk, and optical disk (e.g., CD-ROM, CD-R / W, or DVD). Instructions can also reside wholly or at least partially within memory 1013 and / or controller circuitry 1011 during execution by computer system 1010, which also constitute computer-readable storage media.

[0116] In implementation, the modules, components, and other features described herein can be implemented as discrete components or integrated into the functionality of hardware components such as ASICs, FPGAs, DSPs, or similar devices.

[0117] A "hardware component" is a tangible (e.g., non-transient) physical component (such as a group or one or more processors) capable of performing certain operations and which can be configured or arranged in some physical manner. A hardware component may include dedicated circuitry or logic permanently configured to perform certain operations. A hardware component may include dedicated processors, such as FPGAs or ASICs. A hardware component may also include programmable logic or circuitry that is temporarily configured by software to perform certain operations.

[0118] Additionally, modules and components can be implemented as firmware or functional circuitry within a hardware device. Furthermore, modules and components can be implemented as any combination of hardware devices and software components, or as software only (e.g., code stored or otherwise embodied in a machine-readable medium or transmission medium).

[0119] Unless otherwise specifically stated, it should be understood, as is apparent from the following discussion, that throughout this specification, the use of terms such as “receive,” “execute,” “identify,” and “replace” refers to the actions and processes of a computer system or similar electronic computing device that manipulates data represented as physical (electronic) quantities in the registers and memories of the computer system and transforms it into other data similarly represented as physical quantities in the computer system’s memory or registers or other such information storage, transmission, or display devices.

[0120] While certain embodiments have been described, these embodiments are presented by way of example only and are not intended to limit the scope of the invention. In fact, the methods and apparatus described herein can be embodied in various other forms; furthermore, various omissions, substitutions, and changes can be made to the form of the methods and apparatus described herein.

Claims

1. A computer-implemented tomographic reconstruction method, the method comprising: Receive tomographic projection data from the subject; The initial part of the iterative reconstruction process is performed using the tomographic projection data to generate an initial volume estimate; One or more metallic regions within the initial volume estimate are identified as initial metallic masks; The initial metal mask is projected forward to determine metal projection data; Replace the data in the tomographic projection data that corresponds to the metal projection data with interpolated data to generate interpolated tomographic projection data. The final part of the iterative reconstruction process is performed using the interpolated tomographic projection data to generate a final volume estimate; as well as The final metal volume is inserted into the final volume estimate to generate a corrected final volume estimate.

2. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, Also includes: The intermediate portion of the iterative reconstruction process is performed using the interpolated tomographic projection data to generate an intermediate volume estimate; One or more metal regions within the intermediate volume estimate are identified as intermediate metal masks; The intermediate metal mask is projected forward to determine intermediate metal projection data; as well as The interpolated tomographic projection data is updated by replacing the data in the interpolated tomographic projection data that corresponds to the intermediate metal projection data with interpolated data.

3. The computer-implemented tomographic reconstruction method according to claim 2, characterized in that, Also includes: The interpolated tomographic projection data is repeatedly updated for multiple intermediate parts of the iterative reconstruction process.

4. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, Also includes: Before replacing the data in the tomographic projection data and / or the interpolated tomographic projection data, the tomographic projection data and / or the interpolated tomographic projection data are normalized. as well as The interpolated tomographic projection data is denormalized.

5. The computer-implemented tomographic reconstruction method according to claim 4, characterized in that, The normalization includes a soft tissue thresholding process.

6. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, Identifying one or more metallic regions involves a metal thresholding process.

7. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, The final part of the iterative reconstruction process comprises approximately half the number of iterations of the iterative reconstruction process.

8. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, The iterative reconstruction process includes: (i) Back-projection tomography data to generate candidate volume estimates; (ii) Forward projection of the candidate volume estimate to generate candidate projection data; (iii) Update the candidate volume estimate based on the difference between the tomographic projection data and the candidate projection data; and (iv) Repeat (ii) and (iii) until the stopping criterion is met, and output the candidate volume estimate.

9. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, Also includes: Perform a cleaning operation on any metal mask before determining any metal projection data.

10. The computer-implemented tomographic reconstruction method according to claim 9, characterized in that, The cleaning operations include any or all of the following: expansion, erosion, hole filling, and debris removal.

11. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, Linear interpolation techniques are used to determine any interpolated data.

12. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, Also includes: Scattering correction is performed on the tomographic projection data.

13. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, Also includes: Beam hardening correction is performed on the tomographic projection data.

14. The computer-implemented tomographic reconstruction method according to claim 1, characterized in that, The tomographic projection data is cone-beam computed tomography (CBCT) projection data.

15. A data processing apparatus comprising a memory storing computer-executable instructions and a processor configured to execute the instructions to implement a computer-implemented tomographic reconstruction method, the method comprising: Receive tomographic projection data from the subject; The initial part of the iterative reconstruction process is performed using the tomographic projection data to generate an initial volume estimate; One or more metallic regions within the initial volume estimate are identified as initial metallic masks; The initial metal mask is projected forward to determine metal projection data; Replace the data in the tomographic projection data that corresponds to the metal projection data with interpolated data to generate interpolated tomographic projection data. The final part of the iterative reconstruction process is performed using the interpolated tomographic projection data to generate a final volume estimate; as well as The final metal volume is inserted into the final volume estimate to generate a corrected final volume estimate.

16. A computer program including instructions, which, when executed by a computer, cause the computer to implement a computer-implemented tomographic reconstruction method, the method comprising: Receive tomographic projection data from the subject; The initial part of the iterative reconstruction process is performed using the tomographic projection data to generate an initial volume estimate; One or more metallic regions within the initial volume estimate are identified as initial metallic masks; The initial metal mask is projected forward to determine metal projection data; Replace the data in the tomographic projection data that corresponds to the metal projection data with interpolated data to generate interpolated tomographic projection data. The final part of the iterative reconstruction process is performed using the interpolated tomographic projection data to generate a final volume estimate; as well as The final metal volume is inserted into the final volume estimate to generate a corrected final volume estimate.

17. A computer-readable medium including instructions, which, when executed by a computer, cause the computer to perform a computer-implemented tomographic reconstruction method, the method comprising: Receive tomographic projection data from the subject; The initial part of the iterative reconstruction process is performed using the tomographic projection data to generate an initial volume estimate; One or more metallic regions within the initial volume estimate are identified as initial metallic masks; The initial metal mask is projected forward to determine metal projection data; Replace the data in the tomographic projection data that corresponds to the metal projection data with interpolated data to generate interpolated tomographic projection data. The final part of the iterative reconstruction process is performed using the interpolated tomographic projection data to generate a final volume estimate; as well as The final metal volume is inserted into the final volume estimate to generate a corrected final volume estimate.

Images