Method for metal artifact avoidance in x-ray imaging

Abstract

A system and method for avoiding metal artifacts in 3D X-ray imaging are provided. The method includes: determining the 3D position of the metal in an object or volume of interest to be scanned; estimating a source-detector trajectory that will reduce the severity of the metal artifacts; moving the imaging system to a position consistent with the estimated source-detector trajectory; and scanning the object according to the source-detector trajectory.

Description

[0001] Cross-reference to related applications

[0002] This application claims priority to U.S. Provisional Application No. 62 / 944,615, filed December 6, 2019, the entire disclosure of which is hereby incorporated herein by reference in its entirety. Technical Field

[0003] This teaching relates in general to x-ray imaging, and more specifically to a system and method for avoiding metal artifacts in x-ray imaging. Background Technology

[0004] 3D volumetric X-ray imaging (e.g., X-ray computed tomography (CT) or cone-beam computed tomography (CBCT)) is a common means of imaging the human body in diagnostic and image-guided medical procedures. CT is the backbone of diagnostic radiology, and CBCT systems have emerged for a variety of specialized applications (including diagnostic imaging in dentistry, otolaryngology, orthopedics, and breast imaging) and for guiding interventional procedures, interventional radiology, and image-guided radiotherapy. Such CBCT systems offer the advantages of portability, small footprint, low cost, and the ability to provide both fluoroscopy and 3D imaging. The primary application area of ​​the invention described below is CBCT image-guided surgery, where metallic implants in the human body typically limit image quality. Secondary applications include other areas of CBCT imaging (e.g., specialized diagnostic imaging systems in dentistry, otolaryngology, or orthopedics) and some potential applications of CT (e.g., interventional CT and / or diagnostic CT).

[0005] Factors that challenge image quality in CT and CBCT include image artifacts from high-density objects such as surgical clips and staples, dental fillings, surgical instruments, and orthopedic implants. These objects cause various image artifacts (generally referred to as "metal artifacts") that obscure useful information about the imaging tissue and hinder the outline of the metal object itself. An example scenario for intraoperative imaging is the need for precise visualization of the placement of metal instruments (such as implanted screws) relative to surrounding anatomical structures to guide, navigate, and confirm instrument placement. A large number and / or high density of metal objects in the field of view (FOV) can severely degrade image quality and obscure the visualization of nearby anatomical structures and the confirmation of instrument placement.

[0006] The artifacts described above include dark and bright bands, as well as streaks, commonly referred to as metallic artifacts. These artifacts can be attributed to several effects, including beam hardening (an X-ray energy spectrum shift that introduces error or bias into the detector signal), X-ray scattering (a high proportion of scattered X-rays contributing to the detector signal in highly attenuated regions of metallic objects), and photon starvation (a small number of X-ray photons contributing to the detector signal in highly attenuated regions of metallic objects). Furthermore, this effect is sensitive to the geometric instability and geometric calibration errors of the imaging system (especially for mobile CBCT systems, which have poorer mechanical stability compared to fixed gantry CT or CBCT systems).

[0007] Recognizing the significant challenges associated with metal artifacts, numerous methods have been proposed and implemented for metal artifact removal (MAR), each with different advantages and disadvantages. MAR algorithms are typically software-based and can be broadly categorized as follows: (1) modifying projection data measured in the metal-affected area by correcting detector pixel values ​​(e.g., interpolation / patch). This type of approach is often referred to as projection domain metal artifact correction (PMAC); (2) invoking a physical model of the artifact source (e.g., X-ray beam hardening) and using model-based iterative image reconstruction (MBIR) to improve image quality; and (3) combining (1) and / or (2) to utilize prior information about the metal object (e.g., the exact shape of the object) or the patient (e.g., previous CT / MRI scans).

[0008] The disadvantages of this MAR method include: (1) errors in the localization of the metal-affected area and / or detector pixel value correction; (2) errors in other physical processes of X-ray beam modeling or artifact sources; (3) the requirement for prior information about the metal object or patient that may be unavailable or difficult to obtain; and (4) the potential for significant computational burden if iterative algorithms are involved. Despite decades of research and development of the MAR method and its deployment in various commercially available clinical CT and CBCT systems, metal artifacts remain a major source of image degradation.

[0009] Techniques to overcome the aforementioned shortcomings need to be developed. Furthermore, this method performs better when the quality of the x-ray projection data acquired in 3D scanning is high—that is, when the x-ray projection data is less affected by noise, errors, and / or biases associated with metallic objects in the volume of interest. In other words, the MAR method performs better when fewer artifacts need to be corrected. This invention provides a method to achieve this objective, which provides x-ray projection data less affected by metallic objects (i.e., carrying reduced noise, errors, and / or biases), such that the resulting 3D image reconstruction will exhibit reduced metallic artifacts. This may eliminate the need for MAR and / or allow the application of the MAR method to be more effective than when projection data is acquired in a conventional manner. Summary of the Invention

[0010] Based on examples from this disclosure, a method for avoiding metal artifacts in 3D X-ray imaging is provided. The method includes determining the 3D location of the metal within an object or volume of interest to be scanned; estimating a source-detector trajectory that will reduce the severity of the metal artifacts; moving the imaging system to a location consistent with the estimated source-detector trajectory; and scanning the object according to that source-detector trajectory. In some examples, the estimated location may be infeasible because the system is typically unaware of existing spatial constraints (such as stage location, patient size, tools, etc.). Therefore, the system can allow the user to move the imaging system to an optimal location with respect to the estimated location and spatial constraints.

[0011] Various additional features, including the following, can be implemented in the computer-implemented method. The determination may further include one or more of the following: performing an initial 3D scan of the object or volume of interest; acquiring one or more X-ray projection images of the object or volume of interest; using one or more previously acquired X-ray images; or using a tracking system including one or more cameras or electromagnetic trackers to locate the metal. The source-detector orbit may include the position and orientation of the X-ray source and detector for acquiring projections in the 3D X-ray imaging. Estimating the source-detector orbit further includes calculating an objective function based on the determined 3D position of the metal, wherein the objective function describes a characteristic associated with metal artifacts in the 3D image reconstruction. This characteristic may include estimates of spectral shift, attenuation, or a combination thereof. The objective function may be based on one or more of the following: the standard deviation of a metric plot along the gantry rotation axis, the maximum value of the metric plot along the gantry rotation axis, and the sum of metric plots over the gantry rotation angle range. The severity of metal artifacts can be reduced compared to a circular path in a plane perpendicular to the major axis of the object. The scan may further include acquiring multiple X-ray projections along the source-detector orbit and forming a 3D image reconstruction of the object. Forming the 3D image reconstruction may further include performing one or more 3D image reconstruction algorithms, including 3D filtered backprojection or model-based image reconstruction.

[0012] According to examples in this disclosure, a 3D x-ray imaging system is provided. This 3D x-ray imaging system may include: a 3D x-ray imaging apparatus including a gantry capable of moving at multiple tilt angles along a tilt axis and multiple rotation angles along a rotation axis; and a hardware processor configured to execute instructions including: determining the 3D position of a metal within an object or volume of interest to be scanned; estimating a source-detector trajectory that would reduce the severity of metal artifacts; moving the imaging system to a position consistent with the estimated source-detector trajectory; and scanning the object according to the source-detector trajectory.

[0013] Various additional features, including the following, can be implemented in the computer-implemented method. The determination may further include one or more of the following: performing an initial 3D scan of the object or volume of interest; acquiring one or more X-ray projection images of the object or volume of interest; using one or more previously acquired X-ray images; or using a tracking system including one or more cameras or electromagnetic trackers to locate the metal. The source-detector orbit may include the position and orientation of the X-ray source and detector for acquiring projections in the 3D X-ray imaging. Estimating the source-detector orbit may further include calculating an objective function based on the determined 3D position of the metal, wherein the objective function describes a characteristic associated with metal artifacts in the 3D image reconstruction. This characteristic may include estimates of spectral shift, attenuation, or a combination thereof. The objective function may be based on one or more of the following: the standard deviation of a metric plot along the gantry rotation axis, the maximum value of the metric plot along the gantry rotation axis, and the sum of metric plots over the gantry rotation angle range. The severity of metal artifacts can be reduced compared to a circular path in a plane perpendicular to the major axis of the object. The scan may further include acquiring multiple X-ray projections along the source-detector orbit and forming a 3D image reconstruction of the object. Forming the 3D image reconstruction may further include performing one or more 3D image reconstruction algorithms, including 3D filtered backprojection or model-based image reconstruction.

[0014] According to an example of this disclosure, a non-transitory computer-readable medium is provided, comprising instructions configured, when executed by a hardware processor, to perform a method for avoiding metal artifacts in 3D x-ray imaging, the method comprising: determining the 3D position of a metal in an object or volume of interest to be scanned; estimating a source-detector trajectory that would reduce the severity of the metal artifacts; moving an imaging system to a position consistent with the estimated source-detector trajectory; and scanning the object according to the source-detector trajectory.

[0015] According to an example of this disclosure, a computer-implemented method for avoiding metal artifacts in computed tomography (CT) imaging is provided. The computer-implemented method includes: acquiring multiple initial views of a target location of an imaged volume to provide initial guidance when performing CT imaging using a CT imaging device; forming a coarse 3D attenuation map of the target location by a hardware processor; segmenting the coarse 3D attenuation map into multiple images with similar properties using a segmentation algorithm executed by the hardware processor; calculating a graphical representation by the hardware processor relating to the severity of metal artifacts in the target location; calculating by the hardware processor an objective function defined with respect to the tilt angle of the gantry of the CT imaging device to capture the severity of metal artifacts across a range of rotation angles of the gantry; and determining a desired tilt angle by the hardware processor based on the objective function for the gantry, at which a 3D image scan of the target location is performed to avoid metal artifacts.

[0016] Various additional features, including the following, can be implemented in the computer-implemented method. Multiple initial views may include multiple positioning scan views, wherein the positioning scan views are operational modes of the CT imaging equipment, where CT slices are cut to initially position the CT imaging equipment. A coarse 3D attenuation map can be formed by backprojecting each of the multiple positioning scan views. The desired tilt angle can be determined based on the minimum value of a determined objective function. The graphical representation may include a metric plot relating the rotation angle to the tilt angle of the gantry. The objective function may be based on the standard deviation of the metric plot along the rotation axis of the gantry. The objective function may be based on the maximum value of the metric plot along the rotation axis of the gantry. The objective function may be based on the sum of the metric plots covering the range of gantry rotation angles. The computer-implemented method may further include performing a 3D image scan of the target location based on the determined desired tilt angle. The computer-implemented method may further include providing the desired tilt angle to the operator of the CT imaging equipment performing the 3D image scan.

[0017] According to examples of this disclosure, a computed tomography (CT) imaging system is provided, which may include a CT imaging apparatus including a gantry capable of moving at multiple tilt angles along a tilt axis and multiple rotation angles along a rotation axis; and a hardware processor configured to execute instructions including: acquiring multiple initial views of a target location of an imaged volume to provide initial guidance when performing CT imaging using the CT imaging apparatus; forming a coarse 3D attenuation map of the target location; segmenting the coarse 3D attenuation map into multiple images with similar properties using a segmentation algorithm; calculating a graphical representation relating to the severity of metal artifacts in the target location; calculating an objective function defined with respect to the tilt angles to capture the severity of metal artifacts across a range of rotation angles of the gantry; and determining a desired tilt angle based on the objective function for the gantry, at which a 3D image scan of the target location is performed to avoid metal artifacts.

[0018] Various additional features, including the following, can be implemented in this CT imaging system. Multiple initial views can include multiple positioning scan views, wherein the positioning scan views are the operating modes of the CT imaging device, where CT slices are cut to initially position the CT imaging device. A coarse 3D attenuation map can be formed by backprojecting each of the multiple positioning scan views. The desired tilt angle is determined based on the minimum value of a determined objective function. The graphical representation includes a metric plot relating the rotation angle to the tilt angle of the gantry. The objective function is based on the standard deviation of the metric plot along the rotation axis of the gantry. The objective function can also be based on the maximum value of the metric plot along the rotation axis of the gantry. Alternatively, the objective function can be based on the sum of the metric plots covering the range of gantry rotation angles. The hardware processor can be further configured to execute instructions including performing a 3D image scan of the target location based on the determined desired tilt angle. The hardware processor can be further configured to execute instructions including providing the desired tilt angle to the operator of the CT imaging device performing the 3D image scan. Attached Figure Description

[0019] Embodiments of this teaching are illustrated in conjunction with the accompanying drawings, which are incorporated in and form a part of this specification, and together with the description, serve to explain the principles of this disclosure. In the drawings:

[0020] Figure 1A and Figure 1B A diagram illustrating an example MAA system and its geometry is shown. Tilt angle (φ) * It defines a circular orbital scan plane that can reduce metallic artifacts in 3D image reconstruction.

[0021] Figure 2A flowchart of an example MAA method based on this disclosure is shown.

[0022] Figures 3A to 3C An illustrative simulation study of the MAA method is presented. Figure 3A The volume rendering of a cylindrical water phantom containing various metal spheres and rods is shown. The phantom has two positional scan projection views (AP and lateral), corresponding to p(u, v|θ=0, φ=0) and p(u, v|θ=90, φ=0) in the MAA method. Figure 3B The backprojection volume (rough 3D attenuation map, denoted as μ) derived from two positioning scan views is shown. c ). Figure 3C It shows from Figure 3B The attenuation map in the image is used to derive a simple intensity-based segmentation of a high-density metallic object (referred to as a coarse 3D attenuation map after segmentation, denoted as μ). seg ). Figure 3C The illustration area shows a magnified view of one of the segmented regions.

[0023] Figure 4 The metric q(θ, φ) calculated from a view (simulated orthographic projection) of the full 3D attenuation map of the metal object is shown. q(θ, φ) (representing the number of BH artifacts associated with the metal, equation (1)) is calculated here for all possible C-arm gantry angles and tilts. The results shown here are calculated under the assumption that the metal object is fully known—that is, the full 3D attenuation map of the metal object rather than μ. seg

[0024] Figure 5 Showing according to Figure 4 The objective function Q(φ) is calculated from the metric plot q(θ, φ). The form of Q(φ) is shown here as equation (6). The minimum value of the objective function Q(φ) (Q(φ) = max q(θ, φ)) represents the tilt angle (φ) that best avoids metal artifacts in 3D image reconstruction. * (Marked by arrows 502 and 504).

[0025] Figure 6 Shown for Figure 3A The diagram illustrates how tilt angles can reduce metallic artifacts in the simulated phantom using the MAA method. 3D image reconstructions for various tilt angles (labeled at the top of each subplot, φ = 1-30 degrees) are shown here. The minimum value in Q(φ) is compared with... Figure 5 The images corresponding to arrows 502 and 504 are marked with asterisks 602 and 604, respectively, representing φ = 8 degrees and 27 degrees.

[0026] Figure 7A and Figure 7BThe MAA metric is shown, calculated based on only two positioning scan projection views. Figure 7A The metric graph q(θ, φ) is shown. Note that, with... Figure 4 The similarity (the calculated metric map is based on the assumption of a complete 3D attenuation map of the metallic object). The metric map calculated based on only two views can still successfully locate the view that will cause metallic artifacts, corresponding to high attenuation (hyperspectral shift). Figure 7B It shows that according to Figure 7A The objective function Q(φ) is calculated from the horizontal line in the metric plot. This again reproduces... Figure 5 The idealized results show the ideal tilt angles at approximately φ = 8° and φ = 27° (arrows 702 and 704).

[0027] Figures 8A to 8E The results of the physical phantom experiments using the MAA method are shown. Figure 8A Two projected positioning scan views are shown, p(u, v|θ=0, φ=0) and p(u, v|θ=90, φ=0). Figure 8B It shows that by... Figure 8A A rough 3D attenuation map (μ) is formed by unweighted back projection of the two positioning scan views. c ). Figure 8C Intensity-based segmentation (μ) of a coarse 3D attenuation map is shown. seg ). Figure 8D It shows the use of equation (1) by applying... Figure 8C of (μ) seg The metric graph q(θ, φ) is calculated by orthographic projection. Figure 8E It shows that according to Figure 8D The objective function Q(φ) is calculated using the horizontal line, where the minimum value indicates the tilt angle (φ) that will avoid metal artifacts. * ).

[0028] Figures 9A to 9F Axial views of the chest phantom acquired at different tilt angles are shown. Note the artifacts (light and dark stripes around the spinal screws) at φ=0°, which are reduced in images acquired at tilt angles of φ=20 or 30° predicted by the MAA method.

[0029] Figure 10A The coronal image of simulated phantom #2 is shown. Figure 10B The volume rendering of the simulated model #2 is shown (screws are shown).

[0030] Figure 11A A graph showing the q(θ, φ) metric calculated based on simulated phantom #2 is presented. The optimal non-circular orbit is marked by curve 1102. Figure 11BThe diagram shows the Q(φ) (semi-circular) target plotted at a series of gantry tilt angles. The non-circular Q(φ(θ)) achieves a lower value than any setting of the semi-circular track. Figure 11C An axial slice (screw #3, 0° exterior angle) is shown, imaged using optimal circular (top) and non-circular (bottom) orbits (reconstructed using penalized weighted least squares (PWLS) method). Figure 11D Three example screws (#4-6, with exterior angles of 0°, 11°, and 22° respectively) are shown, obtained using the optimal semi-circular (top) and non-circular (bottom) orbits. Note that the images obtained using the non-circular MAA method exhibit significantly reduced metallic artifacts. An oblique axial view across the axis of each screw is shown. Figure 11E The ground truth axial image of the simulated screw is shown.

[0031] Figure 12 An example of a method for avoiding metal artifacts in 3D X-ray imaging, according to this disclosure, is shown.

[0032] Figure 13A and Figure 13B Enlarged views of the mobile C-arm, C-arm geometry, and detector, as exemplified by this disclosure, are shown respectively. Nine parameters (degrees of freedom) determined via geometric calibration are labeled. (B) Source-detector track illustrated on a sphere. Circular tracks (various settings of gantry tilt φ) are shown. Example non-circular tracks (φ varying with θ) are shown. A scaled-down inset illustrates geometric parameter estimates for a view determined by interpolation of the four nearest vertices in a geometric calibration library of circular scans.

[0033] Figure 14 An example of an end-to-end neural network for 3D localization / segmentation of metal instruments, according to this disclosure, is shown.

[0034] Figure 15A and Figure 15B The degree of reproducibility of the system geometry of a predetermined non-circular orbit, as exemplified by this disclosure, is shown.

[0035] Figures 16A to 16D The impact of geometric calibration accuracy on spatial resolution (modulation transfer function, denoted as MTF) and 3D image quality is shown. Figure 16A It shows the method for using by Figure 13A and Figure 13B The MTF for the three cases defined in the layout, and Figure 16B , Figure 16C and Figure 16D The axial and sagittal magnified views (skull base) of the head phantoms for the three cases are shown respectively.

[0036] Figure 17A, Figure 17B and Figure 17C It shows that in the form of Figure 13A and Figure 13B The expected reduction in cone-beam artifacts generated by non-circular orbits in the sagittal image of the head phantom containing a stack of flat disks, as defined by the arrangement. The apparent thickness (full width at half maximum, representing FWHM) of the edge of the uppermost disk is reduced from 9 mm in the standard “circular case” to the true thickness (approximately 5.5 mm) in its two non-circular cases.

[0037] Figure 18 The performance of segmenting real or simulated metal implants using the Dice coefficient (denoted as DC) is shown in validation tests for three network types (single U-Net, dual U-Net, and end-to-end) as a function of the number of projected views.

[0038] Figures 19A to 19E The segmentation performance of three neural networks is shown on an example test dataset (cadavers with 12 pedicle screws) according to this disclosure. Figures 19A to 19C An example axial slice segmentation overlaid with ground truth is shown. Segmentation for different numbers of positioning scan views is also shown. Figure 19D The end-to-end method is shown to calculate μ based on two views. seg The isosurface. Figure 19E The isosurface of the ground truth segmentation is shown (which is downsampled to μ). seg ).

[0039] Figures 20A to 20F The MAA method, as exemplified in this disclosure, is applied to a chest phantom with eight pedicle screws implanted. Figure 20A A graph of the q(θ,φ) metric superimposed on the optimal non-circular orbit is shown. Figure 20B The amplitude of the halo artifact (FWHM of the screw axis) for each of the eight screws is shown. Figure 20C and Figure 20E Axial images of (medium) circular scans and (MAA) non-circular scans show improvements in visual image quality acquisition, which (in Figure 20D and Figure 20F (The image is further illustrated by an enlarged in-plane quasi-axial slice of each screw.)

[0040] Figure 21 Another method, as illustrated in this disclosure, is shown as a method for avoiding metal artifacts in 3D X-ray imaging.

[0041] Figure 22 This illustrates yet another method for avoiding metal artifacts in 3D X-ray imaging using a CNN network, as exemplified by examples of this disclosure.

[0042] Figure 23 This is an example of a hardware configuration for a computer device based on the examples in this disclosure.

[0043] It should be noted that some details in the accompanying drawings have been simplified in order to facilitate understanding of this teaching, rather than to maintain strict structural accuracy, detail, and proportion. Detailed Implementation

[0044] In general, the examples in this disclosure provide a system and method for metal artifact avoidance (MAA) that differs from and improves upon PMAC and MBIR, does not assume prior information about the patient with the metallic implant, can be used in conjunction with PMAC and / or MBIR (if desired), and adds little or no computational burden on its own. Specifically, this system and method provide MAA by adapting CT or CBCT image acquisition in a manner that reduces the impact of the metallic object on the projected data and thus reduces the severity of metal artifacts. This system and method may not necessarily provide “minimal” or “zero” metal artifacts, but it provides fewer artifacts than conventional scans.

[0045] The differences between this disclosure and previous PMAC methods are as follows: (1) This system and method are hardware-based solutions that attempt to avoid metal artifacts during image acquisition rather than correcting them in the post-acquisition stage. This is more robust than software-based methods and can be easily integrated into current CBCT systems without significant modifications to the artifact correction and reconstruction pipeline—that is, the MAA method is consistent with existing PMAC methods (and can be used in combination), and (2) the information required for automated trajectory design is obtained through a small number (e.g., two) of localization scan views. Unlike some of the methods described above, although prior information about metallic objects or the patient may be included, for example, to improve… Figure 2 The segmentation step in step 208 is not required by the proposed solution, but this prior information is not necessary. The method for determining the gantry tilt angle is based on at least two positioning scan views. Using the method described below, the tilt angle of the MAA (Mountain Angle) using a CBCT imaging system can be calculated based on at least two positioning scan views. The determination of the tilt angle is formulated as a simple optimization problem (with respect to φ) aimed at reducing metal artifacts. For simplification and computational efficiency, the problem is formulated in the projection domain.

[0046] Typically, CT or CBCT systems acquire projected image data via a circular track—that is, the X-ray source and detector travel in a circle around the patient—and the plane containing this circular track is orthogonal to the patient's major axis. Please note the following three points regarding the source-detector track:

[0047] (i) Previous work has reported various imaging approaches involving non-circular orbits—for example, expanding the field of view (FOV) to reduce “cone-beam artifacts” (which are quite different from metal artifacts) or maximizing spatial frequency sampling for a specific imaging task. Compared to simple 3D filtered back projection (which is applicable to circular orbits and is the mainstay of CT or CBCT image reconstruction methods), non-circular orbits typically involve more complex analytical or iterative (MBIR) methods.

[0048] (ii) Furthermore, it is not uncommon to use circular tracks tilted relative to the patient's long axis to acquire CT or CBCT data. For example, in CT diagnosis of the head, the CT scanner gantry is sometimes tilted along the canthal line to reduce beam hardening effects associated with small bones of the skull. This gantry tilting technique is very specific to imaging the skull and is designed to overcome high attenuation in common anatomical sites (and not to avoid metal artifacts). Tilted (circular) tracks are still consistent with 3D filtered back projection (referred to as FBP) and other analytical reconstruction techniques. Non-circular tracks may not be consistent with basic 3D FBP or other analytical reconstruction algorithms. Utilizing knowledge of system geometry, both circular and non-circular tracks can generally be consistent with typical 3D model-based iterative reconstruction (MBIR). Both circular and non-circular tracks can be reconstructed using deep learning (DL) neural networks, convolutional neural networks (CNNs), or other relevant methods. The MAA method is applicable to all such reconstruction and post-processing methods because it applies to the data acquisition process, not the reconstruction process itself.

[0049] (iii) The MAA method adapts the source-detector orbit in a manner that specifically minimizes the impact of metallic objects on the projected data (e.g., signal bias associated with the attenuation of the X-ray beam by metallic objects). A non-limiting example described below involves a tilted circular (or semi-circular) orbit [as shown in (ii)], with the tilt angle selected in a manner that specifically minimizes the impact of highly attenuated metallic objects on the projected data (thus helping to avoid metallic artifacts in 3D image reconstruction).

[0050] In a non-limiting example, MAA involves a tilted circular track whose tilt angle is determined by a small number of projection views (e.g., two views), referred to as the "scout" views. From as few as two scout scan views, a tilted circular track can be determined that minimizes the influence of metal on the projection data—that is, avoids the generation of metal artifacts—rather than correcting for them using PMAC or reducing their impact on image reconstruction using MBIR. In other words, scans acquired using the MAA method remain compatible with PMAC and / or MBIR and can similarly benefit from the functionality of both.

[0051] Figure 1A and Figure 1BAn illustration of an example system and geometry of MAA based on this disclosure is shown. Figure 1A A CT system 102 is shown with a gantry 104 in the form of a movable C-arm or O-arm, the gantry being movable in at least two degrees of freedom, expressed as tilt angles (φ). * 106 and rotation angle (θ) 108. Tilt angle (φ) * )106 defines a circular orbital scan plane that reduces metallic artifacts in 3D image reconstruction. The tilt angle can vary between -30° and +30°. The rotation angle can vary between 0° and 196°. Because the C-arm can electrically control both the tilt and rotation angles, non-circular orbits can be performed during scanning via computer-controlled variations in the tilt and rotation angles. For example, a movable C-arm or O-arm is a typical example of a CBCT imaging system with tilted circular trajectory capabilities. For example, a movable C-arm or O-arm is a typical example of a CBCT imaging system that may have an X-ray tube, a computer-controlled tube output, and an X-ray detector (e.g., a flat panel detector, FPD) with a detector area sufficient to cover the volume of interest in a 3D scan. The system geometry is typically defined by the source-detector distance (SDD) and source-axis distance (SAD), resulting in a volumetric field of view (FOV) covering the volume of interest. A nominal scanning scheme includes multiple projections acquired over a scan arc of 180° or greater at specific settings of the X-ray tube output and scan time. Systems used in this disclosure include isocentric or non-isocentric C-arms, O-arms, fixed-space C-arms (e.g., Axiom Artis Zee), or ring-type gantry diagnostic CT scanners with gantry tilting capability (e.g., Somatom Definition).

[0052] Figure 1A The coordinate system (x,y,z) 110 for the scanned object 112 and the coordinate system (u,v) 114 for the plane of the gantry 104 are also shown. Figure 1A The normal (non-tilted) circular trajectory (tilt angle φ = 0) shown by circle 114 and the tilted circular trajectory shown by circle 116 are also shown. Figure 1B A diagnostic system 114, which can be coupled to a CT system 102, is shown as an example according to this disclosure. A control system 118 may include a computer system for processing and displaying the results of the CT system 102 using the methods further described below.

[0053] Figure 2A computer-implemented method 200 for avoiding metal artifacts in CT imaging, as exemplified by this disclosure, is shown. Before starting method 200, the patient and metal instruments are positioned on the stage of the CT system 102, and a scan is instructed at 202.

[0054] Method 200 continues at point 204 along a normal (non-inclined) circular trajectory (inclination angle φ = 0, as shown below). Figure 1A As shown in circle 114, multiple (e.g., two or more) positioning scan views are obtained. The tilt angle φ is determined... * Then, using this inclined circular trajectory (such as...) Figure 1A A 3D scan of the patient is performed (as shown in circle 116). Two positioning scan views (e.g., a side view and an AP view, 90 degrees apart) are acquired in the case of a nominally non-tilted circular geometry. The projected view acquired at specific θ and φ is denoted as p(u, v|θ). i φ i By this definition, the two localization scan views (i = 1, 2) can be represented as p(u, v|θ = 0, φ = 0) and p(u, v|θ = 90, φ = 0). These localization scan views can be acquired at very low doses because they only need to depict the area associated with the metallic object. Acquiring images of standard localization scan views at the start of the procedure is typical and does not impose additional workflow or dose on the patient.

[0055] Method 200 continues at 206 by backprojecting the positioning scan view acquired at 204 to form a coarse 3D attenuation map (μ). c To locate dense anatomical structures and metallic objects based on the acquired localization scans, the two localization scans were back-projected to form a coarse 3D attenuation map (μ). c Then, the coarse 3D attenuation map is segmented to form a segmented coarse 3D attenuation map (μ) containing only high-attenuation voxels. seg Segmentation can be performed using simple segmentation methods (such as adaptive thresholding) or more advanced methods based on machine learning, deep learning, convolutional neural networks (CNNs).

[0056] Method 200 continues to segment the coarse 3D attenuation map at position 208 (μ seg This is used to locate particularly dense anatomical structures and / or high-density metallic objects. It is used to determine μ from a coarse attenuation map. segExemplary methods include (but are not limited to) intensity thresholding, region growing, and classification based on artificial neural networks. Such segmentation methods can operate with or without prior information about the size, shape, number, and / or material content of metallic objects known or believed to be present in the patient's body. The tilt angle φ for minimizing metal artifacts is from μ seg Yes, that's for sure. The following details the process from μ... seg An exemplary method for determining φ involves an objective function, which is selected in the projection domain for a specific choice of φ. * The objective function is minimized. Note that the objective function can be minimized through exhaustive search (without iterative optimization).

[0057] In a non-restrictive example, such as Figure 2 As shown, the operations described in 204, 206, and 208 are performed serially. This first example is... Figure 18 Figure 19 is referred to as "Single U-Net," where the operation of Single U-Net is... Figure 2 The Chinese character is marked with 208.

[0058] In another non-limiting example, the localization scan view can be segmented to define regions with dense anatomical structures and regions with high-density metallic objects, and then back-projected to form μ. c This method can potentially improve μ. seg The accuracy of the segmentation operation is high, but it is often challenged by the large dynamic range of signals in the overlapping structures and projection data. Several methods (e.g., KC-Recon)

[15] have been developed to overcome these challenges. Figure 2 The second example is performed between the operations described in sections 204 and 206. Figure 18 The one in Figure 19 is called "Double U-Net". Figure 21 Method 2100, illustrating an example of a dual U-Net, is shown. Figure 21 As shown, operations 2102 and 2104 are identical to operations 202 and 204. After operation 2104, which acquires multiple localization scan views, as shown in 2106, the localization scan views are segmented in the projection domain to contain only particularly dense anatomical structures and / or high-density metallic objects. Then, as shown in 2108, method 2100 continues to form a coarse 3D attenuation map by backprojecting the localization scan views. Then, as shown in 2110, method 2100 continues to segment the coarse 3D attenuation map in the image domain. Operations 2104 and 2110 represent the “dual U-Net” nature of this method. Then, method 2100 continues to perform the same operations 2112, 2114, 2116, and 2118 as described in operations 210 through 216.

[0059] In yet another non-limiting example, a coarse 3D attenuation map μ can also be determined via FBP reconstruction of more sparse views (more views compared to other examples, such as 10 views). c FBP does not perform a simple backprojection; instead, it "filters" the projected data before backprojection. According to existing research, filtering does not improve performance for only two localization scan views, but it does improve performance for a larger number of localization scan views. c Then it can be divided to form the same μ as the first example discussed above. seg .

[0060] In yet another non-restrictive example, one can directly use, such as Figure 18 The “end-to-end” convolutional neural network (CNN) method, shown in Figure 19 and discussed further below, determines the coarse 3D attenuation μ after segmentation from the localization scan view. seg This method allows the network to know the output without explicitly producing intermediate results—that is, without calculating μ. c As an intermediate step. Figure 14 An example design of an "end-to-end" CNN is described in the document. In this example, the operations described in sections 206 and 208 are performed in a single operation. Figure 22 Method 2200, illustrating an example of an end-to-end method, is shown. For example... Figure 22 As shown, operations 2202 and 2204 are identical to operations 202 and 204. After operation 2204, which acquires multiple localization scan views, the "end-to-end" CNN determines a coarse 3D attenuation map after segmentation, as shown in 2206 and further described below. Then, method 2200 continues to perform the same operations 2208, 2010, 2112, and 2114 as described in operations 210 through 216.

[0061] Method 200 continues at 210 based on segmentation (μ) of a coarse 3D attenuation map. seg The metric map is calculated using q(θ, φ). The metric map (denoted as q(θ, φ)) is defined in the projection domain and relates to the severity of metallic artifacts that will lead to 3D image reconstruction. Note that q(θ, φ) is defined for each projection view (θ) at each tilt angle (φ), and these attenuation maps are calculated by adjusting μ according to the geometry of the actual imaging system. seg Generate by orthographic projection.

[0062] The metric map (denoted as q(θ, φ)) is defined to relate to the severity of metallic artifacts that will lead to 3D image reconstruction. Note that q(θ, φ) is defined for each projected view in the projection domain (represented by the detector's 2D spatial coordinates (u, v)) to avoid time-consuming searches in the 3D reconstruction domain. Three example definitions of q(θ, φ) are described below.

[0063] Example 1 (Multi-energy conversion): Calculate the amount of metal (high-density object) causing beam hardening (BH) artifacts (spectral shifts) in the projection domain for each rotation angle (θ) and tilt angle (φ):

[0064]

[0065] Where, p mono (u, v|θ, Φ) is the integrated value of the analog monoenergetic line at the detector pixel index (u, v), which is the result of the integration of μ in the previous step. seg The orthographic geometry is determined by orthographic projection, and is defined by θ and φ. Since the orthographic projection operator is monoenergetic, therefore p mono (u, v|θ, Φ) has no BH artifacts. Therefore, the simulated beam hardening artifacts are added using the following formula to generate the corresponding BH-contaminated multi-energy line integral p. poly (u, v|θ, Φ):

[0066]

[0067] Where, α i The polynomial coefficients are determined by the incident spectrum from the X-ray tube and the attenuation properties of the metallic object.

[0068] Example 2 (Maximum Attenuation): Calculate the sum of k maximum monoenergetic line integrals for each θ and φ:

[0069]

[0070] Where, p mono (u, v|θ, Φ) is the analog monoenergetic line integral value at the detector pixel index (u, v) as described above, and maxk is the operator that takes the k maximum values ​​from the independent variables.

[0071] Example 3 (Monoenergetic Threshold): Calculates the sum of the integral values ​​of monoenergetic lines above a certain threshold for each θ and φ. Thresholding can help isolate the influence of high-density metallic objects from the influence of low-density anatomical background (soft tissue, etc.) in the projective domain.

[0072]

[0073] Where, p mono (u, v|θ, Φ) is the analog monoenergetic line integral value at the detector pixel index (u, v) as described above, and Thresh is an operator that takes a value higher than a certain threshold from the independent variable.

[0074] Example 4: Calculate the sum of the monoenergetic line integrals over the entire projected view for each θ and φ.

[0075]

[0076] Where, p mono (u, v|θ, Φ) is the analog monoenergetic line integral value at the detector pixel index (u, v) as described above.

[0077] Of course, alternative formulas for the metric graph q(θ, φ) can be constructed—for example, logarithmic transformations of any of the examples above.

[0078] Example 1 (Equation (1)) is a non-limiting example of the MAA solution described herein. Simulation and phantom experiments show that the formula provides a reliable way to avoid metal artifacts, especially the beam-hardened portion of metal artifacts.

[0079] A simulation study was conducted to illustrate the disclosed method. This simulation simulated... Figure 1A The geometry and X-ray spectrum of the CBCT system are shown. The digital phantom used in this simulation is a water column containing several metal spheres and cylindrical rod inserts. Figure 3A The volumetric rendering of the phantom and the localization scan views (AP and lateral) of the two markers of the phantom are shown, corresponding to p(u, v|θ=90, φ=0) and p(u, v|θ=90, φ=0) respectively. In this example, the two metal spheres overlap in the AP view, which will result in particularly low fidelity of the projected data in this area and will lead to severe metal artifacts in 3D image reconstruction. Figure 3B A coarse 3D attenuation map (coarse 3D attenuation map, denoted as μ) is shown, formed using volumes back-projected from two positioning scan views. c ). Figure 3C It shows Figure 3B The attenuation map is the result of simple intensity-based segmentation. As described in this paper, based on the segmented coarse 3D attenuation map (μ seg The calculation will avoid tilted tracks of metal artifacts as much as possible. Figure 3C The illustration area shows a magnified view of one of the segmented regions.

[0080] For illustrative purposes, use Figure 1A The calculation of the metric map q(θ, φ) for the simulated phantom is shown below, assuming that the metal object is known, i.e., the complete 3D attenuation map of the metal object rather than the segmented coarse 3D attenuation map (μ). seg This article discusses a more practical case where only two views of the unknown phantom are included (i.e., using μ). seg ).

[0081] For the simulated object in Figure 3, the calculation of q(θ, φ) according to equation (1) is performed in... Figure 4As shown in the diagram, the bright areas indicating θ and φ will result in severe metal artifacts (low-fidelity view) and should therefore be avoided. The simplicity of the MAA solution is also noted: the ideal source-detector tilt angle is simply defined by the lower horizontal lines (circular trajectories) in the q(θ, φ) plot, and some horizontal lines (i.e., some tilt angles) will obviously suffer higher attenuation (metal artifacts) than others. Find the tilted circular geometry of φ. * (Minimizing the metal artifacts) is equivalent to finding a horizontal line in the q(θ,φ) plot that minimizes the objective function (denoted as Q(φ)) with respect to the tilt angle φ.

[0082] Figure 4 The metric q(θ, φ) calculated from a view (simulated orthographic projection) of the full 3D attenuation map of the metal object is shown. q(θ, φ) (representing the number of BH artifacts associated with the metal, equation (1)) is calculated here for all possible C-arm gantry angles and tilts. The results shown here are calculated under the assumption that the metal object is fully known—that is, the full 3D attenuation map of the metal object rather than μ. seg .

[0083] With q(θ, φ) ready, the objective function Q can be calculated. This calculation can be divided into the following two cases:

[0084]

[0085] (Case I): If φ(θ) is constant for all θ, then the inclined circular orbit (track) is in Figure 4 The middle corresponds to the horizontal line, making Figure 4 Each horizontal straight line (from left to right) in the diagram corresponds to a (slanted) circular track, and each curve (from left to right) corresponds to a non-circular track, for example, as shown in Figure 11.

[0086] (Case II): If φ(θ) is not constant for all θ, then the non-circular orbit (track) is in Figure 4 The middle corresponds to the curve (e.g., line 1102 in Figure 11).

[0087] The optimal tilt angle for a circular track on a gantry crane: The objective of the above equation can be simplified to finding the tilted (circular / semi-circular) track, denoted as Q(φ). This simplification is equivalent to finding the q(θ, φ) measurement plot with the minimum (or reduced) standard deviation—that is, the scalar value φ(φ *The "horizontal" row, which can be easily solved by exhaustively searching a finite variable space, is a key feature. However, there are several potential challenges associated with increasingly larger tilt angles, including an increased risk of collisions with the control panel and reduced reproducibility of geometric calibration. Therefore, the "optimal" tilt angle corresponds to a φ value that is as small as possible but large enough to avoid or reduce metallic artifacts.

[0088] Non-circular orbits: A growing number of modern C-arm systems are capable of non-circular orbits—for example, large fixed space robot C-arms (such as Artis Pheno, Siemens Healthineers) or even the mobile C-arm shown in Figure 1 (Cios Spin 3D), which provides computer control over θ and φ. This opens up the possibility of fully optimizing φ(θ) (i.e., the non-circular orbit) to achieve a lower objective function (Q(φ(θ)) than the simple inclined circular orbit (Q(φ(θ))) described above. To reduce the dimensionality of the problem and encourage smooth and realistic orbits, we model φ(θ) as a superposition of cubic b-spline kernels, each kernel being one of N = 10 sections (at θ) i The points are evenly distributed along the scanning arc, centered on the point:

[0089]

[0090] Where B is a cubic b-spline and fi is the control parameter of section i. The optimization in equation (7) is usually non-convex and may be difficult to solve using traditional gradient-based methods. An exemplary approach to solving the minimum of equation (7) is the Covariance Matrix Adaptive Evolution Strategy (CMA-ES), as described by N. Hansen and S. Kern in Evaluating the CMA Evolution Strategy on Multimodal Test Functions in Parallel Problem Solving from Nature-PPSN VIII (edited by X. Yao, EKBurke, JALozano et al., pp. 282-291), as a robust solver for equation (7) that runs almost in real time.

[0091] Method 200 continues at 212 by calculating the objective function Q(φ) defined with respect to the tilt angle φ, to capture the overall severity of metal artifacts across all θ. The optimal tilt angle (denoted as φ) for avoiding metal artifacts is then calculated by searching all possible gantry tilt ranges (φ).* To minimize Q(φ). It is important to note that a strict minimum of Q(φ) may not be necessary. For example, in some cases, Q(φ) may monotonically decrease with φ (without a minimum). In such cases, a value of Q(φ) that is as small as possible but large enough to avoid metal artifacts should be found. In this sense, the angle φ * The choice is not really about “minimizing” or “optimizing”, but rather requires important practical considerations (i.e., “as small as possible, but large enough if necessary”).

[0092] The objective function Q(φ) is defined with respect to φ to capture the overall severity of metallic artifacts across all θ. Some example forms of the Q(φ) definition are as follows:

[0093] Example 1 (Changes in the metric graph):

[0094] Q(φ)=σ θ [q(θ,φ)] (8)

[0095] Where, σ θ It is the standard deviation along the θ axis.

[0096] Example 2 (Maximum value of the metric graph):

[0097] Q(φ)=maxq(θ,φ) (9)

[0098] The "max" operator takes the maximum value along the θ axis.

[0099] Example 3 (Sum of metric graphs):

[0100]

[0101] The summation operator measures the sum along the horizontal lines in the graph.

[0102] Alternative formulas for Q(φ) can also be constructed. Example 1 (e.g., equation (8)) is a non-limiting example of the MAA solution described herein, but there may be other cases where the Q objective might be more suitable. Figure 4 The q(θ, φ) graph, using equation (8) to calculate Q(φ) Figure 5 As shown in the image.

[0103] Figure 5 Showing according to Figure 4 The objective function Q(φ) is calculated from the metric plot q(θ, φ). The form of Q(φ) is shown here as equation (8). The minimum value of the objective function Q(φ) (Q(φ) = maxq(θ, φ)) represents the tilt angle (φ) that best avoids metallic artifacts in 3D image reconstruction. * (Marked by arrows 502 and 504).

[0104] Method 200 continues to determine the tilt angle φ for performing the 3D imaging scan at point 214: angle φ * With a tilted circular trajectory (tilted by φ) * Perform a 3D scan of the patient. The optimal tilt angle (φw) to minimize metal artifacts is calculated by searching for possible gantry tilt ranges (φ). For example, a C-arm gantry can typically provide reliable imaging and geometric calibration within a φ range of -30° to +30°. Figure 5 The range is shown. Note that due to the simplicity of the projection domain formulas for q(θ, φ) and Q(φ), exhaustively searching for all possible φ is simpler and more reliable than full iterative optimization. This search for all possible gantry tilts (φ) is equivalent to simply considering... Figure 4 All horizontal lines (metric graph q(θ, φ)) are defined, and Q(φ) is calculated for each horizontal line. That is, Figure 4 The horizontal line that minimizes Q(φ) represents the desired tilt angle (φ) of the MAA. * ).

[0105] exist Figure 5 In the diagram, arrow 502 (φ approximately 8°) and arrow 504 (φ approximately 27°) illustrate two possible tilt angles that minimize Q(φ). Figure 4 As can be seen in the metric plot q(θ, φ), these angles avoid the low-fidelity view (“bright spots”) associated with the metal spheres and rods in the phantom.

[0106] Method 200 may continue to perform collision checks on CT system 102 and / or gantry 104 at 216 and perform 3D imaging scans at the desired tilt angle to obtain 3D images that avoid metal artifacts.

[0107] Figure 6 The image shows images taken at different tilt angles. Figure 3A 3D image reconstruction of the phantom (axial slices). Half the possible tilt angles are shown, marked at the top of each sub-image as φ = 1°, 2°, ... 30°. Metal artifacts associated with the two metal spheres in the axial plane are clearly visible as dark horizontal stripes in the images. Tilting angles with more severe metal artifacts are φ = 1°, 2°, ... Note the two images marked with asterisks 602 and 604 at φ = 8° and φ = 27° respectively. These tilt angles correspond to... Figure 5 Arrows 502 and 504 (minimum values ​​of Q(φ)) are used to predict angles that can avoid metal artifacts. Figure 6 The illustration shows that these angles do indeed avoid metal artifacts.

[0108] Figure 6 It shows that for Figure 3A The diagram illustrates how tilt angles can reduce metallic artifacts in the simulated phantom using the MAA method. 3D image reconstructions for various tilt angles (labeled at the top of each subplot, φ = 1-30 degrees) are shown here. The minimum value in Q(φ) is compared with... Figure 5 The images corresponding to arrows 502 and 504 are marked with asterisks 602 and 604, respectively, representing φ = 8 degrees and 27 degrees.

[0109] For the purpose of explanation, Figure 4 , Figure 5 and Figure 6 The results illustrate the current method in an idealized (impractical) scenario where a complete 3D attenuation map of the metallic object is known. This is merely to provide the reader with an illustrative understanding of the basic principles. A non-limiting embodiment achieves the same effect based on as few as two positioning scan views without knowing the attenuation map of the phantom. Figure 7A and Figure 7B The results based on two positioning scan projection views are shown. First, two projection positioning scan views of the phantom are obtained (e.g., Figure 3A The AP and side view in the image correspond to p(u, v|θ=0, φ=0) and p(u, v|θ=90, φ=0). Unweighted backprojection is performed on these two positioning scan views to form... Figure 3B The rough 3D attenuation map shown (μ) c Then, via as Figure 3C The simple intensity thresholding shown segments the coarse 3D attenuation map, corresponding to μ. seg For the example phantom, this yields the positioning (not the exact shape) of two strongly attenuated metallic objects.

[0110] Then by adjusting μ seg Orthographic projection is used to determine the metric map q(θ, φ). The results are in... Figure 7A As shown in the image. Note again the bright spots; these correspond to views where strong attenuation and metallic artifacts may be expected. Also note... Figure 7A and Figure 4 Differences between them: Figure 7A The metric map q(θ, φ) is based solely on two localization scan views (based on μ). seg ) calculated, and Figure 4 The metric map q(θ, φ) is calculated based on complete knowledge of the metallic object in the digital phantom (a complete 3D attenuation map of the metallic object). This is done even though only two views are used to estimate μ. seg This method correctly identified the projected views associated with strong metal artifacts.

[0111] Then the objective function Q(φ) can be determined based on the metric graph q(θ, φ), such as Figure 7BAs shown. Here again, note... Figure 7B and Figure 5 The difference between the Q(φ) targets. Figure 7B The target was calculated based on only two positioning scan views, but it correctly identified the ideal tilt angles of approximately 8° and 27°. (via μ) seg (As shown in Figure 7) These determined metrics (q(θ, φ) and Q(φ)) are compared with a complete 3D attenuation map of the metallic object (as shown in Figure 7). Figure 4 and Figure 5 The measurements determined (as shown) are relatively consistent, indicating the feasibility of determining the required tilt angle using only two positioning scan views.

[0112] Figure 7A and Figure 7B The MAA metric is shown, calculated based on only two positioning scan projection views. Figure 7A The metric graph q(θ, φ) is shown. Note that, with... Figure 4 The similarity (the calculated metric map is based on the assumption of a complete 3D attenuation map of the metallic object). The metric map calculated based on only two views can still successfully locate the view that corresponds to high attenuation and will cause metallic artifacts. Figure 7B It shows according to Figure 7A The objective function Q(φ) is calculated from the horizontal line in the metric plot. This again reproduces... Figure 5 The idealized results show the ideal tilt angles at approximately φ = 8° and φ = 27° (arrows 502 and 504, respectively).

[0113] use Figure 1A The CBCT system with a movable C-arm shown was used in a physical phantom experiment to further test the method. As shown in Figure 9, an anthropomorphic chest phantom was used in this study, which included two bilateral pedicle screws placed near the spine.

[0114] Obtain two positioning scan projection views of the phantom—such as Figure 8A As shown, this corresponds to p(u, v|θ=0, φ=0) and p(u, v|θ=90, φ=0). Then, by... Figure 8A These two positioning scan views are back-projected (unweighted) to determine a coarse 3D attenuation map [μ]. c , Figure 8B Then, the coarse 3D attenuation map is segmented by intensity thresholding to generate... Figure 8C The coarse 3D attenuation map after segmentation (μ) seg ).

[0115] As explained above, by adjusting μ seg (c) Perform orthographic projection to calculate the metric graph q(θ, φ) (using equation (1)). Figure 8D The image in the diagram shows a bright spot at the center of the measurement plot, θ, approximately 100°, and φ, ranging from -15° to 15°. This means that almost any tilt angle (φ greater than approximately 15° or less than approximately -15°) will largely avoid metallic artifacts on the phantom.

[0116] The objective function Q(φ) is calculated based on the metric graph q(θ, φ) in (d). Figure 8E The function is shown. A broad peak around 0° indicates the range of gantry tilt that should be explicitly avoided. In practice, this illustrates a common problem with imaging at φ = 0°, which will suffer from significant metallic artifacts. The minimum value of Q(φ) (or, alternatively, a very low value in Q(φ) relative to the peak) identifies the ideal tilt angle (φ). * ).

[0117] Figures 9A to 9F 3D image reconstructions of phantoms with tilt angles ranging from -30° to 30° are shown. It can be seen that when φ approaches 0° (the conventional case), the phantom suffers from strong metal artifacts—dark bands appear at the screw tips and a "halo" appears at the screw edges. However, when using tilt angles greater than +15° or -15° (e.g., φ = 30°), the metal artifacts are reduced as expected, resulting in better depiction of the screw boundaries. This observation is consistent with... Figure 8D and Figure 8E The φ optimization results match.

[0118] Figures 8A to 8E The results of the physical phantom experiments using the MAA method are shown. Figure 8A Two projected positioning scan views are shown, p(u, v|θ=0, φ=0) and p(u, v|θ=90, φ=0). Figure 8B It shows that by... Figure 8A A rough 3D attenuation map (μ) is formed by unweighted back projection of the two positioning scan views. c ). Figure 8C Intensity-based segmentation (μ) of a coarse 3D attenuation map is shown. seg ). Figure 8D It shows the use of equation (1) by applying... Figure 8C of (μ) seg The metric graph q(θ, φ) is calculated by orthographic projection. Figure 8E It shows according to Figure 8D The objective function Q(φ) is calculated using the horizontal line, where the minimum value indicates the tilt angle (φ) that will avoid metal artifacts. * ).

[0119] Figures 9A to 9FAxial views of the chest phantom acquired at different tilt angles are shown. Note the artifact (light and dark stripes around the spinal screws) at φ=0°, which is avoided in images acquired at tilt angles of φ=20° or 30° as predicted by this method.

[0120] The 3D image is reconstructed using established methods. Since the orbit is circular, the 3D image can be formed using common 3D filtered backprojection methods. Alternatively, the 3D image can be reconstructed via iterative MBIR, as is common in MBIR, which improves the overall signal-to-noise ratio properties. Optionally, the 3D image can be reconstructed via various PMAC methods through processing of the projection data and / or image reconstruction to further suppress metal artifacts. Other artifact corrections (e.g., X-ray scattering / hysteresis or motion artifacts), as is typical in the prior art, can also be optionally applied. The 3D image reconstruction and processing chain remains relatively unchanged. Of course, an important consideration is that the system geometry on which the 3D image reconstruction process is based must take into account the gantry tilt; for example, the system's geometric calibration must account for the gantry tilt.

[0121] Therefore, a source-detector trajectory (i.e., a tilted circular track) to avoid metal artifacts was determined prior to the actual 3D scan. This avoids metal artifacts without requiring prior information or modifications to the reconstruction and artifact correction image processing pipeline.

[0122] The second simulation study extended the MAA method to more realistic anatomical structures and metal instruments on non-circular orbits. Figure 10A The coronal image of simulated phantom #2 is shown. Figure 10B The volume rendering of simulated model #2 is shown. Figure 10A and 10B As shown, Experiment #2 simulated the anatomy of a human chest (extracted from the Cancer Image Archive (TCIA) dataset) with three pairs of bilateral pedicle screws, labeled #1-6, implanted (at -22°, -11°, and 0° out-of-plane relative to the central axis on one side of the spine, and 0°, +11°, and +22° on the other side). To increase complexity, the attenuation of screws #1-3 was increased by 10% compared to screws #4-6. We extended track optimization to non-circular tracks. We analyzed the reduction of metal artifacts by non-circular tracks compared to the optimal inclined circular tracks, from the perspectives of visual image quality and RMSE with ground truth.

[0123] Figures 11A to 11E The results of simulation #2 are shown: MAA predictions for the optimal non-circular trajectory, and simulations of realistic anatomy and screws. Specifically, Figure 11A A graph of the q(θ, φ) metric is shown. The optimal non-circular orbit is marked in green. Figure 11B The diagram shows the Q(φ) (semi-circular) target plotted at a series of gantry tilt angles. The non-circular Q(φ(θ)) achieves a lower value than any setting of the semi-circular track. Figure 11C An axial slice (screw #3, 0° exterior angle) is shown, imaged using optimal circular (top) and non-circular (bottom) orbits (reconstructed using penalized weighted least squares (PWLS) method). Figure 11D Three example screws (#4-6, with exterior angles of 0°, 11°, and 22° respectively) are shown, obtained using optimal semi-circular (top) and non-circular (bottom) tracks. An oblique axial view across the axis of each screw is shown. Figure 11E The ground truth axial image of the simulated screw is shown.

[0124] The results of the simulation: Figures 11A to 11E A graph showing the q(θ, φ) metric calculated for a simulated spinal phantom is presented. Note the two bright areas passing through the screw axis that correspond to the low-fidelity view (strong bias caused by metal). The optimal tilted semicircular track based on Q(φ) suggests a tilt angle φ of approximately 4°, which avoids some, but not all, low-fidelity views. Figure 11C As shown in the image, it exhibits fairly strong residual metal artifacts (but is still an improvement over the non-tilted track (not shown for simplicity)). Note that Q(φ) is the target of the overall metal artifacts produced by all screws clearly visible in the projected data; therefore, while a larger tilt angle (e.g., φ = -20°) may reduce the artifacts associated with screw #3 (exterior angle: 0°), Figure 11C However, this leads to increased metal artifacts on other screws (e.g., screws #1 and #2) (with exterior angles of -22° and -11° respectively). Therefore, in this case, no single gantry tilt angle can significantly reduce artifacts on all screws.

[0125] This problem was solved by extending the MAA method to non-circular orbit optimization. The optimal non-circular orbit (by...) Figure 11A The overlaid curve (marked 1102) avoids most low-fidelity views, such as Figure 11B As shown, compared to the optimal circular track, this results in a much lower Q(φ(θ)) characteristic. The optimal non-circular MAA track significantly reduces metal artifacts for all screws—from Figure 11C The image below and Figure 11D The panel clearly shows this. Although some residual artifacts remain (especially in the #1-3 screws with higher attenuation), the MAA orbital reduced the total RMSE by approximately 46%, confirming the effectiveness of the MAA method for defining non-circular orbits.

[0126] Figure 12A method 1200 for avoiding metal artifacts in 3D X-ray imaging, as exemplified by this disclosure, is shown. Method 1200 begins at 1202 by determining the 3D position of the metal within an object or volume of interest to be scanned. At 1204, this determination may further include one or more of the following: at 1204, performing an initial 3D scan of the object or volume of interest; at 1206, acquiring one or more X-ray projection images of the object or volume of interest using one or more previously acquired X-ray images; or at 1208, locating the metal using a tracking system comprising one or more cameras or electromagnetic trackers.

[0127] Method 1200 continues by estimating a source-detector orbit at 1210 that will reduce the severity of metal artifacts. This source-detector orbit includes the position and orientation of the X-ray source and detector for acquiring projections in the 3D X-ray imaging. The severity of metal artifacts is reduced compared to a circular path in a plane perpendicular to the major axis of the object. Estimating the source-detector orbit at 1210 may further include calculating an objective function at 1212 based on the determined 3D position of the metal. The objective function describes a characteristic associated with metal artifacts in 3D image reconstruction. For example, this characteristic may include an estimate of spectral shift or attenuation.

[0128] Method 1200 continues at 1214 by moving the imaging system to a position consistent with the estimated source-detector orbit, or to a position close to the estimated source-detector orbit and within existing spatial constraints. At 1214, the movement can be based on the constraints of the imaging system (e.g., electrical or mechanical limits) and the location of the volume of interest (such that the resulting 3D image will contain the volume of interest). The estimated source-detector orbit can be adjusted as needed to align with these factors and constraints. For example, if the spatial constraints of the imaging system render the determined source-detector orbit unusable, a source-detector orbit close to the determined source-detector orbit can be used. A source-detector orbit considered close can, for example, be within a few degrees of tilt angle (φ), such as less than 1°, approximately 1° to 3°, 1° to 5°, approximately 1° to 7°, or approximately 1° to 10°, where these ranges include the limits of the range.

[0129] Method 1200 continues scanning the object at 1216 according to the source-detector orbit. The scanning at 1216 may further include acquiring multiple X-ray projections along the source-detector orbit at 1218 and forming a 3D image reconstruction of the object at 1220. Forming the 3D image reconstruction at 1220 may further include performing one or more 3D image reconstruction algorithms at 1222, including 3D filtered backprojection or model-based image reconstruction.

[0130] In summary, this MAA method accurately predicts tilted circular and / or non-circular tracks that reduce the amplitude of metal artifacts in CBCT reconstruction. The method is able to locate complex distributions of metal instruments using 2–6 low-dose localization scan projections acquired during routine pre-scan collision checks. It was found that even in complex anatomical scenes, simple U-Net segmentation is sufficient to localize (segment) metal objects in sparse views and generalizes well to a wide variety of metal objects without the need for manufacturer-specific device models. Simulations show that MAA tracks reduce RMSE in 3D image reconstruction by approximately 46%–70% and reduce “halo” artifacts (FWHM of screws) by approximately 20%–45%. The non-circular tracks defined by MAA reduce RMSE by approximately 46% compared to the optimal circular tracks. The results demonstrate that the MAA method accurately predicts tilted circular / non-circular tracks that minimize metal artifacts in all experiments.

[0131] Additionally and / or alternatively to the above description, the capability for 3D imaging via a non-circular track on a mobile C-arm with motorized tilting and gantry angles is further described. Specifically, the geometric calibration problem for general (non-predefined) non-circular tracks has been addressed, and the 3D imaging performance characteristics of non-circular tracks compared to conventional circular tracks have been evaluated. The ability to extend the MAA method to non-circular tracks is utilized to minimize metal-induced bias. Results show that the method operates well using only two low-dose localization scan views (without requiring additional prior information on patient anatomy / metallic objects) and is compatible with MAR and multipotential reconstruction methods, which can further improve image quality.

[0132] exist Figure 13A The movable C-arm used in this disclosure is shown (Cios Spin 3D from Siemens Healthineers, Fockeheim, Germany). The C-arm has electrically controllable rotation angles (θ: 0° to 196°) and gantry tilt angles (φ: -30° to 30°, due to practical considerations regarding patient / table collisions), allowing for the execution of non-circular tracks during scanning by variations in θ and φ. Established methods can be used to calibrate tilted circular tracks (constant φ). However, for non-circular scan trajectories defined by methods such as “task-driven” imaging and MAA, the track is dynamically designed in real time, and given the various feasible combinations of (θ, φ), the track may not fit the previously obtained calibration. As detailed below, an interpolation-based approach is used to address this issue.

[0133] Alternatively, instead of using the system geometry determined by conventional geometric calibration methods or interpolation geometric calibration methods to solve for image reconstruction, the image reconstruction can be solved by the system geometry determined by the "self-calibration of cone-beam CT geometry using 3D-2D image registration," a paper published by S Ouadah, JW Stayman, GJ Gang, T Ehtiati, and J Hsiewerdsen on March 10, 2016. 2017, Institute of Physics and Engineering in Medicine, Physics in Medicine & Biology, Volume 61, No. 7 and Hopkins Intellectual Property Case C13072, Patent Publication No.: US 2017 / 0238897 A1.

[0134] Figure 13A and Figure 13B Examples of a movable C-arm, C-arm geometry, and a magnified view of the detector are shown, respectively, according to this disclosure. Figure 13A Similar to Figure 1A A movable C-arm. Nine parameters (degrees of freedom) determined by geometric calibration are marked at point 1302. Figure 13B The source-detector track is illustrated on a sphere. Curve 1304 shows a circular track (various settings of gantry tilt φ). Curve 1306 shows an example non-circular track (φ varies with θ). The zoomed-in inset illustrates the geometric parameter estimates of a view (point 1308) determined by interpolation of the four nearest vertices (point 1302) in a geometric calibration library of circular scans.

[0135] First, using a cylindrical BB phantom placed at the same tilt angle as the C-arm gantry, a geometric calibration library is established for a circular track with a tilt range varying from φ = -30° to +30° in 5° intervals. The affine transformation of the library registers all calibrations to a common coordinate system (common registration):

[0136] P φ＝0 =P φ =φ * ×T 倾斜 ×T w (11)

[0137]

[0138]

[0139] Where Pφ = φ* is the projection matrix at the tilt angle φ*, and T_tilt is the transformation of the tilt angle φ* (read from the C-arm motor encoder). Transformation Tw explains the minute differences in displacement (t) and tilt (rz) between the BB phantom and the gantry for each member in the library, which are determined by 3D-3D registration. This common registration step: (i) ensures that the orientation of the reconstructed image remains unchanged regardless of the track; (ii) improves the accuracy of the following interpolation steps.

[0140] The co-registration calibration is then decomposed into nine parameters: SDD (source detector distance), (position in x, y, z), u0, v0 (puncture point), and (detector rotation matrix in x, y, z), as follows: Figure 13A Point 1302 is shown in the diagram. Then, the geometric parameters from the calibration library—that is, as shown in the diagram—can be adjusted. Figure 13B As shown, four recent views from a calibrated inclined circular track are used to interpolate and estimate the system geometry of a general non-circular track. Parameters SDD, u0, and v0 are estimated by dispersed linear interpolation, and C and R are estimated by spherical linear interpolation (slerp). The slerp operation is non-interchangeable; because the θ direction is sampled more finely in the calibration library, the slerp operation is performed first on θ and then on φ to reduce interpolation error.

[0141] Given two localization scan views, the MAA method will determine a binary 3D reconstruction (coarse segmentation, μ-segmentation) describing the 3D position of a metallic object within the FOV. seg Unlike in MAR, it was found that coarse segmentation is sufficient for MAA, requiring only approximations of the shape, size, and orientation of the metallic object (see MAR method, which typically requires very precise segmentation to avoid introducing secondary artifacts).

[0142] The coarse segmentation uses an end-to-end convolutional neural network (CNN) as discussed above, which can learn features from both the projection domain and the image domain simultaneously. Specifically, the projection domain feature extraction layer is connected to the image domain 3D segmentation layer via a frozen (parameterless) backprojection layer. The intuition behind this network is that image domain segmentation is improved by fusing shape information from the projection domain (before it is erased by backprojection), while projection domain feature extraction is guided by a segmentation loss backpropagated from the image domain.

[0143] Figure 14 An example of an end-to-end neural network for 3D localization / segmentation of metal instruments, according to this disclosure, is shown. The projection domain feature extraction layer is connected to the image domain segmentation layer via a frozen backprojection layer (determined by C-arm geometry), enabling joint learning in both the projection and image domains.

[0144] Each projection domain feature extraction layer contains two 3×3 convolutions, each followed by a Corrected Linear Unit (ReLU) with batch normalization, and finally a 2×2 max pooling layer. The 3D segmentation layer is implemented as a three-layer multichannel 3D U-Net with the general Dice coefficient as the loss function. Figure 14 (Details are provided below). At the beginning of the 3D segmentation layer, the derivative of the loss function with respect to the input of the backprojection layer is the orthographic operation (determined by C-arm geometry and Siddon ray tracing, not as a learnable parameter of the network), allowing for joint learning in both the projection and image domains. The network is initialized and trained using the Adam optimizer with a "Henormal" initialization and training rate of 5 × 10⁻⁴ for 100 generations.

[0145] One of the key hyperparameters of the network is the number of feature maps extracted in the projection domain. Figure 14 The value is denoted as Nc, which is also the number of input channels used in a multi-channel 3D U-Net in the image domain. A backprojection step is performed separately for each channel (within the backprojection layer).

[0146] For simplicity and to avoid using vendor-specific metal instrument models, the network was trained using only simulated data generated from 10 abdominal and chest CT images extracted from the Cancer Imaging Archive (TCIA) and a random number of classful metal objects: ellipses of varying sizes and eccentricities (major and minor axes of 10–80 mm) and coarsely segmented spinal screws extracted from the CT images. Future work will certainly involve adding higher-fidelity object models (e.g., vendor-specific designs) and other application-specific objects (e.g., fracture fixation plates) to improve performance. The DRR generation incorporates the effects of data truncation, beam hardening (via multi-energy orthographic projection), scattering, hysteresis, glare, and noise (quantum and electronic). Data augmentation includes variations in the position, size, orientation, aspect ratio, and attenuation of the simulated metal objects in each DRR, resulting in approximately 8,400 images in total (8,000 training images + 400 validation images).

[0147] The results of the experimental study are presented below. First, the geometric calibration of non-circular tracks is discussed. The reproducibility of the geometric parameters of non-circular tracks (φ linearly increases from -20° to +20°, while θ linearly increases from 0° to 196°) was evaluated by performing four geometric calibrations (non-interpolation) during an 8-hour normal usage period. Three cases were evaluated: (i) a conventional pre-calibrated circular track (denoted as "calibrated circular"); (ii) a pre-determined and pre-calibrated non-circular track (denoted as "calibrated non-circular"); and (iii) a general non-circular track with a projection matrix determined by the above-described interpolation-based method (denoted as "interpolated non-circular"). In a head phantom containing various custom inserts, the basic image quality characteristics of the scans in these three cases were evaluated in terms of spatial resolution (modulation transfer function MTF) and 3D sampling characteristics (cone beam artifacts). The axial plane MTF was determined by the edge spread function measured by a high-contrast (approximately 300 HU) cylindrical rod insert. The amplitude of the cone-beam artifact was measured based on the full width at half maximum (FWHM) of the upper and lower edges of the flattened disc inserted into the skull (“Defrise phantom”). All scans consisted of 400 projections at an arc of θ of 196°, 110 kV, 1.65 mAs / view, and a scan time of 30 s.

[0148] Next, we discuss metal artifact avoidance (MAA) using non-circular orbits. The performance of the CNN-based metal object localization step is evaluated in terms of Dice coefficients on the validation and test datasets [scans of cadavers with six pairs of pedicle screws (Nuvasive, San Diego, USA)]. The performance of the proposed end-to-end approach is compared with two other approaches: (i) conventional image domain U-Net segmentation on direct backprojection of the localization scan view (referred to as “single U-Net”); [5] and (ii) projection domain U-Net segmentation of the metal object, followed by backprojection and image domain U-Net segmentation (referred to as “dual U-Net”). Note that the two U-Nets in (ii) are trained separately (non-end-to-end).

[0149] The performance of the MAA method was evaluated in a phantom study involving an anthropomorphic chest phantom containing natural human skeleton in tissue-equivalent plastic and eight pedicle screws (Renahm DePuy-Synthes, USA; lengths varying between 30–60 mm). The screw plane exterior angles ranged from -20° to +30° (positive on one side of the spine, negative on the other). For both standard circular orbit scans and the optimal non-circular orbit defined by MAA, the amplitude of metallic artifacts was evaluated in terms of the “halo” (the screw’s FWHM minus its true width) about the screw axis.

[0150] The results of the geometric calibration study on non-circular orbits will now be discussed. Figure 15A and Figure 15B The reproducibility of the geometric parameters of the aforementioned predetermined non-circular orbit is shown, wherein, Figure 15A The location of the puncture point (u0, v0) is shown, and Figure 15B The SDD (Self-Depth Determination) is shown in four repeated trials within a standard 8-hour use interval. Reduced reproducibility was observed: for example, the standard deviation of the puncture site location (u0, v0) was approximately 1.7 mm in the repeated trials (compared to approximately 0.8 mm for a conventional circular track). Note that the mobile C-arm is not electromechanically adjustable outside its standard clinical deployment and therefore does not support non-circular tracks in standard use. However, as shown below, despite the reduced mechanical reproducibility, the interpolation-based calibration still provides a reasonable estimate of the system geometry that supports 3D image reconstruction.

[0151] Figures 16A to 16D The figure illustrates the relationship between geometric calibration error and spatial resolution reduction in the three cases described above. Figure 16A It shows the method for using by Figure 13A and Figure 13B The MTF for the three cases defined in the layout, and Figure 16B , Figure 16C and Figure 16D Axial and sagittal magnified views (skull base) of the head phantom are shown for three different cases. Compared to the standard “calibrated circular” case, the “calibrated non-circular” case shows a slight decrease in MTF (approximately 3.5% reduction in spatial frequency, denoted as f10, at MTF = 0.10) due to reduced reproducibility of the calibration parameters. The “interpolated non-circular” case shows a further reduction in resolution (approximately 12% reduction in f10), attributed to two factors: smoothing of geometric parameters in the interpolation; and neglecting the gantry momentum differences of continuous non-circular tracks compared to a discrete circular scan library. Despite the reduced MTF, the image obtained in the “interpolated non-circular” case visually appears comparable to... Figure 16B and Figure 16C The other images in the dataset are comparable. Therefore, the interpolation-based approach appears to provide a reasonable estimate of the system geometry for non-circular scans that have neither pre-calibrated nor “self-calibrated” (using previous scans and 3D-2D registration).

[0152] Figures 17A to 17C It shows that in the form of Figure 13A and Figure 13B The expected reduction of cone-beam artifacts generated by non-circular orbits in the sagittal image of a head phantom comprising a stacked disk, as defined by the arrangement, wherein, Figure 17A Corresponding to the calibration circular track, Figure 17B Corresponding calibration for non-circular tracks, and Figure 17CCorresponding to interpolated non-circular tracks. The apparent thickness (FWHM) of the top disk is reduced from 9 mm in the standard "circular case" to the actual thickness (approximately 5.5 mm) in its two non-circular cases.

[0153] The results of the study on non-circular orbital MAA will now be discussed. Figure 18 The performance of segmenting real or simulated metal implants using the Dice coefficient (denoted as DC) in validation tests for three network types (single U-Net, dual U-Net, and end-to-end) as a function of the number of projected views is illustrated, showing the improved performance of the end-to-end network compared to single U-Net and dual U-Net. The increase in DC (especially for fewer localization scan views) confirms the effectiveness of end-to-end training: for only two localization scan views, the end-to-end method increases Dice by approximately 29% on the validation dataset and approximately 42% on the test dataset compared to dual U-Net. This is consistent with other literature utilizing end-to-end training.

[0154] Figures 19A to 19E The segmentation performance of three neural networks is shown in the test dataset (cadavers with 12 pedicle screws). Figures 19A to 19C An example axial slice segmentation overlaid with ground truth is shown. Segmentation for different numbers of positioning scan views is also shown. Figure 19D The end-to-end method is shown to calculate μ based on two views. seg The isosurface. Figure 19E The isosurface of the ground truth segmentation is shown (which is downsampled to μ). seg ).

[0155] Figures 20A to 20F The q(θ,φ) metric plot is shown, calculated using only two localized scan views of the chest phantom and spinal screws via the MAA method. Clearly, there is no circular track [the horizontal line in (a)] that significantly reduces metal artifacts on all screws, because there is always at least one region of strong deviation caused by metal in the (θ,φ) track space (roughly corresponding to the exterior angle of the screw). This problem is solved by 2D minimization of equations (6) and (7), resulting in the non-circular track (a) marked with the cyan curve. The non-circular track avoids most low-fidelity views, guiding a path in (θ,φ) that minimizes changes in spectral shift. The resulting image quality is... Figures 20C to 20F As shown, the halo artifacts around the screw are greatly reduced (optimal non-circular track achieves an improvement of approximately 70% in screw shaft halo).

[0156] In summary, a method for geometric calibration of non-circular C-arm CBCT tracks is described, providing a practical means of estimating system geometry from a discrete circular scan calibration library. Although the geometric calibration method significantly reduces MTF, its impact on visual image quality is relatively small. Results show that, as expected, CBCT images acquired from non-circular tracks improve 3D sampling characteristics (e.g., reduce cone-beam artifacts). This capability enables a method (called MAA) to identify non-circular tracks with minimal metal-induced bias. An end-to-end neural network is described that can locate metallic objects from only two positioning scan views without requiring sufficient prior information about patient anatomy or metal instruments. Integrating the end-to-end network with the MAA method for non-circular tracks demonstrates a significant reduction in metal artifacts in phantom and cadaver studies. Furthermore, the method is compatible with established MAR and multi-energy reconstruction algorithms for further artifact reduction.

[0157] Figure 23 This is an example of a hardware configuration for computer device 2300, which can be used to perform one or more of the processes described above, and / or implement the CT system 102 and / or control system 118 described above. Computer device 2300 can be any type of computer device, such as a desktop computer, laptop computer, server, etc., or mobile device, such as a smartphone, tablet computer, cellular phone, personal digital assistant, etc. Figure 23 As shown, computer device 2300 may include one or more processors 2302 with different core configurations and clock frequencies. Computer device 2300 may also include one or more memory devices 2304 that serve as main memory during operation of computer device 2300. For example, during operation, a copy of the software supporting the above operations may be stored in one or more memory devices 2304. Computer device 2300 may also include one or more peripheral interfaces 2306, such as a keyboard, mouse, touchpad, computer screen, touchscreen, etc., for enabling human-computer interaction and manipulation of computer device 2300.

[0158] Computer device 2300 may also include one or more network interfaces 2308 for communicating via one or more networks, such as Ethernet adapters, wireless transceivers, or serial network components, to communicate using protocols over wired or wireless media. Computer device 2300 may also include one or more storage devices 2310 with different physical dimensions and storage capacities, such as flash drives, hard disk drives, random access memory, etc., for storing data, such as images, files, and program instructions executed by one or more processors 2302.

[0159] Additionally, computer device 2300 may include one or more software programs 2312 that implement the above-described functions. The one or more software programs 2312 may include programs that cause one or more processors 2302 to execute, for example, the functions described herein. Figure 2 , Figure 12 , Figure 14 , Figure 21 and / or Figure 22 The process describes the procedures, functions, and instructions for operation. Copies of one or more software programs 2312 may be stored in one or more memory devices 2304 and / or one or more storage devices 2310. Similarly, data used by one or more software programs 2312 may be stored in one or more memory devices 2304 and / or one or more storage devices 2310. A data bus 2316 provides a data communication path between the various components of the computer device 2300.

[0160] In this implementation, computer device 2300 can communicate with other devices via network 2314. These other devices can be any type of device as described above. Network 2314 can be any type of network, such as a local area network (LAN), wide area network (WAN), virtual private network (VPN), the Internet, an intranet, an extranet, a public switched telephone network (PSTN), an infrared network, a wireless network, or any combination thereof. Network 2314 can support communication using any of a variety of commercially available protocols, such as TCP / IP, UDP, OSI, FTP, UPnP, NFS, CIFS, AppleTalk, etc. Network 2314 can be, for example, a LAN, WAN, VPN, the Internet, an intranet, an extranet, a PSTN, an infrared network, a wireless network, or any combination thereof.

[0161] Computer device 2300 may include various data storage devices as discussed above, as well as other memories and storage media. These storage devices may reside in various locations, such as on storage media located locally on one or more computers (and / or residing therein) or on storage media remotely from any or all computers in a network. In some embodiments, information may reside in a storage area network (“SAN”) familiar to those skilled in the art. Similarly, any necessary files that will be used to perform functions belonging to a computer, server, or other network device may be stored locally and / or remotely as needed.

[0162] In implementations, the components of the computer device 2300 described above need not be enclosed in a single housing, nor even adjacent to each other. Those skilled in the art will understand that the components described above are merely examples, as the computer device 2300 may include any type of hardware components for performing the disclosed embodiments, including any necessary accompanying firmware or software. The computer device 2300 may also be implemented, in part or in whole, by electronic circuit components or processors such as application-specific integrated circuits (ASICs) or field-programmable gate arrays (FPGAs).

[0163] If implemented in software, functionality can be stored or transmitted as one or more instructions or code on or through a computer-readable medium. Computer-readable media include tangible non-transitory computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one place to another. Storage media can be any available tangible non-transitory medium that is accessible to a computer. By way of example, and not limitation, such tangible non-transitory computer-readable media can include RAM, ROM, flash memory, EEPROM, CD-ROM or other optical disc storage devices, disk storage devices or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that is accessible to a computer. As used herein, disks and optical discs include CDs, laser discs, optical discs, DVDs, floppy disks, and Blu-ray discs, where disks typically copy data magnetically, while optical discs copy data optically using lasers. Furthermore, any connection can be appropriately referred to as a computer-readable medium. For example, if software is transmitted from a website, server, or other remote source using coaxial cable, optical fiber, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the definition of media includes coaxial cable, optical fiber, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave. Combinations of the above should also be included within the scope of computer-readable media.

[0164] The foregoing description is illustrative, and those skilled in the art can make changes to the configuration and implementation. For example, the various illustrative logics, logic blocks, modules, and circuits described in connection with the embodiments disclosed herein can be implemented or performed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a cryptographic coprocessor, or other programmable logic devices designed to perform the functions described herein, discrete gate or transistor logic, discrete hardware components, or any combination thereof. A general-purpose processor may be a microprocessor, but alternatively, it may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other such configuration.

[0165] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. For software implementations, the techniques described herein may be implemented in conjunction with modules (e.g., processes, functions, subroutines, programs, routines, subroutines, modules, software packages, classes, etc.) that perform the functions described herein. Modules may be coupled to another module or hardware circuit by passing and / or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc., may be passed, forwarded, or transmitted using any suitable means, including memory sharing, messaging, token passing, network transmission, etc. Software code may be stored in memory cells and executed by a processor. Memory cells may be implemented inside or outside the processor; in the case of external implementations, memory cells may be communicatively coupled to the processor via various means known in the art.

[0166] Compared to existing MAR solutions, the MAA method offers several advantages. First, it can acquire as few as two localization scan views in addition to the nominal 3D scan. Since localization scan views are typically acquired as standard practice to confirm patient placement, they do not expose the patient to additional radiation. Second, prior information about the shape of the metal insert or the patient's pre-instrument 3D scan is not required. Third, the underlying algorithm for calculating the optimal trajectory (tilt angle) is independent of existing reconstruction and artifact correction pipelines and can be deployed as a separate module, making it easier to integrate into current CBCT systems. Finally, as a hardware-based solution, it is more robust than its software-based counterpart and eliminates the risk of introducing unrealistic information (through interpolation, etc.) into the reconstructed volume.

[0167] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. For software implementations, the techniques described herein may be implemented in conjunction with modules (e.g., processes, functions, subroutines, programs, routines, subroutines, modules, software packages, classes, etc.) that perform the functions described herein. Modules may be coupled to another module or hardware circuit by passing and / or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc., may be passed, forwarded, or transmitted using any suitable means, including memory sharing, messaging, token passing, network transmission, etc. Software code may be stored in memory cells and executed by a processor. Memory cells may be implemented inside or outside the processor; in the case of external implementations, memory cells may be communicatively coupled to the processor via various means known in the art.

[0168] In addition, this disclosure includes embodiments pursuant to the following terms:

[0169] Clause 1. A method for avoiding metal artifacts in 3D X-ray imaging, the method comprising:

[0170] Determine the 3D position of the metal within the object or volume of interest to be scanned;

[0171] The source-detector orbit is estimated to reduce the severity of metal artifacts;

[0172] Move the imaging system to a location aligned with the estimated source-detector orbit, or to a location close to the estimated source-detector orbit but within existing spatial constraints; and

[0173] The object is scanned based on the source-detector orbit.

[0174] Clause 2. The method as described in Clause 1, wherein the determination further comprises one or more of the following: performing an initial 3D scan of the object or volume of interest, acquiring one or more X-ray projection images of the object or volume of interest, or locating the metal using a tracking system comprising one or more cameras or electromagnetic trackers.

[0175] Clause 3. The method as described in Clause 1 or 2, wherein the source-detector orbit includes the position and orientation of the x-ray source and detector for acquiring a projection in the 3D x-ray imaging.

[0176] Clause 4. The method as described in Clauses 1 to 3, wherein estimating the source-detector orbit further comprises calculating an objective function based on the determined 3D position of the metal, wherein the objective function describes characteristics associated with metal artifacts in the 3D image reconstruction.

[0177] Clause 5. The method as described in Clauses 1 to 4, wherein the characteristic includes an estimate of spectral shift, attenuation, or a combination thereof.

[0178] Clause 6. The method as described in Clauses 1 to 5, wherein the objective function is based on one or more of the following: the range of the metric plot along the gantry rotation axis, the standard deviation of the metric plot along the gantry rotation axis, the maximum value of the metric plot along the gantry rotation axis, and the sum of the metric plots within the rotation angle range of the gantry.

[0179] Clause 7. The method as described in Clauses 1 to 6, wherein the severity of metal artifacts is reduced compared to a circular path in a plane perpendicular to the major axis of the object.

[0180] Clause 8. The method as described in Clauses 1 to 7, wherein the scan further comprises: acquiring multiple X-ray projections along the source-detector orbit and forming a 3D image reconstruction of the object.

[0181] Clause 9. The method as described in Clauses 1 to 8, wherein forming the 3D image reconstruction further comprises: performing one or more 3D image reconstruction algorithms, including 3D filtered backprojection, model-based image reconstruction, deep learning, or neural network reconstruction.

[0182] In addition, this disclosure includes embodiments pursuant to the following terms:

[0183] Clause 1. A 3D x-ray imaging system, comprising:

[0184] A 3D x-ray imaging apparatus comprising a gantry capable of moving at multiple tilt angles along a tilt axis and multiple rotation angles along a rotation axis; and

[0185] A hardware processor configured to execute instructions including the following:

[0186] Determine the 3D position of the metal within the object or volume of interest to be scanned;

[0187] The source-detector orbit is estimated to reduce the severity of metal artifacts;

[0188] Move the imaging system to a location aligned with the estimated source-detector orbit, or to a location close to the estimated source-detector orbit but within existing spatial constraints; and

[0189] The object is scanned based on the source-detector orbit.

[0190] Clause 2. The 3D x-ray imaging system as described in Clause 1, wherein the determination further includes one or more of the following: performing an initial 3D scan of the object or volume of interest, acquiring one or more x-ray projection images of the object or volume of interest, or locating the metal using a tracking system comprising one or more cameras or electromagnetic trackers.

[0191] Clause 3. A 3D x-ray imaging system as described in Clause 1 or 2, wherein the source-detector orbit includes the position and orientation of the x-ray source and detector for acquiring a projection in the 3D x-ray imaging.

[0192] Clause 4. The 3D X-ray imaging system as described in Clauses 1 to 3, wherein estimating the source-detector orbit further comprises calculating an objective function based on the determined 3D position of the metal, wherein the objective function describes a characteristic associated with metal artifacts in the 3D image reconstruction.

[0193] Clause 5. A 3D x-ray imaging system as described in Clauses 1 to 4, wherein the characteristic includes an estimate of spectral shift, attenuation, or a combination thereof.

[0194] Clause 6. The 3D X-ray imaging system as described in Clauses 1 to 5, wherein the objective function is based on one or more of the following: the range of the metric plot along the gantry rotation axis, the standard deviation of the metric plot along the gantry rotation axis, the maximum value of the metric plot along the gantry rotation axis, and the sum of the metric plots within the rotation angle range of the gantry.

[0195] Clause 7. The 3D X-ray imaging system as described in Clauses 1 to 6, wherein the severity of metal artifacts is reduced compared to a circular path in a plane perpendicular to the long axis of the object.

[0196] Clause 8. The 3D x-ray imaging system as described in Clauses 1 to 7, wherein the scan further comprises: acquiring multiple x-ray projections along the source-detector orbit and forming a 3D image reconstruction of the object.

[0197] Clause 9. The 3D x-ray imaging system as described in Clauses 1 to 8, wherein forming the 3D image reconstruction further comprises: performing one or more 3D image reconstruction algorithms, including 3D filtered backprojection, model-based image reconstruction, deep learning, or neural network reconstruction.

[0198] Although the numerical ranges and parameters described in this teaching are approximate, the numerical values ​​presented in the specific examples are reported as precisely as possible. However, any numerical value inherently contains some error, which must be caused by the standard deviation of their respective test measurements. Furthermore, all ranges disclosed herein should be understood to encompass any and all subranges contained therein. For example, the range “less than 10” can include any and all subranges (including endpoints) between a minimum value of 0 and a maximum value of 10; that is, any and all subranges where the minimum value is equal to or greater than zero and the maximum value is equal to or less than 10, such as 1 to 5. In some cases, the numerical values ​​of the parameters can be negative. In such cases, example values ​​for the range stated as “less than 10” can be negative, such as -1, -2, -3, -10, -20, -30, etc.

[0199] Although this teaching has been described with respect to one or more embodiments, changes and / or modifications may be made to the illustrated examples without departing from the spirit and scope of the appended claims. For example, it will be appreciated that although the process is described as a series of actions or events, this teaching is not limited to the order of these actions or events. Some actions may occur in a different order and / or simultaneously with other actions or events other than those described herein. Moreover, not all process stages may be required to implement the method according to one or more aspects or embodiments of this teaching. It will be understood that structural components and / or processing stages may be added, or existing structural components and / or processing stages may be removed or modified. Furthermore, one or more actions described herein may be performed in one or more separate actions and / or stages. In addition, with regard to the extent to which the terms “including,” “includes,” “having,” “has,” “with,” or variations thereof are used in the Detailed Description and Claims, these terms are intended to be inclusive rather than inclusive, in a manner similar to the term “comprising.” The term “at least one of” is used to indicate that one or more of the listed items may be selected. As used herein, the term "one or more of," for example, A and B, refers to A alone, B alone, or A and B. Furthermore, in the discussion and claims herein, the term "on," when referring to two materials, means that there is at least some contact between the materials, while "over" means that the materials are close together but may have one or more other intermediate materials, allowing for contact, but not necessarily requiring it. Neither "on" nor "over" as used herein implies any directionality. The term "about" indicates that the listed values ​​may be varied, as long as such variation does not result in a process or structure inconsistent with the illustrated implementation. Finally, "exemplary" indicates that the description is used as an example and not implying that it is ideal. Other embodiments of this teaching will be apparent to those skilled in the art upon consideration of the specification and practice of the disclosure herein. The specification and examples are intended to be considered merely exemplary, and the true scope and spirit of this teaching are indicated by the appended claims.

Claims

1. A method for avoiding metal artifacts in 3D X-ray imaging, the method comprising: Determine the 3D position of the metal within the object or volume of interest to be scanned; The source-detector orbit is estimated to reduce the severity of metal artifacts, wherein the source-detector orbit is defined by the tilt and rotation angles of the gantry that vary during the scanning operation, and wherein estimating the source-detector orbit further includes calculating an objective function based on the determined 3D position of the metal, wherein the objective function describes the characteristics associated with metal artifacts in 3D image reconstruction. Move the imaging system to a location aligned with the estimated source-detector orbit, or to a location close to the estimated source-detector orbit but within existing spatial constraints; and The object is scanned based on the source-detector orbit.

2. The method as described in claim 1, wherein, The determination further includes one or more of the following: performing an initial 3D scan of the object or volume of interest, acquiring one or more X-ray projection images of the object or volume of interest, or locating the metal using a tracking system including one or more cameras or electromagnetic trackers.

3. The method as described in claim 1, wherein, The source-detector orbit includes the position and orientation of the X-ray source and detector for obtaining projections in this 3D X-ray imaging.

4. The method of claim 1, wherein, This feature includes estimations of spectral shift, attenuation, or a combination thereof.

5. The method of claim 1, wherein, The objective function is based on one or more of the following: the range of the metric plot along the gantry rotation axis, the standard deviation of the metric plot along the gantry rotation axis, the maximum value of the metric plot along the gantry rotation axis, and the sum of the metric plots within the rotation angle range of the gantry.

6. The method of claim 1, wherein, The severity of metal artifacts is reduced compared to a circular path in a plane perpendicular to the object's major axis.

7. The method of claim 1, wherein, The scan further includes acquiring multiple X-ray projections along the source-detector orbit and forming a 3D image reconstruction of the object.

8. The method of claim 7, wherein, The formation of the 3D image reconstruction further includes: performing one or more 3D image reconstruction algorithms, including 3D filtered backprojection or model-based image reconstruction.

9. The method of claim 8, wherein, The model-based image reconstruction is based on deep learning.

10. The method of claim 8, wherein, The model-based image reconstruction is a neural network reconstruction.

11. A 3D x-ray imaging system, comprising: A 3D x-ray imaging device, comprising a gantry capable of moving at multiple tilt angles along a tilt axis and multiple rotation angles along a rotation axis; as well as A hardware processor configured to execute instructions including the following: Determine the 3D position of the metal within the object or volume of interest to be scanned; The source-detector orbit is estimated to reduce the severity of metal artifacts, wherein the source-detector orbit is defined by the tilt and rotation angles of the gantry that vary during the scanning operation, and wherein estimating the source-detector orbit further includes calculating an objective function based on the determined 3D position of the metal, wherein the objective function describes the characteristics associated with metal artifacts in 3D image reconstruction. Move the imaging system to a location aligned with the estimated source-detector orbit, or to a location close to the estimated source-detector orbit but within existing spatial constraints; and The object is scanned based on the source-detector orbit.

12. The 3D x-ray imaging system of claim 11, wherein, The determination further includes one or more of the following: performing an initial 3D scan of the object or volume of interest, acquiring one or more X-ray projection images of the object or volume of interest, or locating the metal using a tracking system including one or more cameras or electromagnetic trackers.

13. The 3D x-ray imaging system as claimed in claim 11, wherein, The source-detector orbit includes the position and orientation of the X-ray source and detector for obtaining projections in this 3D X-ray imaging.

14. The 3D X-ray imaging system as claimed in claim 11, wherein, This feature includes estimations of spectral shift, attenuation, or a combination thereof.

15. The 3D x-ray imaging system of claim 11, wherein, The objective function is based on one or more of the following: the range of the metric plot along the gantry rotation axis, the standard deviation of the metric plot along the gantry rotation axis, the maximum value of the metric plot along the gantry rotation axis, and the sum of the metric plots within the rotation angle range of the gantry.

16. The 3D x-ray imaging system of claim 11, wherein, The severity of metal artifacts is reduced compared to a circular path in a plane perpendicular to the object's major axis.

17. The 3D x-ray imaging system of claim 11, wherein, The scan further includes acquiring multiple X-ray projections along the source-detector orbit and forming a 3D image reconstruction of the object.

18. The 3D x-ray imaging system of claim 17, wherein, The formation of the 3D image reconstruction further includes: performing one or more 3D image reconstruction algorithms, including 3D filtered backprojection or model-based image reconstruction.

19. The 3D x-ray imaging system as claimed in claim 18, wherein, The model-based image reconstruction is based on deep learning.

20. The 3D x-ray imaging system of claim 18, wherein, The model-based image reconstruction is a neural network reconstruction.

21. A non-transitory computer-readable medium comprising instructions configured, when executed by a hardware processor, to perform a method for avoiding metal artifacts in 3D x-ray imaging, the method comprising: Determine the 3D position of the metal within the object or volume of interest to be scanned; The source-detector orbit is estimated to reduce the severity of metal artifacts, wherein the source-detector orbit is defined by the tilt and rotation angles of the gantry that vary during the scanning operation, and wherein estimating the source-detector orbit further includes calculating an objective function based on the determined 3D position of the metal, wherein the objective function describes the characteristics associated with metal artifacts in 3D image reconstruction. Move the imaging system to a location aligned with the estimated source-detector orbit, or to a location close to the estimated source-detector orbit but within existing spatial constraints; and The object is scanned based on the source-detector orbit.

Images