Anchored kernel scattering estimation

Abstract

A radiation imaging apparatus is provided that includes a radiation source for emitting radiation, a radiation detector arranged to receive radiation emitted by the radiation source and generate radiation data, the radiation data including a primary component and a secondary component, and a data processing system configured to apply an image transform to the primary component using a generating function, construct a scattering model basis using the transform, adjust parameters in the scattering model to fit the scattering using the scattering model basis, generate an estimated scattering image using the fitted scattering model, and modify the radiation data using the scattering image to reduce scattering in the radiation data, thereby generating a scatter corrected image.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application is related to U.S. patent application Ser. No. 16 / 694,145, filed Nov. 25, 2019, entitled "MULTIMODAL RADIATION APPARATUS AND METHODS," and U.S. patent application Ser. No. 16 / 694,148, filed Nov. 25, 2019, entitled "APPARATUS AND METHODS FOR SCALABLE FIELD OF VIEW IMAGING USING A MULTI-SOURCE SYSTEM," both of which are incorporated herein by reference in their entireties.

[0002] Aspects of the disclosed technology relate to improving quality during radiological image processing, including, for example, reducing noise and artifacts associated with scatter and scatter correction, and more specifically, processing scatter corrected images as non-scatter corrected and scatter only components. [Background technology]

[0003] Tomography is a noninvasive radiological imaging technique used to generate cross-sectional images of three-dimensional (3D) objects without overlaying tissue. Tomography can be classified as transmission tomography, such as computed tomography (CT), and emission tomography, such as single-photon emission computed tomography (SPECT) and positron emission tomography (PET). CT is a technique that creates cross-sectional views of the body based on X-rays passing through the patient. Photon emission computed tomography and positron emission tomography provide 3D imaging information about radionuclides injected into the patient, showing metabolic and physiological activity within organs.

[0004] In a tomographic scan, projections are acquired from many different angles around the body by one or more rotating detectors (as with a rotating radiation source in CT). These data are then reconstructed to form a 3D image of the body. For example, reconstruction of a tomographic image can be achieved by filtered backprojection and iterative methods.

[0005] The quality of the final image is limited by several factors. Examples include attenuation and scattered photons, detection efficiency, and the spatial resolution of the collimator-detector system. These factors can cause low spatial resolution, low contrast, and / or high noise levels. Image data processing (e.g., filtering) techniques can be used to improve the quality of the image.

[0006] In CT, including cone-beam CT, the primary signal detected by a detector element represents the x-rays that exit the tube, penetrate the patient's body, and reach the detector. The x-rays in the primary signal travel along the x-ray path connecting the tube focal point to the detector element. The scatter signal detected by the same element also represents x-rays scattered to the element from other x-ray paths. The primary signal allows for the reconstruction of the CT image. However, the scatter signal can degrade the CT image both quantitatively and qualitatively.

[0007] Scatter in various radiological imaging modalities, including CT and cone-beam CT, can account for a significant portion of detected photons. Scatter can adversely affect image quality, including contrast, uniformity, and quantitative accuracy. Therefore, scatter measurement, estimation, and correction are applicable to data processing and image reconstruction, including those associated with image-guided radiation therapy (IGRT). IGRT can utilize medical imaging techniques, such as CT, to collect images of patients before, during, and / or after treatment. Summary of the Invention

[0008] In one variant, there is provided a radiation imaging apparatus including a radiation source for emitting radiation, a radiation detector arranged to receive the radiation emitted by the radiation source and generate radiation data, the radiation data including a primary component and a secondary component, and a data processing system configured to apply an image transform to the primary component using a generating function, construct a scattering model basis using the transform, adjust parameters in the scattering model to fit the scatter using the scattering model basis, generate an estimated scatter image using the fitted scattering model, and modify the radiation data using the scatter image to reduce scatter in the radiation data, thereby generating a scatter-corrected image.

[0009] The radiation source may comprise a rotating X-ray source emitting a radiation beam. The radiation detector may comprise an X-ray detector positioned to receive radiation from the X-ray source. The apparatus may further comprise a beamformer configured to adjust the shape of the radiation beam emitted by the X-ray source such that a primary region of the X-ray detector is directly exposed to the radiation beam and at least one shadow region of the X-ray detector is blocked from direct exposure to the radiation beam by the beamformer.

[0010] Estimating the scatter component of the radiation data may be based on measured scatter data in at least one shadow region. Correcting the radiation data using the scatter image to reduce scatter in the radiation data may include subtracting the scatter image from the radiation data.

[0011] The generating function may include at least one of a narrow-angle kernel, a wide-angle kernel, a Gaussian kernel, a filter or filter bank, an orthonormal basis, a Fourier transform, a wavelet basis, and a continuous wavelet transform. The scattering model may include a convolutional neural net. The scattering model may include a singular value decomposition or an eigenvalue decomposition. The scattering model may include at least one of a least squares method, a weighted least squares method, a conjugate gradient least squares method, and a nonlinear optimization technique. The scattering model basis may include an orthonormal basis. The scattering model basis may include a Gaussian kernel. The scattering model basis may include at least one of an asymmetric Gaussian kernel, a symmetric Gaussian kernel, and a Gaussian kernel with non-zero distortion. Separating the radiation data into a primary component and a scatter component may include filtering the radiation data. Separating the radiation data into a primary component and a scatter component may include segmenting the radiation data. Adjusting parameters in the scattering model to fit the scatter may include an iterative analysis of the radiation data. The iterative analysis may analyze a segmented image including the primary radiation data. The data processing system may be configured to at least one of offset the radiation data, normalize the radiation data, correct the radiation data, and weight one or more portions of the radiation data. The apparatus may further include dynamic positioning of a collimator to manipulate radiation emitted by the radiation source before the radiation is received by the radiation detector.

[0012] Variations may include a method of processing radiation data acquired by a radiation imaging device, the method including receiving the radiation data, separating the radiation data into primary and secondary components, applying an image transform to the primary component using a generating function, constructing a scatter model basis using the transform, adjusting parameters in the scatter model to fit scatter in at least one of the primary and secondary components using the scatter model basis, generating an estimated scatter image using the fitted scatter model, and modifying the radiation data using the scatter image to reduce scatter in the radiation data, thereby generating a scatter-corrected image.

[0013] Features described and / or illustrated with respect to one variation may be used in the same or similar way in one or more other variations and / or in combination with or in place of features of the other variations.

[0014] The description of the invention does not limit the words used in the claims or the scope of the claims or the invention in any way. The words used in the claims have all of their full ordinary meanings.

[0015] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate variations of the present invention and, together with the general description of the invention above and the detailed description below, serve to illustrate variations of the present invention. It will be understood that element boundaries (e.g., boxes, groups of boxes, or other shapes) shown in the figures represent one variation of the boundary. In some variations, one element may be designed as multiple elements, or multiple elements may be designed as one element. In some variations, an element shown as an internal component of another element may be implemented as an external component, and vice versa. Additionally, elements are not necessarily drawn to scale. [Brief explanation of the drawings]

[0016] [Figure 1] 1 is a perspective view of an exemplary imaging device in accordance with one aspect of the disclosed technology; [Figure 2] FIG. 1 is a schematic diagram of an imaging apparatus incorporated into an exemplary radiation therapy device in accordance with an aspect of the disclosed technology. [Figure 3] FIG. 2 is a schematic diagram of an exemplary collimated projection onto a detector. [Figure 4] 1 is a flowchart illustrating an exemplary method for scatter correction. [Figure 5] 1 is a flowchart illustrating an exemplary method of IGRT using an imaging device in a radiation therapy device. [Figure 6] FIG. 6 is a flow diagram of an algorithm 600 for frame-by-frame estimation and correction of scatter in an image using an anchored kernel model. [Figure 7] An exemplary processed image 700 according to the algorithm 600 is shown. [Figure 8A] An exemplary Boolean mask 800a is shown that includes a primary region C and secondary (shadow) regions A and B that can be used for image segmentation in step 606. [Figure 8B] Another exemplary mask 800b is shown. [Figure 9] For example, FIG. 8B shows a separated secondary A / B image 900 that may be produced by application of mask 800b. [Figure 10] An exemplary primary image 1000 is shown in which a mask 800a has provided a primary region C with data and two secondary regions A and B that are set to constants (eg, zero). [Figure 11] Two potential scattering kernels 1102 and 1104 that may be used for the convolution of step 614 are shown. [Figure 12] The output 1200 of the wide angle kernel 1102 is shown. [Figure 13] The output 1300 of the narrow angle kernel 1104 is shown. [Figure 14] The result 1400 of applying the secondary mask 800b to the output 1200 of the wide angle scatter kernel 3202 is shown. [Figure 15] The result 1500 of applying the secondary mask 800b to the output 1300 of the narrow angle scatter kernel 1104 is shown. [Figure 16] 16 shows the overall estimated scatter image 1600 produced by algorithm 600. [Figure 17] 16 shows the measured scatter in the primary region for comparison with the estimated scatter image 1600. DETAILED DESCRIPTION OF THE INVENTION

[0017] The following include definitions of exemplary terms that may be used throughout this disclosure. Both the singular and plural forms of all terms apply to each meaning.

[0018] As used herein, a "component" may be defined as a portion of hardware, a portion of software, or a combination thereof. A portion of hardware may include at least a processor and a portion of memory, where the memory contains instructions to execute. A component may be associated with a device.

[0019] As used herein, "logic," synonymous with "circuitry," includes, but is not limited to, hardware, firmware, software, and / or combinations of each for performing a function(s) or operation(s). For example, depending on the desired application or needs, logic may include discrete logic such as a software-controlled microprocessor, an application-specific integrated circuit (ASIC), or other program logic device and / or controller. Logic may also be embodied entirely in software.

[0020] As used herein, a "processor" includes, but is not limited to, one or more of a microprocessor, a microcontroller, a central processing unit (CPU), and virtually any number of processor systems or standalone processors, such as a digital signal processor (DSP), a field programmable gate array (FPGA), a graphics processing unit (GPU), in any combination. A processor may be associated with various other circuits that support the operation of the processor, such as random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), clocks, decoders, a memory controller, or an interrupt controller. These support circuits may be internal or external to the processor or its associated electronic package. The support circuits are in operative communication with the processor. The support circuits are not necessarily shown separately from the processor in a block diagram or other drawing.

[0021] As used herein, a "signal" includes, but is not limited to, one or more electrical signals, including analog or digital signals, one or more computer instructions, bits or bitstreams, etc.

[0022] "Software," as used herein, includes, but is not limited to, one or more computer-readable and / or executable instructions that cause a computer, processor, logic, and / or other electronic device to function, operate, and / or behave in a desired manner. The instructions may be embodied in various forms, such as a routine, algorithm, module, or program, including separate applications or code from dynamically linked sources or libraries.

[0023] While exemplary definitions are provided above, it is Applicant's intention that these and other terms be given the broadest reasonable interpretation consistent with this specification.

[0024] In a CT scan, the X-ray reference data (I0) is the signal when the patient (and patient table) is not present. The raw data or patient data (I d Once the signal-to-flux ratio (Pr) is acquired, the ratio of flux to signal at each detector element is calculated. If the patient data has only the primary signal (Pr), the logarithm of the ratio is the line integral of the patient's linear attenuation along the corresponding x-ray path. CT images can then be reconstructed from the line integrals measured by all detector elements at many angles around the patient.

[0025] The detected signal includes both the primary signal (Pr) and the scattered signal (Sc) (where I d = Pr + Sc), direct calculation of the ratio of reference signal to detector signal, I0 / Id, is possible by using the scatter component of the signal (S c ), it is no longer the integral of the patient's linear attenuation along the x-ray path (l). Explicitly, the correct line integral must be l = log(I / Pr). Including scatter, the calculated ratio is shown in Equation 1:

number

number

[0026] Scattered S c has a positive value. Without scatter correction, the calculated line integral is smaller than the true line integral without Sc. Reconstruction using the contaminated line integral ld introduces quantitative bias into the image and, qualitatively, reduces contrast and introduces artifacts into the image.

[0027] To address the above scattering issues, clinical CT systems can use hardware techniques to minimize scattering upfront during data acquisition and once data has been acquired, or they can apply software techniques to correct for residual scattering in the measured data, which can be referred to as scatter correction.

[0028] The principle of scatter correction is to estimate the scatter (Sc_est), remove or subtract the estimated scatter from the patient data, and calculate the corrected line integral according to Equation 2.

number

[0029] If the scatter estimate (Sc_est) accurately represents the scatter component of the measured data, Equation 2 yields line integrals that facilitate accurate CT image reconstruction.

[0030] Scatter correction increases noise in the calculated line integrals, which in turn increases noise in the reconstructed image. The variance (noise) of the line integrals without scatter correction is shown in Equation 3:

number

[0031] The variance of the line integral after scatter correction is shown in Equation 4.

number

[0032] Comparing the noise in the scatter-corrected line integral in Equation 4 with the noise in the non-scatter-corrected line integral in Equation 3 shows that even if the scatter estimate (Sc_est) is noise-free, the noise in the calculated line integral is amplified by the factor shown in Equation 5.

number

[0033] The noise amplification described by Equation 5 increases as the percentage of scatter in the measurement data increases. For example, if 50% of the measurement data is scatter, the noise is amplified by a factor of 4. In a cone-beam CT system that uses an anti-scatter grid, residual scatter can be 30% of the data. Using Equation 5, the noise amplification is predicted to be approximately 2. In a cone-beam CT with a flat-panel detector without an anti-scatter grid, scatter can exceed 50% of the total measurement data. The larger the patient, the greater the scatter.

[0034] Noise reduction techniques for scatter correction can include (a) reducing the noise in the estimated scatter, which corresponds to reducing the noise in the second term on the right-hand side of Equation 4; (b) reducing the noise in the scatter-corrected raw data or line integrals; (c) modeling the scatter in the iterative reconstruction as an additive term to the estimated first order and comparing the sum of the estimated first order and the scatter to the measured data; (d) normalizing the noise during reconstruction; and / or (e) filtering / denoising the scatter-corrected image.

[0035] All of these approaches have potential advantages but also drawbacks. Even if method (a) can obtain a nearly noise-free scatter image, the noise is still amplified by the factor shown in Equation 5. This factor can be very large for CT scans with a large amount of scatter, especially for cone-beam CT scans with a large field of view and no anti-scatter grid. Method (b) not only suffers from the drawbacks of method (a), but also potentially loses signal (resolution and contrast). This is because the raw data is filtered to reduce noise. Method (c) requires more sophisticated reconstruction algorithms and longer reconstruction times. It also has limitations in noise reduction. Method (d) can be challenging due to the design of the normalization. The scatter-corrected image is considered a combination of the non-scatter-corrected and scatter-corrected components. The non-scatter-corrected component has lower noise than the scatter-corrected component. Therefore, applying normalization to the entire image tends to over-normalize the low-noise non-scatter-corrected component. The post-reconstruction image processing (filtering) / noise reduction in method (e) shares the same challenges as method (d).

[0036] In the variations disclosed herein, the scatter-corrected image can be treated as a combination of two components: the first is a non-scatter-corrected component, and the second is a scatter-only component. The line integral in Equation 2 can be rewritten as a sum of two components, as shown in Equation 6:

number

[0037] The first term in Equation 6 is the line integral without scatter correction. The second term in square brackets is the scatter-corrected component of the line integral. For analytical reconstruction, reconstructing the corrected line integral corresponds to reconstructing the two terms separately to produce two images, and then summing the two images to obtain the final scatter-corrected image. The image reconstructed from the first term is equivalent to a conventional non-scatter-corrected image (noSC image). The image reconstructed from the second term can be called a scatter-only image (scatter-only image). Equation 7 explicitly shows these components. CT image = noSC image + scatter-only image (7)

[0038] The scatter estimate (Sc_est) can be removed from the patient data using Equation (7). The scatter-only image contains noise and artifacts associated with scatter correction. The noSC image has much less noise than the scatter-only image, and the full CT image is a combination of the two.

[0039] Conventional noise reduction associated with scatter correction, whether performed within the raw data, the reconstruction process, or post-reconstruction, essentially works on the combined noSC and scatter-only components of the image, even though the two components have very different noises. Suppressing noise in the scatter-only component can result in over-smoothing (and therefore resolution degradation) of the noSC component. Techniques that minimize resolution degradation may not be effective at suppressing noise associated with scatter correction.

[0040] The variations described herein achieve improved image quality after scatter correction. Improvements include, for example, both substantial noise reduction and low resolution degradation. In these variations, the scatter-only image can be processed separately and differently from the noSC image to reduce noise and artifacts associated with scatter and scatter correction. More powerful noise-suppression data processing techniques (e.g., filters) can be applied to the noSC image, which has higher noise, to improve noise reduction. At the same time, lighter noise-suppression data processing techniques (e.g., filters) can be applied to the noSC image to minimize resolution loss. These and other methods can independently improve the two image components of Equation 7. The resulting combined image (e.g., a CT image) can achieve an improved compromise between noise reduction and resolution preservation.

[0041] In some variations, using a noSC image to guide noise reduction of a high-noise scatter-only image can be an additional advantage. Because the noSC image has lower noise than the scatter-only image, noise reduction of the scatter-only image can be guided. This guidance can include determining a filter kernel and any associated parameters. Guided noise reduction of the scatter-only image can result in a scatter-only image with comparable or lower noise than the noSC image. Edges in the noSC image can provide reliable edge-preserving guidance for data processing (e.g., filtering) of the scatter-only image. The combined final image (e.g., a CT image) has a noise level comparable to the noSC image, but edge preservation can be improved.

[0042] In a variation, the two components of Equation 6 may be reconstructed in different ways, for example, a higher resolution filter (kernel) may be used to reconstruct the noSC component, while a higher smoothing filter (lower resolution kernel) may be used to reconstruct the scatter-only component.

[0043] The following flow charts and block diagrams illustrate exemplary configurations and methods related to scatter correction and / or image generation. The exemplary methods may be implemented in logic, software, hardware, or a combination thereof. Additionally, while the procedures and methods are presented in a sequence, the blocks may be executed in a different order, including serially and / or in parallel. Thus, the following steps, including imaging, image-based pre-delivery steps, and treatment delivery, are shown sequentially but may be executed simultaneously, including in real time. Furthermore, additional or fewer steps may be used.

[0044] A variation of the disclosed technology involves the use of patient data from imaging scans (I dThe present invention relates to correcting scatter in imaging data, including utilizing a scatter estimate (Sc_est) and a scatter estimate (Sc_est). Imaging scans can be performed with any radiological imaging device, including X-ray, CT, CBCT, SPECT, PET, MR, etc., for the type of scan. These methods can be used for scatter correction of imaging data from these imaging scans, e.g., noise and artifact reduction. While CT scanners and cone-beam CT scanners are highlighted in some exemplary variations, the techniques can also be applied to image reconstruction / data processing based on removal of unwanted counts / signals from original counts, e.g., scatter correction for SPECT, PET, MR, SPECT / CT, PET / CT, PET / MR, etc., to generate corrected images.

[0045] In a variation, imaging scans can be performed using a dedicated imaging device or an imaging device integrated with the radiation therapy delivery device. For example, a radiation therapy delivery device can utilize an integrated low-energy radiation source for CT for use with or as part of IGRT. In particular, a low-energy collimated radiation source for imaging within a gantry using rotational (e.g., helical or step-and-shoot) image acquisition can be combined with a high-energy radiation source for therapeutic treatment, as described, for example, in U.S. patent application Ser. No. 16 / 694,145, entitled "MULTIMODAL RADIATION APPARATUS AND METHODS," filed November 25, 2019, and U.S. patent application Ser. No. 16 / 694,148, entitled "APPARATUS AND METHODS FOR SCALABLE FIELD OF VIEW IMAGING USING A MULTI-SOURCE SYSTEM," filed November 25, 2019, both of which are incorporated herein by reference in their entireties. In these variations, a low energy radiation source (e.g., kilovolts (kV)) can produce higher quality images than using a high energy radiation source (e.g., megavolts (MV)) for imaging.

[0046] Imaging data acquisition methods may include, for example, multiple rotational scans, which may be continuous scans (e.g., with a helical radiation source trajectory about a central axis along with longitudinal movement of the patient support through the gantry bore), non-continuous circumferential stop-and-reverse scans with incremental longitudinal movement of the patient support, step-and-shoot circumferential scans, etc., or may be otherwise used.

[0047] According to a variant, the imaging device collimates the radiation source, for example using a beamformer, into a cone beam or a fan beam, etc. In one variant, a collimated beam can be combined with a gantry that rotates continuously while the patient is moving, thereby enabling helical image acquisition.

[0048] As described in more detail below, detectors (with various row / slice sizes, configurations, dynamic ranges, etc.), scan pitch, and / or dynamic collimation are additional features of the variations, including selectively exposing portions of the detector and selectively defining the active readout area.

[0049] The imaging apparatus and method can provide selective and variable collimation of a radiation beam emitted by a radiation source, including adjusting the radiation beam shape to expose less than the entire active area of ​​an associated radiation detector (e.g., a radiation detector positioned to receive radiation from the radiation source). Exposing only the primary region of the detector to direct radiation allows regions of the detector that are shadowed to receive only scatter. Scatter measurements of the detector's shadow region (and in some variations, measurements of the penumbra region) can be used to estimate scatter in the primary region of the detector that receives projection data.

[0050] 1 and 2, an exemplary imaging device 10 (which may include, for example, an X-ray imaging device) is shown. It will be understood that the imaging device 10 may be associated with and / or integrated into a radiation therapy device (as shown in FIG. 2) that can be used for various applications, including, but not limited to, IGRT. The imaging device 10 includes a rotatable gantry system, referred to as a gantry 12, supported or housed by a support unit or housing 14. Gantry herein refers to a gantry system comprising one or more gantries (e.g., rings or C-arms) that can support one or more radiation sources and / or associated detectors as they rotate around a target. The rotatable gantry 12 defines a gantry bore 16 within which a patient can be moved and positioned for imaging and / or treatment. According to one variation, the rotatable gantry 12 is configured as a slip-ring gantry that provides continuous rotation of the imaging radiation source (X-rays) and associated radiation detectors while providing sufficient bandwidth suitable for high-quality imaging data received by the detectors.

[0051] The patient support 18 is disposed adjacent to the rotatable gantry 12 and is configured to support a patient, typically in a horizontal position, for longitudinal movement into the rotatable gantry 12. The patient support 18 can move the patient, for example, in a direction perpendicular to the plane of rotation of the gantry 12 (along or parallel to the axis of rotation of the gantry 12). The patient support 18 can be operatively coupled to a patient support controller for controlling movement of the patient and the patient support 18. The apparatus 10 is capable of volume-based and planar-based imaging acquisition. For example, in a variant, the apparatus 10 can be used to acquire volumetric and / or planar images and perform the associated processing methods described above.

[0052] As shown in FIG. 2, imaging device 10 includes an imaging radiation source 30 coupled to or supported by rotatable gantry 12. Imaging radiation source 30 emits a radiation beam (generally designated 32) for producing high-quality images. In this variation, the imaging radiation source is an X-ray source 30 configured as a kilovoltage (kV) source (e.g., a clinical X-ray source having an energy level ranging from about 20 kV to about 150 kV). The imaging radiation source can be any type of transmission scanning source suitable for imaging. Other imaging transmission scanning sources can be used interchangeably in various other variations.

[0053] The imaging device 10 may include another radiation source 20 coupled to or supported by the rotatable gantry 12. According to one variation, the radiation source 20 is configured as a therapeutic radiation source, such as a high-energy radiation source used to treat a tumor in a patient within a region of interest. It will be understood that the therapeutic radiation source may be a high-energy X-ray beam (e.g., a megavoltage (MV) X-ray beam). Generally, the radiation source 20 has a higher (e.g., peak and / or average) energy level than the imaging radiation source 30. While FIGS. 1 and 2 show the X-ray imaging device 10 with the radiation source 30 mounted on the ring gantry 12, other variations may include other types of rotatable imaging devices, including, for example, C-arm gantries and robotic arm-based systems.

[0054] A detector 34 (e.g., a two-dimensional flat or curved detector) may be coupled to or supported by the rotatable gantry 12. The detector 34 (e.g., an X-ray detector) is positioned to receive radiation from the imaging radiation source 30 and may rotate with the imaging radiation source 30. The detector 34 may detect or measure the amount of unattenuated radiation and thus infer the amount of radiation actually attenuated by the patient or associated patient ROI (compared to the amount of radiation originally generated). The detector 34 may detect or collect attenuation data from different angles as the radiation source 30 rotates and emits radiation toward the patient.

[0055] A collimator or beamformer assembly (generally designated 36) is positioned relative to the imaging source 30 to selectively control and adjust the shape of the radiation beam 32 emitted by the imaging radiation source 30 to selectively expose a portion or region of the active area of ​​the detector 34. The beamformer can also control how the radiation beam 32 is positioned on the detector 34. For example, in one variation, each readout may capture 3-4 centimeters of projection image data, with approximately 1-2 centimeters of unexposed detector area on one or both sides that can be used to capture scatter data.

[0056] A detector 24 may be coupled to or supported by the rotatable gantry 12 and positioned to receive radiation 22 from the therapeutic radiation source 20. The detector 24 may detect or measure the amount of unattenuated radiation and thus infer the amount of radiation actually attenuated by the patient or associated patient ROI (compared to the amount of radiation originally generated). The detector 24 may detect or collect attenuation data from different angles as the therapeutic radiation source 20 rotates and emits radiation toward the patient.

[0057] The therapeutic radiation source 20 may be mounted, configured, and / or moved in the same plane or in a different plane (offset) than the imaging source 30. In some variations, scattering caused by simultaneous activation of the radiation sources 20, 30 can be reduced by offsetting the radiation planes.

[0058] When integrated with a radiation therapy device, the imaging device 10 can provide images used to set up (e.g., align and / or register), plan, and / or guide a radiation delivery procedure (treatment). Typical setup is accomplished by comparing current (in-treatment) images with pre-treatment image information. Pre-treatment image information may include, for example, X-ray, CT, CBCT, MR, PET, SPECT, and / or 3D rotational angiography (3DRA) data, and / or any information obtained from these or other imaging modalities. In some variations, the imaging device 10 can track the movement of the patient, target, or ROI during treatment.

[0059] A reconstruction processor 40 may be operatively coupled to the detectors 24, 34. In one variation, the reconstruction processor 40 is configured to generate patient images based on radiation received by the detectors 24, 34 from the radiation sources 20, 30, as described above. It will be understood that the reconstruction processor 40 may be configured to perform the methods described herein. The apparatus 10 may also include a memory 44 suitable for storing information including, but not limited to, data processing and reconstruction algorithms and software, including filters and data processing / filter parameters, imaging parameters, image data from previous or otherwise previously acquired images (e.g., planning images), treatment plans, etc.

[0060] The imaging device 10 may include an operator / user interface 48 to enable an operator of the imaging device 10 to interact with or otherwise control the imaging device 10 to provide input related to scan or imaging parameters, etc. The operator interface 48 may include any suitable input device, such as a keyboard, mouse, voice-activated controller, etc. The imaging device 10 may also include a display 52 or other human-readable element for providing output to the operator of the imaging device 10. For example, the display 52 may allow the operator to observe reconstructed patient images and other information, such as imaging or scan parameters, related to the operation of the imaging device 10.

[0061] 2 , imaging apparatus 10 includes a controller (generally designated 60) operably coupled to one or more components of apparatus 10. Controller 60 controls the overall function and operation of apparatus 10, including providing power and timing signals to imaging source 30 and / or therapeutic radiation source 20, and a gantry motor controller that controls the rotational speed and position of rotatable gantry 12. It will be understood that controller 60 can include one or more of a patient support controller, a gantry controller, a controller coupled to therapeutic radiation source 20 and / or imaging source 30, a beamformer 36 controller, a controller coupled to detector 24 and / or detector 34, etc. In one variation, controller 60 is a system controller that can control other components, devices, and / or controllers.

[0062] In variations, the reconstruction processor 40, operator interface 48, display 52, controller 60, and / or other components may be combined into one or more components or devices.

[0063] The device 10 may include various components, logic, and software. In one variation, the controller 60 includes a processor, memory, and software. By way of example and not limitation, the imaging device and / or radiation therapy system may include various other devices and components (e.g., a gantry, a radiation source, a collimator, a detector, a controller, a power supply, a patient support, among others) that can perform one or more routines or steps related to imaging and / or IGRT for a particular application, where the routines may include imaging, image-based pre-delivery steps, and / or treatment delivery, including the settings, configurations, and / or positions (e.g., paths / trajectories) of the respective devices, which may be stored in memory. Additionally, the controller(s) may directly or indirectly control one or more devices and / or components according to one or more routines or processes stored in memory. An example of direct control is the setting of various radiation source or collimator parameters (power, speed, position, timing, modulation, etc.) related to imaging or treatment. An example of indirect control is the communication of position, path, speed, etc. to the patient support controller or other peripheral devices. The hierarchy of various controllers that may be associated with the imaging device may be arranged in any suitable manner to communicate appropriate commands and / or information to the desired devices and components.

[0064] Furthermore, those skilled in the art will appreciate that the systems and methods may be implemented in other computer system configurations. The illustrated aspects of the present invention may be implemented in a distributed computing environment where certain tasks are performed by local or remote processing devices linked through a communications network. For example, in one variation, the reconstruction processor 40 may be associated with a separate system. In a distributed computing environment, program modules may be located in both local and remote memory storage devices. For example, the imaging device 10 may utilize a remote database, a local database, a cloud computing platform, a cloud database, or a combination thereof.

[0065] The imaging device 10 can utilize an exemplary environment for implementing various aspects of the present invention, including a computer, which includes a controller 60 (including, for example, a processor and memory, which may be memory 44) and a system bus. The system bus can couple system components, including, but not limited to, the memory, to the processor, and can communicate with other systems, controllers, components, devices, and the processor. The memory can include read-only memory (ROM), random access memory (RAM), hard drives, flash drives, and any other form of computer-readable media. The memory can store various software and data, including routines and parameters that may include, for example, a treatment plan.

[0066] Image quality is determined by many factors (e.g., imaging source focal spot size, detector dynamic range, etc.). A limitation of many imaging techniques and image quality is scatter. Various techniques can be used to reduce scatter. One technique is to use an anti-scatter grid (which collimates the scatter). However, implementing a scatter grid with motion tracking and correction on a kV imaging system can be problematic. As mentioned above, to improve the quality of the image data, an accurate estimation of scatter in the projection data is necessary. In a variant, scatter in the projection data acquired in the primary region of the detector 34 can be estimated based on data measured in the shadow region (and penumbra) of the detector 34.

[0067] FIG. 3 is a schematic diagram of an exemplary collimated projection 300 onto a detector 302. A rotating radiation source 306 (e.g., X-ray) is shown emitting a radiation beam 308 that exposes a primary or central (C) region 310 of the detector 302 to direct radiation from the rotating radiation source 306 (e.g., having passed through a target) as the rotating radiation source 306 rotates about the y-axis. Movement of the patient support (not shown) can be axial (longitudinal) movement along the y-axis, which is included as part of the scan as described above. The detector 302 also has a posterior (B) shadow region 312 and a front (A) shadow region 314 that are blocked from direct exposure to the radiation beam 308 by a beamformer / collimator 320. The beamformer / collimator 320 is configured to adjust the shape and / or position of the radiation beam 308 emitted by the rotating radiation source 306 onto the detector 302. The shadowed regions 312, 314 receive only scattered radiation.

[0068] The opening of the collimator 320 is configured so that the posterior (B) end 312 and the anterior (A) end 314 of the detector 302 in the axial or longitudinal direction (toward the patient table or along the y-axis) are not illuminated by direct radiation 308. These shadow regions of the posterior (B) 312 (in the negative longitudinal direction along the rotational y-axis) and anterior (A) 314 (in the positive longitudinal direction along the rotational y-axis) do not receive direct radiation and can therefore be utilized for scatter measurements. For example, the readout range of the detector 302 can be configured to read out all or a portion of the data within one or more shadow regions 312, 314 and use the data for scatter estimation of the primary region 310. The primary or central (C) region 310 receives both direct projection and scatter.

[0069] In variations, the data processing system (e.g., including the processor 40) may be configured to receive measured projection data in the primary region 310 and measured scattering data in at least one shadow region 312, 314, and then determine estimated scattering in the primary region 310 based on the measured scattering data in the at least one shadow region 312, 314. In some variations, determining the estimated scattering in the primary region 310 during a current rotation may be based on measured scattering data in at least one shadow region 312, 314 during an adjacent (previous and / or subsequent) rotation. In other variations, measurement data from penumbra region(s) (bordering the primary region and shadow region) may also be used for scattering estimation.

[0070] Various techniques and methods can utilize different scan geometries, detector placements, and / or beamformer window shapes. In some variations, the detectors may also be laterally offset.

[0071] 4 is a flowchart illustrating an exemplary method 400 of scatter estimation and correction as described above. Inputs can include any previous data and / or scan design. In this variation, step 410 includes data acquisition. For example, during rotation of a radiation source that projects a collimated radiation beam toward a target and a radiation detector, the method measures projection data (primary + scatter) of a central (primary) region of the radiation detector and measures scatter using a front shadow peripheral region and / or a back shadow peripheral region of the detector. Data acquisition in step 410 can also include adjusting the shape / position of the radiation beam using a beamformer before and / or during the scan and / or adjusting the readout range (including determining the active region).

[0072] Next, step 420 includes scatter estimation. For example, the method uses scatter measurements from the shadow region(s) to estimate scatter in the projection data from the central (primary) region. Next, step 430 includes scatter correction, which can include either of the two component techniques described above. The output includes scatter-corrected projection data suitable for imaging. Various variations can utilize different scan geometries, detector placement / active areas, beamformer placement / window shapes, etc.

[0073] FIG. 5 is a flowchart illustrating an exemplary method 500 of IGRT using a radiation therapy device (e.g., including imaging device 10). Patient prior image data 505 can be used. The prior image data can be previously acquired planning images, including prior CT images. The prior data 505 can also include treatment plans, phantom information, models, prior information, etc. In some variations, the prior image data 505 is generated by the same radiation therapy device, but at an earlier time. In step 510, imaging of the patient is performed using a low-energy radiation source (e.g., kV radiation from X-ray source 30). In one variation, the imaging includes a helical scan with a fan or cone beam geometry. Step 510 can generate high-quality (HQ) image(s) or imaging data 515 using the scatter estimation and correction techniques described above. In some variations, the image quality can be adjusted to improve the balance between image quality / resolution and radiation dose. In other words, not all images need to be of the highest quality, or the image quality can be adjusted to improve or trade off the balance between image quality / resolution and image acquisition time. The imaging step 510 may also include image / data processing to generate patient images based on the imaging data (e.g., according to the methods described above). Although the image processing step 520 is shown as part of the imaging step 510, in some variations, the image processing step 520 is a separate step, including when image processing is performed by a separate device.

[0074] Next, in step 530, one or more image-based pre-delivery steps, described below, are performed based at least in part on the imaging data 515 from step 510. As described in more detail below, step 530 can include determining various parameters related to the therapeutic treatment and (subsequent) imaging planning. In some variations, the image-based pre-delivery step (530) can require additional imaging (510) before treatment delivery (540). Step 530 can include adapting the treatment plan based on the imaging data 515 as part of an adaptive radiation therapy routine. In some variations, the image-based pre-delivery step 530 can include real-time treatment planning. Variations can also include simultaneous, overlapping, and / or alternating activation of imaging and therapeutic radiation sources. Real-time treatment planning can include any or all of these types of imaging and therapeutic radiation (simultaneous, overlapping, and / or alternating) activation techniques.

[0075] Next, in step 540, therapeutic treatment delivery is performed using high-energy radiation (e.g., MV radiation from therapeutic radiation source 20). Step 540 delivers a therapeutic dose 545 to the patient according to the treatment plan. In some variations, IGRT method 500 may include returning to step 510 at various intervals for additional imaging, followed by image-based pre-delivery steps (530) and / or treatment delivery (540) as needed. In this manner, high-quality imaging data 515 can be generated and utilized during IGRT using a single device 10 capable of adaptive therapy. As described above, steps 510, 520, 530, and / or 540 may be performed simultaneously, overlapping, and / or interleaved.

[0076] In variations, the various methods described above can be utilized for scatter correction regardless of whether the imaging data is generated using a dedicated imaging device or an imaging device integrated with the radiation therapy delivery device.

[0077] In one variation, the CT device includes a rotating X-ray source and a set of raw data for CT image generation (e.g., I d ), and hardware and / or software for measuring and / or generating a set of scatter data (e.g., Sc_est) to compensate / correct for scatter contamination in the raw data. A non-scatter corrected image is reconstructed from the raw data, and a scatter-only image is reconstructed from the scatter data. In this variation, the raw data can be used to calculate the non-scatter corrected line integrals for the reconstruction of a non-scatter corrected CT image. The scatter data can be used to calculate the scatter-only line integrals for the reconstruction of a scatter-only image according to Equation 6. The non-scatter corrected and scatter-only images are processed independently, with the latter being more heavily filtered due to its higher noise. The processed non-scatter corrected image and the processed scatter-only image can be combined to create a final scatter corrected CT image.

[0078] In another variation, volumetric image subtraction may be used to generate a scatter-only image. Here, the scatter data is used with the raw data to generate scatter-corrected line integrals for reconstructing a scatter-corrected image. The raw data can be used to calculate non-scatter-corrected line integrals to reconstruct a non-scatter-corrected image. The non-scatter-corrected image can be subtracted from the scatter-corrected image to obtain a scatter-only image. The non-scatter-corrected and scatter-only images are processed independently, with the latter being more heavily filtered due to its higher noise. The processed non-scatter-corrected and processed scatter-only images can be combined together to create a final scatter-corrected CT image.

[0079] In a variation, the non-scatter corrected image can be used to guide the processing of the scatter-only image to achieve effective noise and artifact reduction in the scatter-only image while preserving edges in the image. For example, the filter can be a Gaussian filter that uses voxel differences in the non-scatter corrected image to determine kernel weights for the scatter-only image filter. In this way, edge information in the non-scatter corrected image is used to preserve corresponding edges in the scatter-only image. The non-scatter corrected image can also be used in more advanced edge-preserving processing schemes to improve processing of the scatter-only image. For example, the scatter-only image can be processed based on anisotropic derivative filter parameters obtained in the non-scatter corrected image.

[0080] In another variation, the non-scatter-corrected image and the scatter-only image can be reconstructed using different reconstruction schemes. For example, the non-scatter-corrected image can be reconstructed using a higher resolution kernel than the scatter-only image, and the scatter-only image can be reconstructed using a customized streak artifact reduction algorithm. The scatter-only image can be reconstructed using a different grid to reduce reconstruction time. For example, if the reconstruction of the non-scatter-corrected image uses a 512 x 512 matrix, the reconstruction of the scatter-only image can use a 256 x 256 matrix for reconstruction to reduce reconstruction time. The reconstructed scatter-only image can then be resampled to the same grid as the non-scatter-corrected image. The non-scatter-corrected image can then be used to guide the processing of the scatter-only image. The resulting scatter-only image can be combined with the non-scatter-corrected image to create a final scatter-corrected image.

[0081] In addition to the CT environment highlighted in some of the exemplary variations, various other variations involve scattering (e.g., I dVarious imaging modalities acquire or generate raw data with scatter data (e.g., Sc_est) and scatter data (e.g., Sc_est), such as SPECT and PET, can use the scatter data to correct the raw data. The scatter data can be used to correct / correct the line integrals, which can be decomposed into a linear combination of components without and with scatter correction, similar to Equation 6. The non-scatter-corrected image has less noise than the scatter-only image. The two images can be reconstructed differently to improve the quality of both, and then combined to obtain the final image. The reconstructed non-scatter-corrected and scatter-only images can be processed independently to improve the quality of both, and then combined to obtain the final image. The non-scatter-corrected image can also be used as a guide image to determine the weights of the filtering kernel when processing the scatter-only image.

[0082] In addition to variants that utilize the non-scatter-corrected image to guide the processing of the scatter-only image (i.e., operate in the image domain), other variants can operate in the data domain. In these variants, the processing of the generated line integrals of the scatter-only component can be based on the line integral data of the non-scatter-corrected component as guiding data that preserves edges in the scatter-only component. The resulting line integrals of the scatter-only component can be reconstructed separately or together with the line integrals of the non-scatter-corrected component.

[0083] In a variant, the raw data (e.g., I d) and the measured scatter data (e.g., Sc_est) are used together to reconstruct a scatter-corrected image, and the raw data is used to reconstruct a non-scatter-corrected image using various reconstruction algorithms to obtain the image. In some variations, the reconstruction may be analytical reconstruction. In some variations, the reconstruction may be iterative reconstruction. In variations, the scatter-only image is processed (including filtering, artifact reduction, etc.) separately from the non-scatter-corrected image and then combined with the non-scatter-corrected image to obtain the final image. In some variations, the scatter-only image is generated by subtracting the non-scatter-corrected image from the scatter-corrected image. Furthermore, the non-scatter-corrected image can be used to guide the processing of the scatter-only image for optimal noise and artifact reduction and edge preservation.

[0084] 6 is a flow diagram of an algorithm 600 for estimating scattering in an image on a frame-by-frame basis using an anchored kernel model. Algorithm 600 can be used to estimate scattering in conjunction with any of the other methods described herein (e.g., methods 400 or 500).

[0085] More specifically, algorithm 600 estimates scattering using selective beam blocking in conjunction with a mathematical representation of scattering.

[0086] An appropriate mathematical representation of scattering may include, for example, a Gaussian Mixture Model (GMM). In a GMM representation, the Gaussian kernel parameters (e.g., amplitude and / or variance) of the detected beam are estimated by fitting to measurement parameters, including scattering measured from beam interception. GMMs are particularly useful for fitting lower spatial frequency image components associated with physical scattering mechanisms imparted by the projection of physical objects. Briefly, GMMs can provide a multi-resolution form of blurring operator for capturing both wide-angle and narrow-angle scattering.

[0087] GMM is just one way to represent scattering. It should be understood that algorithm 600 is general with respect to mathematical representations. Other representations that may be used in conjunction with algorithm 600 include those using non-Gaussian representations, scattering kernel representations, continuous wavelet transforms, Fourier transforms, machine learning (e.g., neural nets), etc.

[0088] The method 600 uses a collimator (e.g., collimator 320 in FIG. 3 ) to generate an image I having secondary A / B and primary C portions. The collimator provides an X-ray source-side beam stop (e.g., on the side of source 306 shown in FIG. 3 ). The collimator 320 can have collimator blades 320a and 320b positioned cranially, caudally, ipsilaterally, and / or contralaterally from the patient's perspective. While single blades 320a and 320b are shown in FIG. 3 , other collimator configurations are possible. For example, patterned beam stops can be used, as can collimators with other shapes and configurations. While static beam stop device positions are shown in FIG. 3 , dynamically positioned beam stops and / or collimators during a scan may also be used. Exemplary applications include using dynamic beam collimation to perform region-of-interest (ROI) volume imaging and radiation-dose-reduced target tracking. The collimating blades can be "planned" to follow the search space of the target's motion or, for example, for ROI reconstruction during treatment. In the case of dynamic motion, at each frame instant the position of the shadow can be used in the described way.

[0089] Such beam collimation can form rectangular shadows 312 and 314 on the flat panel X-ray detector 302. The three regions that can form (e.g., primary C, penumbra (region between C and A and B, not shown), and secondary (i.e., blocked A and B)) can be automatically detected, e.g., frame by frame. They can also be known from geometry calibration (e.g., periodic calibration, previous calibration) of the beam 308.

[0090] Algorithm 600 represents anchored scatter image estimation using measured secondary data and a scattering basis model 600a (e.g., as a GMM where Gaussian component amplitudes are estimated from X-ray absorption data and scattering from a collimation scheme such as that shown in FIG. 3 ). While algorithm 600 is described in detail with respect to collimation scheme 300, it should be understood that the concepts described herein with respect to algorithm 600 are general with respect to the details of the collimation scheme and may be used with collimation schemes other than 300.

[0091] Although a particular mathematical model 600a for fitting scatter (e.g., a particular GMM configuration and construction / fitting protocol) is described below with respect to algorithm 600, algorithm 600 can be configured in several different ways. The scatter model can be pre-configured for imaging system 300 by fitting it to measurements of scatter images from a phantom (patient and / or other imaging sample) and the angle of gantry 12. This can include, for example, adjusting for Gaussian variance in the GMM. This fitting can include measured differences, such as images between a full panel exposure of detector 302 and a pencil raster scan or slit scan sweep of source 306 and detector 302. Other types of pre-configuration are possible and within the scope of this disclosure. However, it should be understood that such pre-configuration is optional and not required to implement algorithm 600.

[0092] As described in more detail below, in certain variations, model 600a can be constructed by solving a system of equations formed by vectorizing image pixels in the shadow or "secondary" regions (regions A, B in FIG. 3) after applying a mask into a data vector (b). In this construction based on the secondary regions A and B, the unknowns can be, for example, the amplitudes of the GMM components (x). Because the secondary regions A and B contain pure background scattering signals (i.e., scattering uncontaminated by signal from the sample), this construction allows the GMM to be calibrated for scattering. The primary region C contains both scattering and signal from the sample (e.g., patient or phantom).

[0093] A solvable system matrix A can then be formed by mathematically convolving the components x of the initial model 600a with pixels in the primary region (region C in FIG. 3) of the projection image frame I. When the convolution begins, the components x of the model 600a may have arbitrary amplitudes, or the components of the model 600a may be pre-configured, as described briefly above and in more detail below. The result of the convolution at each pixel in the shadow region (A and B) corresponding to the ordering of the data vector b is an element of the system matrix A, an estimate of the scatter in the primary region C. Any suitable regression algorithm (e.g., constructing A via weighted least squares provides an estimate of the GMM model scatter amplitude) can be used.

[0094] The result of the convolution at each pixel in the shadow regions (A and B) is an estimate of the scatter at that pixel. The corresponding element of the data vector (b) is the measured scatter at that pixel. Matrix A can be solved by matching the estimated scatter with the measured scatter in the data vector (b). Once A is solved, it can be used to convolve with the data in the primary region C to estimate the estimated scatter in region C. The result can be used to scatter correct the primary data in region C. The method used to solve matrix A can be least squares, weighted least squares, steepest descent, or any other suitable technique for image convolution. Any method that solves a system of equations based on input (measurement projections including primary and secondary regions), prior assumptions (e.g., kernel models such as GMM), and outputs a scatter image in the primary region. The specific steps of algorithm 600 are described below.

[0095] The following description applies algorithm 600 and model 600a to a single image frame. It should be understood that steps may be applied to each frame within a multi-frame image. Once processed and scatter corrected, multiple image frames may be reconstructed to form a tomographic 3D representation. It should be understood that each step described below is optional. Algorithm 600 may be performed without one or more of the steps described below. It may also be performed with one or more steps performed in a different order than described below. It may also be performed with one or more additional steps not described below. All such variations should be considered within the scope of this disclosure and of algorithm 600.

[0096] In step 602, the algorithm 600 performs corrections (e.g., offsets, bad pixel corrections, and gain Normalization) can be applied to the input signal, which is the raw X-ray signal detected by the detector 302. The input signal may be in the form of an image frame I.

[0097] In step 602, statistically (or otherwise) anomalous pixels may be identified, modified, and / or removed. This may correspond to detecting pixels whose intensity differs from neighboring pixels by a threshold amount. These anomalous pixels may arise from several causes, including defective parts of the detector, interference, stray x-rays scattered from other parts of the acquisition setup 300, and / or other problems in data acquisition. Once identified, the anomalous pixels may be replaced with averaged (or other valued) pixels.

[0098] In step 602, image frame I may be altered for faster and / or more efficient image processing. For example, image I may be downsampled or otherwise reduced or decomposed in resolution to a larger pixel size or other basis function. Multi-resolution techniques can extract lower spatial frequency components from the primary image for use in the method. This can reduce the complexity of further processing and potentially speed up image processing, such as scatter correction, as described below. Image downsizing is particularly advantageous for image I or portions of image I that do not have significant spatial variation or spatial frequency. In these cases, downsizing preserves much of the desired information in the original image frame I. Scatter signal image components typically have relatively low spatial variation. For this reason, it may be advantageous to downsize the primary A / B image portion and the secondary A / B image portion in this preprocessing step. This may enable the scatter signal to be quickly and accurately processed to correct image frame I.

[0099] Step 602 results in an image I, such as exemplary processed image 700 shown in FIG. 7. Exemplary image 700 is a projection of a lung phantom in primary beam C from source 306 after preprocessing according to step 602. For example, two sections 702 and 704 of the lung phantom can be seen as bright regions in primary section C of image 700. Here, X-rays passing through the lungs are less attenuated, resulting in higher values ​​in the lungs (i.e., they appear brighter in the image). Primary section C of image 700 corresponds to image section C of detector 302 shown in FIG. 3. In other words, it is the detection region where gaps in collimator 320 allow radiation from source 302 to pass through the lung phantom sample. However, regions A and B in image 700 represent pure scattering because they correspond to regions where incident radiation from source 306 was blocked by collimator blades 320a and 320b.

[0100] The shadows or secondary regions A and B on either side of primary region C in image 700 each contain scatter common to primary beam region C. Primary region C may contain multiple scatter types (e.g., scatter from the object, patient, radiation source housing, filtration, collimation device components or bore cover, and various layers of x-ray detector 302 imaged in primary beam region C). Thus, the pure scatter from secondary regions A / B may be useful for fitting model 600a to represent much of the scatter present throughout image 700.

[0101] In step 606, image I can be segmented into a primary C region and secondary (A and B) regions. In this context, "segmenting" means creating two different images from image 700 that can be processed separately. Each of the two images represents a segment or portion of image 700. In this example, one segmented image can include a portion of image 700 from secondary portions A / B (FIG. 7). The other segment can include a portion of image 700 from primary portion C. However, these are merely examples, and it should be understood that any suitable and advantageous segmentation of image 700 is possible during this step. The segmentation in step 606 can be performed using any suitable method. A suitable segmentation method includes, for example, applying an imaging mask to image I that enables operations on subregions of the image data. Segmentation can be based on X-ray projection images through air to clearly delineate the primary, penumbra, and shadow regions of the image. Frame-by-frame segmentation based on measured object data may also be possible. Segmentation can be based on a computational model of the imaging geometry and the physical radiation source device.

[0102] FIG. 8A shows an exemplary Boolean mask 800a including a primary region C and secondary (shadow) regions A and B that can be used for image segmentation in step 606. Mask 800a can separate data in primary region C from secondary regions A and B by indexing or other software abstraction. Mask 800a can index everything except primary region C. Mask 800b in FIG. 8B can also be used for segmentation step 606. Mask 800b can index everything except secondary regions A and B. In addition to masking, step 606 can include and apply any suitable imaging segmentation routine based on a known imaging geometry (e.g., geometry 300 shown in FIG. 3 including 320 and detector 302). Segmentation can be performed using any of the methods described herein.

[0103] Another advantage of image segmentation is that the images from the primary C and secondary A / B may have different characteristics that are suitable for different processing methods. For example, the pixel size or other basis decomposition may be different, the filtering may be different, and the weighting factors may be different between the primary C and secondary A / B regions. On the other hand, the image segment from the primary segment C may contain greater spatial resolution and finer spatially resolved details than the secondary A / B portion. In that sense, the segmented image corresponding to the primary portion C may benefit from the application of image processing routines that can advantageously process such details.

[0104] Figure 9 shows a separated secondary A / B image 900 that may be produced, for example, by application of mask 800b in Figure 8B. Secondary image 900 contains data for only secondary regions A and B.

[0105] 10 shows an example image 1000 in which a mask 800a has isolated a primary region C. The secondary image 900 and the primary image 1000 can be subjected to different processing and then recombined to form a scatter-corrected composite image I. The output from the scatter correction may be only the corrected primary region. Although the secondary region may no longer be available for radiographic or CT applications, the corrected secondary region can be evaluated for estimation quality.

[0106] In step 606, an analysis of at least one of images 900 and 1000 is performed. As an example, the analysis of image 1000 is shown schematically in FIG. 10 . An exemplary analysis includes an analysis region 1002 in a secondary segment and an analysis region 1004 in a primary segment. Region 1002 can be scanned across image 1000 so that the entire image 1000 is analyzed. It should be understood that the scattering estimation algorithm utilizes primary data from the entire image 1000 or region 1004 and secondary data from the entire image 1000 or region 1002. An advantage of traversing image 1000 region by region is that the estimated model parameters can vary laterally because scattering behavior can change with ray angle depending on object thickness or other factors. An advantage of regions 1002 and 1004 possibly having different widths is that the scattering data of 1002 contributes to a wider range of the object under illumination.

[0107] Image regions 1002 and 1004 may be the focus of image processing calculations (e.g., pixel vectorization, maximum / minimum determination, averaging, Fourier transform, etc.) The calculations may include any suitable processing method or may combine one or more such methods.

[0108] Image regions 1002 and 1004 may be processed simultaneously with other image regions (not shown) so that parallel analysis of image 1000 is performed at once. Instead of using image regions, image 1000 may be analyzed in a single step. Processing in step 602 may continue iteratively with other steps described below. In particular, processing in step 606 may be performed until the complete scatter image has been estimated.

[0109] One approach is to iteratively estimate the scattering data of the secondary segments after scatter correction of the primary segments using previous estimates. The advantage is that the physical scattering mechanism is related to attenuation through the object, but the raw image data in the primary region also contains the scattering signal. In another approach, the scattering basis functions can be estimated and corrected continuously across the entire image. For example, the kernels used in the scattering model 600a can be estimated individually, moving from higher spatial frequencies to lower spatial frequencies, or vice versa. More generally, the selected basis functions can be estimated individually and iteratively.

[0110] In step 612, algorithm 600 can apply weighting to emphasize or de-emphasize particular features of image I. For example, weighting can be selected to de-emphasize the effect of air-only paths on the scatter estimate or to amplify the relative contribution of low image values ​​to the scatter estimate. Any suitable mathematical operation can be used to weight the images. Examples of suitable operations include image subtraction or the application of a logarithmic (“log”) operator (e.g., a negative log operator to remove portions of the image). A suitable negative log operator can, for example, transform the gain-normalized projection frame of image I (or images 900 and 1000) so that high image values ​​(e.g., those received via paths that do not traverse a patient, phantom, or other sample and therefore do not include object scattering effects) are attenuated or zeroed out.

[0111] In step 614, the algorithm 600 creates basis functions for the model 600a (e.g., a GMM) using data from the image 1000. FIG. 11 shows two exemplary scattering kernels 1102 and 1104 from a GMM variation that generate scattering model basis functions through convolution with the image 1000. It should be understood that the kernels 1102 and 1104 are merely exemplary, and any suitable kernel, generating function, or basis decomposition may be used. Any number of scattering kernels may be used in this step. This step may further include any suitable selection of a basis function decomposition and a selection of basis functions therein. For example, any Gaussian or non-Gaussian basis and / or combinations thereof may be used. Other bases may also be used in this step, including orthonormal bases, wavelet bases, continuous wavelet transforms, and / or combinations thereof. If kernels are used, they may be symmetric or asymmetric, skewed or undistorted.

[0112] Scattering kernel 1102 is an exemplary wide-angle kernel, and scattering kernel 1104 is an exemplary narrow-angle scattering kernel. Using both wide-angle and narrow-angle scattering kernels together may provide model 600a with geometrically different scattering mechanisms. Combinations of one or more of these and / or other kernel types may further improve the accuracy of the model. More generally, model 600a may include one or more kernel types and multiple kernels for more accurate modeling of scattering. Any suitable number of kernels or basis functions may be used in algorithm 600.

[0113] In convolution-based variants, the convolution in step 614 can be performed using any suitable mathematical convolution algorithm or integral transform. For example, if a GMM is used in model 600a, the convolution may be performed by integrating the product of each basis function of the GMM and the image data over the image. One or more of the basis functions used by model 600a to represent the image data may be inverted and / or shifted during convolution or filtering. Other suitable convolution processes are also possible. The convolution in this step may be achieved using a series of convolution steps that include one or more of these convolution processes. Other algorithms. In a regional approach using image domain 1002, the convolution-based variant can be applied to a wider area of ​​the primary domain C, and then only those narrower primary domains can be extracted from the output. Scatter from the entire primary domain C contributes to the narrower primary domain. Performing the convolution in the Fourier domain can provide computational advantages. Generalization to kernel convolution can be achieved, for example, via a filter bank or multi-resolution decomposition. The weighting of the filter bank or basis set can be imposed or calibrated by physical scattering processes.

[0114] Typically, the convolution in step 614 includes image region data from the primary image C (i.e., segmented image 1000 of FIG. 10 ), since scattering is generated by the primary beam. This step constructs the basis functions for model 600 a. However, in some cases, the convolution step may also include data from the secondary image 900. The convolution can be applied to the primary C and the penumbra area. It can be assumed that there is little or no primary radiation on the secondary (shadow) region A / B, and therefore little additional scattering contribution from the direct path from the radiation source to image I pixels within region A / B.

[0115] In step 616, the algorithm 600 creates a scattering basis by reconstructing model 600a basis functions, for example, from a convolution of the scattering kernel of step 614 with the image data. FIG. 12 shows the output 1200 of this process for a wide-angle kernel 1102. The wide-angle kernel convolution produces output 1200 with low variance 1202 across the image. FIG. 13 shows the output 1300 of this process for a narrow-angle kernel 1104. As shown in FIG. 13, the convolution with the narrow-angle kernel produces local maxima 1302 and 1304 near low-attenuation regions of the image. The outputs 1200 and 1300 comprise an exemplary model basis 600a.

[0116] In step 616, a secondary mask 800b can be applied to the convolved image data outputs 1200 and 1300 from step 616. FIG. 14 shows the result 1400 of applying the secondary mask 800b to the output 1200 of the wide-angle scattering kernel 3202. As shown in FIG. 14, the effect of the wide-angle kernel maximum 1202 can be seen in the masked image 1400. FIG. 15 shows the result 1500 of applying the secondary mask 800b to the output 1300 of the narrow-angle scattering kernel 1104. FIG. 15 shows the effect of the two maxima 1302 and 1304 from the narrow-angle kernel on the masked image 1500.

[0117] In step 616, the quadratic domain (A and B) model 600a basis 1...N is vectorized and added to the matrix (A). The basis is a scalar x1...x N The weighted basis sum (Ax) is then calculated. More specifically, the secondary image can be vectorized into a data vector (b) via the following equation:

number

[0118] In step 616, equation (1) may be weighted by a weighting matrix (W).

number

[0119] In step 626, the system of equations x in Equation (3) can be solved for the scattering estimate (x̂). The solution can be found in any number of suitable ways, including through regression techniques, machine learning techniques, various types of (e.g., convolutional) neural networks, or other techniques such as singular value decomposition or eigenvalue decomposition. One such exemplary method is a weighted least squares (WLS) solution estimate (x̂). This estimate x̂ represents a model component (e.g., a GMM component) suitable for estimating scattering S using Equation (1) above.

[0120] In step 628, an estimated scatter image (Ŝ) is calculated as the sum of fitted scatter kernel convolutions (Kx̂).

number

[0121] In step 630, the solution estimate S from step 626 can be modified with respect to the original image frame I, specifically so that the output data matches the pixel size of the input data. For example, S can be reshaped into a 2D image. S can also be upsampled to the original frame I (700) resolution to undo the downsampling described in step 602 above.

[0122] In step 632, the algorithm 600 generates an overall estimated scatter image 1600, shown in Figure 16. Once the model is constructed, in this case the kernel amplitude x̂ is estimated, and the model is then applied to the primary image C to obtain the estimated scatter image Ŝ.

[0123] In step 634, algorithm 600 uses scatter image 600 to modify the original input image 700. One way scatter image 600 can be used in this step is to remove the scatter in the image by subtracting scatter image 600 from image 700. Other ways of using scatter image 600 to address the scatter in image 700 are also possible. For example, a selected portion of scatter image 600 may be subtracted from image 700. The selected portion may correspond to an image area with particularly high scatter.

[0124] Figure 17 shows a measured scatter image 1700 that can be directly compared to the scatter estimate image 1600 of Figure 16. As can be seen by comparison, the scatter estimate 1600 shows a qualitative agreement with the measured scatter 1700.

[0125] A generalized operator representation of the mathematics of method 600 in a particular variation is given below:

[0126] In step 632, the scattered image component I S is a one-dimensional or two-dimensional primary x-ray projection I after radiating through the object of interest (e.g., a patient, a phantom, or other object to be imaged). P can be approximated by an operator F acting on

number

[0127] Measurement image by X-ray detector I M is the primary (C) image component (I P ) and secondary (A / B) image components (I S ) and I M may be represented, for example, by image I above or image 700 shown in Figure 7. The secondary image component (A / B) may include scattered x-rays and off-focal source radiation. The measurement image is the sum of its primary and secondary components.

number

[0128] Measurement image I M can include one or more regions blocked by a beam blocking device (e.g., collimator 320) on the source side 306 of the object. These regions are

number

number

number

[0129] Estimated scattering signal components

number

number

[0130] Estimated primary image

number

number

number

[0131] For large scatter to primary signal ratios (SPR), the method 600:

number

[0132] The modeling operator can consider parametric degrees of freedom, shown here as the vectorized unknowns x. Then, fitting the measured image data to the modeled scattering data allows these parameters to be calculated using the beam blocking region.

number

number

number

number

[0133] In one variant of the operator F, the set of scattering kernels includes a finite number of Gaussian generating functions, each Gaussian generating function having a variance

number

number

[0134] Within the above framework (i.e., method 600 and equations 4-12), multiple variations of operator F are possible. For example, in operator F, the Gaussian kernel may be replaced with any other generating filter type. Operator F may include an image transformation to an orthonormal basis, such as a Fourier transform, from which a scatter estimate may be generated from a weighted recombination or finite selection of basis functions. Operator F may include an image transformation to a wavelet basis, such as a continuous wavelet transform (CWT), from which a scatter estimate may be generated from a weighted recombination or finite selection of wavelets. Operator F may also include image decomposition, such as singular value decomposition (SVD) or eigenanalysis, in which a scatter image estimate is formed by reconstruction of a truncated or finite selection of basis components. Input to operator F may include a set of training data or a network trained from the training data, such as a convolutional neural network (CNN). The input training data may include pairs of primary or measurement images including a beam-blocking region and a corresponding scatter image.

[0135] It should be understood that steps 602-634 of algorithm 600 represent a sequential process performed for each image frame I (700) in a series of images acquired by system 300. Once algorithm 600 is performed for each image frame I, the image frames can be reconstructed to form a 3D image of the patient's area. Scatter correction can also be beneficial in radiography applications. Algorithm 600 may be performed on multiple image frames simultaneously. In fact, a set of image frames may be run all at once, which may produce a final result more quickly. Furthermore, algorithm 600 may use features of a particular image frame in the analysis of features of other image frames when processing is performed simultaneously and / or in parallel. In other variations, specific parameters for the analysis of one image frame may be stored and used in the analysis of other image frames, even when the analysis of multiple frames is not performed simultaneously or in parallel.

[0136] While the disclosed technology has been shown and described with respect to particular embodiments, variants, or variations, it will be apparent that equivalent alterations and modifications will occur to those skilled in the art upon reading and understanding this specification and the accompanying drawings. In particular, with respect to the various functions performed by the above-described elements (components, assemblies, devices, members, compositions, etc.), the terms used to describe such elements (including references to "means") are intended, unless otherwise specified, to correspond to any element that performs the designated function of the described element (i.e., an element that is functionally equivalent), even if it is not structurally equivalent to the disclosed structure that performs that function in the exemplary embodiment, variant, or variations of the disclosed technology shown herein. Furthermore, while particular features of the disclosed technology may be described above with respect to only one or more of some illustrated embodiments or variants, such features may be combined with one or more other features of the other variants, as may be desirable and advantageous for any given or particular application.

[0137] While the variations described herein relate to the systems and methods described above, these variations are intended to be exemplary and are not intended to limit the applicability of these variations to only the descriptions set forth herein. While the invention has been illustrated by a description of its variations and the variations have been described in some detail, it is not the intention of the applicants to restrict or in any way limit the scope of the appended claims to such details. Additional advantages and modifications will be readily apparent to those skilled in the art. Accordingly, the invention in its broader aspects is not limited to the specific details, representative apparatus and methods, and illustrative examples shown and described. Accordingly, departures may be made from such details without departing from the spirit or scope of applicants' general inventive concept.

Claims

1. a radiation source for emitting radiation; a radiation detector arranged to receive radiation emitted by the radiation source and to generate radiation data, the radiation data including a primary component and a secondary component; 1. A data processing system comprising: receiving the radiation data; separating the radiation data into the primary and secondary components; applying an image transform to at least one of the first-order and second-order components using at least one of a generating function and a basis decomposition; constructing a scattering model basis using the image transformation; adjusting parameters in a scattering model to fit scattering in the first component using the scattering model basis; generating an estimated scattering image using the fitted scattering model; and a data processing system configured to modify the radiation data using the scatter image to reduce the scatter in the radiation data, thereby generating a scatter-corrected image; An imaging device comprising:

2. the radiation source comprises a rotating X-ray source emitting a radiation beam; the radiation detector comprises an X-ray detector positioned to receive the radiation from the X-ray source; The device, a beamformer configured to adjust the shape of the radiation beam emitted by the X-ray source such that a primary region of the X-ray detector is directly exposed to the radiation beam and such that at least one shadow region of the X-ray detector is shielded from direct exposure to the radiation beam. The imaging device according to claim 1 .

3. The imaging device of claim 1 or 2, wherein estimating the scatter component of the radiation data is based on measured scatter data in at least one shadow region.

4. 4. The imaging device of claim 1, wherein modifying the radiation data using the scatter image to reduce the scatter in the radiation data comprises subtracting the scatter image from the radiation data.

5. The generating function is narrow angle kernel, Wide-angle kernel, Gaussian kernel, at least one of a filter and a filter bank; Orthonormal basis, Fourier transform, wavelet basis, and 5. The imaging device according to claim 1, further comprising at least one of a continuous wavelet transform.

6. The imaging device of claim 1 , wherein the scattering model comprises a convolutional neural network.

7. The imaging device according to any one of claims 1 to 6, wherein the scattering model comprises singular value decomposition or eigenvalue decomposition.

8. The scattering model is a least squares model, a weighted least squares model, a conjugate gradient least squares model, 8. The imaging device of claim 1, further comprising at least one of a gradient descent method and a non-linear optimization technique.

9. The imaging device according to any one of claims 1 to 8, wherein the scattering model basis includes an orthonormal basis.

10. The imaging device of any preceding claim, wherein the scattering model basis comprises a Gaussian kernel convolved with the primary image.

11. The imaging device of any one of claims 1 to 10, wherein the scattering model basis comprises a primary image convolved with at least one of an asymmetric Gaussian kernel, a symmetric Gaussian kernel, and a Gaussian kernel with non-zero distortion.

12. The imaging apparatus of any preceding claim, wherein separating the radiation data into the primary and secondary components comprises filtering the radiation data.

13. The imaging apparatus of any preceding claim, wherein separating the radiation data into the primary and secondary components comprises segmenting the radiation data.

14. The imaging device of any preceding claim, wherein adjusting parameters in the scattering model to fit scattering comprises iterative analysis of the radiation data.

15. The imaging device of claim 14 , wherein the iterative analysis analyzes a segmented image including primary emission data.

16. The imaging device of any preceding claim, wherein the data processing system is configured to offset the radiation data.

17. The imaging device according to any one of the preceding claims, wherein the data processing system is configured to correct the radiation data.

18. 18. The imaging apparatus of claim 1, wherein the data processing system is configured to at least one of normalize the radiation data and weight one or more portions of the radiation data.

19. The imaging device of claim 16, wherein the correction corresponding to the scatter-corrected image includes adjusting the gain of a portion of the radiation data.

20. 20. The imaging device of any one of claims 1 to 19, further comprising a dynamic positioning of a collimator for manipulating the radiation emitted by the radiation source before it is received by the radiation detector.

21. 1. A method for processing radiation data acquired by a radiation imaging device, comprising: receiving the radiation data; separating the radiation data into primary and secondary components; applying an image transform to at least one of the primary and secondary components using a generating function; constructing a scattering model basis using the transformation; adjusting parameters in a scattering model to fit scattering in the at least one of the first and second order components using the scattering model basis; generating an estimated scattering image using the fitted scattering model; and correcting the radiation data using the scatter image to reduce the scatter in the radiation data, thereby generating a scatter-corrected image.

Images