Systems and methods for deep learning-based image reconstruction

Abstract

The present invention discloses a method and system for deep learning-based image reconstruction. An example method includes receiving a set of imaging projection data; identifying voxels to be reconstructed; receiving a trained regression model; and reconstructing the voxels. The voxels are reconstructed by the following steps: projecting the voxels onto each imaging projection in the set of imaging projections according to an acquisition geometry; extracting neighboring pixels around each projected voxel; feeding the regression model with the extracted neighboring pixel data to generate a reconstructed value for the voxel; and repeating the reconstruction for each voxel to be reconstructed to generate a reconstructed image.

Description

Technical Field

[0001] The present disclosure relates generally to image reconstruction, and more particularly to systems and methods for deep learning based image reconstruction. Background Art

[0002] In recent years, digital breast tomosynthesis (DBT) and contrast-enhanced digital breast tomosynthesis (CE-DBT) have proven to be effective cancer detection techniques. DBT uses x-rays to produce three-dimensional (3D) images of the breasts. By taking multiple x-ray pictures of each breast from many angles, a computer can generate the 3D images used to detect any abnormalities. An important part of the DBT / CE-DBT process is image reconstruction because it directly affects what data the radiologist will look at to determine any diagnosis. To reconstruct the image, algorithms are trained and used to reduce noise and any streaking lines. Despite the complexity of the algorithm, the DBT process typically produces an incomplete image reconstruction. Summary of the invention

[0003] Certain examples provide methods and systems for deep learning based image reconstruction.

[0004] An example method disclosed in the present invention includes receiving a set of imaging projection data; identifying voxels to be reconstructed; and receiving a trained regression model. The example method also includes reconstructing the voxels by the following steps: projecting the voxels onto each imaging projection in the set of projections according to an acquisition geometry; extracting neighboring pixels around the voxel of each projection; feeding the regression model with the extracted neighboring pixel data to generate reconstruction values ​​of the reconstructed model; and repeating the reconstruction for each voxel to be reconstructed to generate a reconstructed image.

[0005] The example system disclosed in the present invention includes: a regression model trainer, which trains a regression model; a voxel identifier, which identifies voxels to be reconstructed; an imaging projection data receiver, which receives a set of imaging projections; and a voxel reconstructor. The voxel reconstructor includes: a voxel projector, which projects the voxel onto each imaging projection in the set of imaging projections according to an acquisition geometry; a neighboring pixel extractor, which extracts neighboring pixels around the voxel of each projection; and a regression model supplier, which supplies the regression model with the extracted neighboring pixel data to generate a reconstructed value of the voxel.

[0006] An example non-transitory computer-readable storage medium disclosed in the present invention includes instructions that, when executed, cause a machine to at least: receive a set of imaging projection data; identify voxels to be reconstructed; receive a trained regression model; and reconstruct the voxels by: projecting the voxels onto each imaging projection in the set of imaging projections according to an acquisition geometry; extracting neighboring pixels around the voxel in each projection; supplying the regression model with the extracted neighboring pixel data to generate a reconstructed value for the voxel; and repeating the reconstruction for each voxel to be reconstructed to generate a reconstructed image.

[0007] An exemplary method disclosed in the present invention includes: receiving a set of imaging projection data; identifying pixels to be reconstructed; receiving a trained regression model; and reconstructing the pixels. Reconstructing the pixels includes: mapping the pixels to each imaging projection in the set of projections according to an acquisition geometry; extracting neighboring pixels around each mapped pixel; feeding the regression model with the extracted neighboring pixel data to generate a reconstructed value for the pixel; and repeating the reconstruction for each pixel to be reconstructed.

[0008] The example system disclosed in the present invention includes: a regression model trainer, which trains a regression model; a pixel identifier, which identifies pixels to be reconstructed; an imaging projection data receiver, which receives a set of imaging projections; and a pixel reconstructor. The pixel reconstructor includes: a pixel mapper, which maps the pixel to each imaging projection in the set of imaging projections according to an acquisition geometry; a neighboring pixel extractor, which extracts neighboring pixels around each mapped pixel; and a regression model supplier, which supplies the regression model with the extracted neighboring pixel data to generate a reconstructed value of the pixel.

[0009] An example non-transitory computer-readable storage medium includes instructions that, when executed, cause a machine to at least: receive a set of imaging projection data; identify pixels to be reconstructed; receive a trained regression model; and reconstruct the pixels. Reconstructing the pixels includes: mapping the pixels onto each imaging projection in the set of imaging projections according to an acquisition geometry; extracting neighboring pixels around each mapped pixel; feeding the regression model with the extracted neighboring pixel data to produce a reconstructed value for the pixel; and repeating the reconstruction for each pixel to be reconstructed.

[0010] An example method disclosed in the present invention includes: receiving a volume; identifying pixels to be reconstructed; receiving a trained regression model and reconstructing the pixels. Reconstructing the pixels includes: mapping the pixels to voxels from the volume according to an acquisition geometry; extracting neighboring voxels around each mapped pixel; feeding the regression model with the extracted neighboring voxel data to produce a reconstructed value for the pixel; and repeating the reconstruction for each pixel to be reconstructed.

[0011] An example system disclosed in the present invention includes: a regression model trainer that trains a regression model; a pixel identifier that identifies pixels to be reconstructed; a volume receiver that receives a volume; and a pixel reconstructor. The pixel reconstructor includes: a pixel mapper that maps pixels to voxels from the volume according to an acquisition geometry; a neighboring pixel extractor that extracts neighboring voxels around each mapped pixel; and a regression model feeder that feeds the regression model with the extracted neighboring voxel data to produce a reconstructed value for the pixel.

[0012] An example non-transitory computer-readable storage medium includes instructions that, when executed, cause a machine to at least: receive a volume; identify pixels to be reconstructed; receive a trained regression model; and reconstruct the pixels. Reconstructing the pixels includes: mapping the pixels onto voxels from the volume according to an acquisition geometry; extracting neighboring voxels around each mapped pixel; feeding the regression model with the extracted neighboring voxel data to produce a reconstructed value for the pixel; and repeating the reconstruction for each pixel to be reconstructed.

[0013] An example method disclosed in the present invention includes: receiving a set of imaging projection data; receiving a volume; identifying a pixel to be reconstructed; receiving a trained regression model; and reconstructing the pixel. Reconstructing the pixel includes: mapping the pixel to each imaging projection in the set of projections according to an acquisition geometry; mapping the pixel to a voxel from the volume according to the acquisition geometry; extracting neighboring pixels around each mapped pixel in the projection; extracting neighboring voxels around each mapped pixel in the volume; feeding the regression model with the extracted neighboring pixel data and the extracted neighboring voxel data to produce a reconstructed value for the pixel; and repeating the reconstruction for each pixel to be reconstructed.

[0014] An example system disclosed in the present invention includes: a regression model trainer, which trains a regression model; a pixel identifier, which identifies pixels to be reconstructed; an imaging projection data receiver, which receives a set of imaging projections; a volume receiver, which receives a volume; and a pixel reconstructor. The pixel reconstructor includes: a pixel mapper, which maps the pixel to each imaging projection in the set of imaging projections according to an acquisition geometry, and maps the pixel to a voxel from the volume according to the acquisition geometry; a neighboring pixel extractor, which extracts neighboring pixels around each mapped pixel in the imaging projection; a neighboring voxel extractor, which extracts neighboring voxels around the mapped pixel in the volume; and a regression model supplier, which supplies the regression model with the extracted neighboring pixel data and the extracted neighboring voxel data to produce a reconstructed value for the pixel.

[0015] An example non-transitory computer-readable storage medium includes instructions that, when executed, cause a machine to at least: receive a set of imaging projection data; receive a volume; identify a pixel to be reconstructed; receive a trained regression model; and reconstruct the pixel. Reconstructing the pixel includes: mapping the pixel onto each imaging projection in the set of imaging projections according to an acquisition geometry; mapping the pixel onto a voxel from the volume according to the acquisition geometry; extracting neighboring pixels around each mapped pixel in the projection; extracting neighboring voxels around each mapped pixel in the volume; feeding the regression model with the extracted neighboring pixel data to produce a reconstructed value for the pixel; and repeating the reconstruction for each pixel to be reconstructed.

[0016] An example method disclosed in the present invention includes: receiving a volume; receiving an orientation parameter and a thickness parameter for a slab to be reconstructed; identifying voxels from the slab to be reconstructed; receiving a trained regression model; and reconstructing the voxels. Reconstructing the voxels includes: mapping voxels from the slab onto voxels from the volume according to the orientation parameter and thickness parameter of the slab; extracting neighboring voxels from the volume surrounding each mapped voxel from the slab; feeding the regression model with neighboring voxels extracted from the volume data to produce a reconstructed value for the voxel from the slab; and repeating the reconstruction for each voxel from the slab to be reconstructed.

[0017] An example system disclosed in the present invention includes: a regression model trainer that trains a regression model; a volume receiver that receives a volume; an orientation parameter receiver that receives an orientation parameter for reconstructing a thick slice; a thickness parameter receiver that receives a thickness parameter for reconstructing a thick slice; a voxel identifier that identifies voxels from a thick slice to be reconstructed; and a voxel reconstructor. The voxel reconstructor includes: a voxel mapper that maps voxels from the thick slice to voxels from the volume based on the orientation parameter and thickness parameter of the thick slice; a neighboring voxel extractor that extracts neighboring voxels from the volume surrounding each mapped voxel from the thick slice; and a regression model supplier that supplies the regression model with neighboring voxels extracted from the volume data to generate reconstructed values ​​of voxels from the thick slice.

[0018] An example non-transitory computer-readable storage medium includes instructions that, when executed, cause a machine to at least: receive a volume; receive an orientation parameter and a thickness parameter for a slab to be reconstructed; identify voxels from the slab to be reconstructed; receive a trained regression model; and reconstruct the voxels. Reconstructing the voxels includes: mapping voxels from the slab onto voxels from the volume according to the orientation parameter and thickness parameter of the slab; extracting neighboring voxels from the volume surrounding each mapped voxel from the slab; feeding the regression model with neighboring voxels extracted from the volume data to produce a reconstructed value for the voxel from the slab; and repeating the reconstruction for each voxel from the slab to be reconstructed. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1A The diagram shows an example diagram of reconstructing a voxel using imaging projection data.

[0020] Figure 1B The diagram shows an example diagram of reconstructing a pixel using imaging projection data.

[0021] Figure 1C The figure shows an example graph of pixels reconstructed using a given volume.

[0022] Figure 2A The diagram shows an example of reconstructing a slab using imaging projection data.

[0023] Figure 2B The figure shows an example of reconstructing a slab using a given volume.

[0024] Figure 3A Is implemented Figures 1A-2B A diagram of a system for reconstructing voxels using imaging projections.

[0025] Figure 3B Is implemented Figures 1A-2B A diagram of a system for reconstructing pixels using imaging projections.

[0026] Figure 3C Is implemented Figures 1A-2B A systematic diagram of how to reconstruct pixels using a given volume.

[0027] Figure 3D Is implemented Figures 1A-2B A diagram of a system for reconstructing pixels using imaging projections and a given volume.

[0028] Figure 3E Is implemented Figures 1A-2B A systematic map of voxels within a slab is reconstructed with a given volume, given orientation parameters, and given thickness parameters.

[0029] Figure 4A Graphics Figure 3A An example implementation of a system diagram of .

[0030] Figure 4B Graphics Figure 3B An example implementation of a system diagram of .

[0031] Figure 4C Graphics Figure 3C An example implementation of a system diagram of .

[0032] Figure 4D Graphics Figure 3D An example implementation of a system diagram of .

[0033] Figure 4E Graphics Figure 3E An example implementation of a system diagram of .

[0034] Figure 5 is a flow chart representing an example method of reconstructing an image.

[0035] Figure 6 is a flow chart representing an example method for selecting a training method to train a regression model.

[0036] Figure 7 is a flow chart representing an example method of using a regression model trained on a database consisting of digital simulation human models.

[0037] Figure 8 is a flow chart representing an example method of using a regression model trained on a database consisting of computed tomography (CT) reconstructed data.

[0038] Fig. 9is a flow chart representing an example method of using a regression model trained on a database consisting of imaging acquisitions that have been reconstructed using a given algorithm.

[0039] Fig.10 is a flowchart representing an example method for deploying a regression model.

[0040] Fig.11 is a flow chart representing an example method of using a regression model for reconstruction.

[0041] Fig.12 is a flow chart representing an example method of providing an output of a reconstructed image.

[0042] Fig.13 is a diagram of a processor that can be used to implement the method of Figures 1-12.

[0043] These figures are not to scale. Rather, in order to make multiple layers and multiple regions clear, the thickness of the layers may be exaggerated in the figures. Where feasible, the same reference numerals will be used throughout the drawings, and the accompanying written description refers to the same or similar parts. As used in this patent, indicating that any component (such as a layer, film, region, or plate) is positioned on (such as located, positioned on, set on, or formed on, etc.) another component in any way means that the referenced component is either in contact with the other component, or the referenced component is on the other component with one or more intermediate components located therebetween. Indicating that any component is in contact with another component means that there are no intermediate components between the two components. DETAILED DESCRIPTION

[0044] In the following detailed description, reference is made to the accompanying drawings which form a part of this paper, and in the accompanying drawings, specific embodiments that can be practiced are shown by way of illustration. These examples are described in sufficient detail so that those skilled in the art can practice the subject matter, and it is understood that other examples can be used. Therefore, the following detailed description is provided to describe exemplary embodiments, and the following detailed description should not be considered as limiting the scope of the subject matter described in this disclosure. Certain features from the different aspects described below can be combined to form other new aspects of the subject matter discussed below.

[0045] When introducing elements of various embodiments of the present invention, the articles “a,” “an,” and “said” are intended to mean that there are one or more of the elements. The terms “comprising,” “including,” and “having” are intended to be inclusive and mean that there may be additional elements other than the listed elements.

[0046] As used in this specification, the terms "system", "unit" or "module" and the like may include hardware and / or software systems used to perform one or more functions. For example, a module, unit or system may include a computer processor, a controller and / or other logic-based device that performs operations based on instructions stored on a tangible and non-transitory computer-readable storage medium such as a computer memory. Alternatively, a module, unit or system may include a hard-wired device that performs operations based on the hard-wired logic of the device. The various modules, units, engines and / or systems shown in the accompanying drawings may represent hardware that operates according to software or hard-wired instructions, software that directs the hardware to perform operations, or a combination thereof.

[0047] As used in the present invention, the term "mapping" refers to converting a location located in an imaged object into a corresponding location in one or more images of the object obtained. Alternatively or in addition, mapping can refer to the correlation between common points in multiple images or views, so that points in a first image are mapped to the same points in other related images, so that their location coordinates are correlated when forming a composite two-dimensional image, three-dimensional volume, etc. For example, each element (e.g., pixel, voxel, etc.) in a 3D object has a location on a coordinate system. Mapping elements in a 3D object refers to converting data points from the 3D object into corresponding data points in the generated 2D or 3D image.

[0048] As used in the present invention, the term "projection" or "projection image" refers to an image obtained from the emission of x-rays from a specific angle or view. Projection can be considered a specific example of mapping, in which a set of projection images are captured from different angles of a 3D object and mapped or combined / fused to reconstruct the volume and / or create a synthetic 2D image. Each projection image is captured relative to a central projection (e.g., base projection, direct projection, zero-angle projection, etc.). The image resulting from the projection is either a 3D reconstructed image that is approximately the same as the original 3D object or a synthetic 2D image that merges each projection together and benefits from the information in each view.

[0049] As used in this disclosure, the term "acquisition geometry" is the specific path or motion of an x-ray source with respect to a 3D object (eg, a detector) to obtain a series of 2D projections.

[0050] Although certain examples are described below in the context of a medical or healthcare workplace, other examples may be implemented outside of a medical setting.

[0051] In many different applications, deep learning techniques have utilized learning methods that allow a machine to be given raw data and determine the representations needed to classify the data. Deep learning uses a back-propagation algorithm to find out the structure in the data set, which is used to change the internal parameters of the deep learning machine (such as node weights). Deep learning machines can use a variety of multi-layer architectures and algorithms. While machine learning, for example, involves identifying features to use when training a network, deep learning processes raw data to identify features of interest without external identification.

[0052] Deep learning in a neural network environment involves many interconnected nodes called neurons. Input neurons are stimulated by external sources, and based on connections to other neurons governed by machine parameters, the input neurons stimulate those other neurons. A neural network behaves in a specific way based on its own parameters. Learning improves the machine parameters, and by extension, the connections between neurons in the network, so that the neural network behaves in the desired way.

[0053] Deep learning operates on the premise that many datasets consist of high-level features that consist of lower-level features. For example, when examining an image, rather than looking for objects, it is more effective to look for edges that form patterns, which form parts, which in turn form the objects being looked for. These architectures of features can be found in many different forms of data, such as speech and text.

[0054] An example use of deep learning techniques in the medical field is mammography. Mammography is used to screen for breast cancer and other abnormalities. Traditionally, mammography has been formed on x-ray film. However, recently, flat-panel digital imagers have been introduced that collect mammography in digital form, thereby facilitating the analysis and storage of the collected images. In addition, a lot of attention and technological development has been focused on obtaining three-dimensional images of the breast. Three-dimensional (3D) mammography is also called digital breast tomosynthesis (DBT). Two-dimensional (2D) mammography is full-field digital mammography, and by combining individual enhanced slices (e.g., 1 mm, 2 mm, etc.) of the DBT volume, the synthesized 2D mammography produces a 2D picture derived from the 3D data. The breast tomosynthesis system reconstructs a 3D image volume from a series of two-dimensional (2D) projection images, each of which is obtained at a different angular displacement of the x-ray source. The reconstructed 3D image volume is typically presented as multiple slices of image data, which are geometrically reconstructed in planes parallel to the imaging detector.

[0055] Example system association method

[0056] Figure 1AAn example diagram 100 is illustrated, which represents the reconstruction of a voxel 102 using imaging projection data. The voxel 102 within a 3D space 104 is reconstructed. The voxel 102 is projected onto a first 2D projection view 106, a second 2D projection view 108, and a third 2D projection view 110, each 2D projection representing a slice 112 of the voxel 102. Each 2D projection view 106, 108, 110 has a unique image value and a position indicator. For example, the first 2D projection view 106 may have a unique image value of six and represent the top of the voxel 102. In the same example, the second 2D projection view 108 may have a unique image value of nine and represent the bottom of the voxel 102. Still in the same example, the third 2D projection view 110 may have a unique image value of five and represent the side of the voxel 102. In this example, there are three 2D projection views. However, in other examples, there may be any number of 2D projection views, each representing an image slice 112 of the volume. The 2D projection views 106, 108, 110 provide input for a regression model 114. This input for the regression model 114 is used to reconstruct 116 the voxels 102 into approximately identical voxels. The process described in FIG. 100 is repeated until each voxel 102 within the 3D space 104 has been reconstructed. In this example, voxels are being reconstructed. However, in other examples, any graphical element may be reconstructed (e.g., pixels, thick slices, etc.).

[0057] Figure 1B An example diagram 120 is shown, which illustrates the reconstruction of pixel 122 within image 124 using image projection data. Figure 1A The voxels 102 are projected onto imaging projections, but in this example, pixels 122 are mapped 126 onto each imaging projection 128 in a set of projections according to the acquisition geometry. The mapping 126 of the pixels 122 produces approximately the same pixels 130 on the imaging projections 128.

[0058] Figure 1C An example graph 140 is illustrated showing reconstruction of pixels 142 within an image 144 using a given volume 146. In this example, pixels 142 are mapped 148 onto voxels 150 from volume 146 according to acquisition geometry. For example, the view of pixels 142 in 2D image 144 is extrapolated to provide a representation of voxel 150 in volume 146.

[0059] Figure 2AAn example diagram 200 is illustrated, which shows reconstruction of a slab 202 using imaging projection data. A slab 202 consisting of several voxels 204 within a 3D space 206 is to be reconstructed. In this illustrated example, the slab 202 is projected onto a first 2D projection view 208, a second 2D projection view 210, and a third 2D projection view 212, each 2D projection representing a slice 214 of each voxel 204 in the slab 202. The first 2D projection view 208, the second 2D projection view 210, and the third 2D projection view 212 each have a unique image value and a projection image value. The projection image value 214 of the second 2D projection view 210 is the same as the unique image value 216 because the second 2D projection view 210 contains a base image value 218. In this example, the second 2D projection view 210 contains the base image value 218. However, in other examples, any of the 2D projection views may contain a base image value. The projected image value 220 of the first 2D projection view 208 represents the distance of the unique image value 222 of the first 2D projection view 208 from the base image value 216. Similarly, the projected image value 224 of the third 2D projection view 212 represents the distance of the unique image value 226 of the third 2D projection view 212 from the base image value 216. For example, if the base image value 216 of the second 2D projection is 8; the unique image value 222 of the first 2D projection 208 is 10; and the unique image value 226 of the third 2D projection 212 is 6, then the projected image value 220 of the first 2D projection 208 is +2 and the projected image value 224 of the third 2D projection 212 is -2. In this example, the unique image values ​​are relatively small, however, in other examples, they may be much longer, which increases computational time and complexity.

[0060] Figure 2A Also illustrated in FIG. 2 is a base pixel 228 located in the 2D projection view and containing the base image value 218. In this example, the base pixel 228 is located in the second 2D projection view 210. Figure 2A Also included are lines 230, 232 on each 2D projection view corresponding to other views of the base pixel 228 (e.g., projection images acquired from different angles of the detector from the X-ray source). In this example, the lines 230, 232 are located on the first 2D projection view 208 and the third 2D projection view 212. The lines 230, 232 correspond to the base pixel 228 in the second 2D projection view 210 from different angles. Thus, for example, the pixel 228 may appear as a point in the first view 210, and appear as a line 230, 232 or a value with depth in the other views 208, 212. These locations can be mapped so that the correspondence is determined when generating a synthetic 2D image, a 3D reconstructed image, etc.

[0061] Moreover, in this example, there are only three 2D projection views, whereas in other examples, there may be many 2D projection views based on how many voxels 204 there are in the slab 202, each 2D projection view corresponding to a slice 214 of the voxel 204. This also increases computational time and complexity. By using projection image values ​​instead of unique image values ​​for each 2D projection view, computational time and complexity are reduced. The amount of memory required to store the image values ​​is also reduced.

[0062] Figure 2A 200 also illustrates that the 2D projection views 208, 210, 212 provide input to a regression model 234. The input to the regression model 234 is then used to reconstruct 236 each voxel 204 within the slab 202 into approximately the same voxel, resulting in approximately the same slab. The process described in FIG. 200 is repeated until each voxel 204 within the slab 202 has been reconstructed. In this example, the reconstructed slab 202 is composed of voxels. However, in other examples, the slab can be composed of any type of graphical elements (e.g., pixels, etc.).

[0063] Figure 2B An example diagram is shown that shows a slab 252 within a 3D space 254 being reconstructed using a given volume 256. Figure 2A The mid-slab 202 is projected onto a set of imaging projections, but in this example, the slab 252 is mapped 258 onto the volume 256 , producing an approximately identical slab 260 .

[0064] For example, a 3D geometry may be projected onto a 2D surface at a particular viewpoint. Thus, a 3D volume may be represented using a set of 2D views from image projections at different angles of the original geometry. A 2D projection may include lines corresponding to pixels in other 2D projections.

[0065] In some examples, a point may be selected and / or determined in the view of the pixel to be generated that is mapped to a set of pixels in each associated individual view. For example, a point in the target image view may correspond to a line in the adjacent image projection (see, e.g. Figure 2A ,exist Figure 2A In the projection image, the points in the center image are pixels that are mapped to lines in the two other projection views). For example, in the projection image, pixels are mapped to voxels, and projections are generated using voxels.

[0066] Thus, image data about an object (e.g., a breast, etc.) can be used to determine, starting with a voxel element of a projection, where the voxel is mapped to another projection. In some examples, starting with a pixel in an image projection, the voxel from which the pixel can originate (e.g., from a projection to a volume) is determined. In some examples, starting with a pixel in a first projection, the corresponding pixel in the other projections can be identified. In some examples, the voxel can be mapped to a voxel in an adjacent slice to calculate a thick slice. Thus, a volume constructed in a 3D image and / or a synthesized 2D image can be created from the projection views and the correlation between pixels, voxels, etc.

[0067] Figures 3A-3E is to reconstruct e.g. Figures 1A-2B 300 of an example system for an example image shown in FIG. The example system 300 includes a reconstruction controller 301. The reconstruction controller 301 initiates a reconstruction process, such as the example process described in more detail below. The reconstruction controller 301 may also decide which method is best for training a regression model at a training model 302, which will be described in detail in FIG. Figures 4A-4E The example system 300 also includes a deployed model 304 within a reconstructor 306. The deployed model 304 is a regression model that has been trained at the training model 302. Figures 4A-4E , the reconstructor 306 is described in more detail. Figures 3A-3E Also included is a computer workstation 308 that includes a reconstructor 306 and a deployed model 304. In this example, the training and reconstruction of the regression model occur at two different locations. However, in other examples, the training and reconstruction of the regression model may occur at the same location. The computer workstation 308 receives the deployed model 304 from the training model block 302. The computer workstation 308 also receives pixels, voxels, and / or slabs to be reconstructed and data used to reconstruct the pixels, voxels, and / or slabs. Figure 3A In the example of FIG. 3 , a computer workstation 308 receives imaging projections 310 to reconstruct voxels 312. Figure 3B In the example of FIG. 3 , computer workstation 308 receives imaging projection 310 to reconstruct pixel 314. Figure 3C In the example of FIG. 3 , computer workstation 308 receives volume 316 to reconstruct pixel 314. Figure 3D In the example of , computer workstation 308 receives both imaging projection 310 and volume 316 to reconstruct pixel 314. Figure 3E In the example of FIG. 3 , computer workstation 308 receives volume 316, orientation parameters 318, and thickness parameters 320 to reconstruct voxels 314 within the slab. Figures 3A-3E are shown separately, but in the specific example, Figures 3A-3E Can be combined and implemented as a single system that accommodates multiple graphic elements, such as voxels, pixels, for 2D and / or 3D image generation.

[0068] Figures 4A-4E A more detailed description of the diagram Figures 3A-3E An example implementation of a system 300 of the invention. The example system 300 includes Figures 3A-3E The example also illustrates a regression model trainer 402, which is located at Figures 3A-3E The training model block 302 of the regression model trainer 402 includes a digital simulation phantom (DAP) modeler 404, a computed tomography (CT) modeler 406, and an algorithm modifier 408. The DAP modeler 404 includes a DAP acquisition simulator 410, which simulates the acquisition based on a digital simulation phantom obtained from a DAP database 412. The simulation phantom acts as a model for complete reconstruction. The DAP modeler also includes a DAP algorithm creator 414, in which the regression model creates a new algorithm that is different from the existing reconstruction algorithm.

[0069] As in Figures 4A-4E As illustrated in FIG. 4 , the CT modeler 406 includes a CT acquisition simulator 416 that uses the CT reconstructed data as a fully reconstructed model to simulate an acquisition. The CT reconstructed data is stored in a CT database 418 within the CT modeler 406. The CT modeler 406 also includes a CT algorithm creator 420 that allows the regression model to create a new algorithm that is different from an existing reconstruction algorithm based on the CT reconstruction data.

[0070] Figures 4A-4E The algorithm modifier 408 includes an algorithm database 422, which includes acquired samples that have been reconstructed using the algorithm in the algorithm reconstructor 424. The reconstruction algorithm manages noise artifacts in the image to obtain high image quality. The regression model trained in the algorithm modifier 408 reduces the computational time and computational complexity of the reconstruction algorithm.

[0071] If the appropriate information is accessible by the corresponding modeler 404, 406, 408, the regression model trainer 402 commands the DAP modeler 404, the CT modeler 406, and the algorithm modifier 408 to perform the corresponding training technique. For example, if the regression model trainer 402 receives a regression model for which only DAP is available to the regression model trainer 402, the DAP modeler 404 performs the regression model training. However, in other examples, more than one of the DAP modeler 404, the CT modeler 406, and the algorithm modifier 408 may have information for a given regression model. In these examples, each of the DAP modeler 404, the CT modeler 406, and the algorithm modifier 408 that has the appropriate information performs the regression model training. In these examples, the reconstruction controller 301 may select a result based on one or more criteria, such as the highest accuracy percentage, the fastest response time, etc.

[0072] exist Figures 4A-4E Also illustrated in FIG. 1 is a type of receiver that receives data for reconstruction together with an identifier that identifies which element is to be reconstructed. Figure 4A An example of includes an imaging projection data receiver 430 that receives a set of imaging projections 310 ( FIG. 3 ) and sends the set of imaging projections 310 to a voxel reconstructor 434 within the reconstructor 306 . Figure 4A Also illustrated in FIG. 4 is a voxel identifier 432 for identifying the voxels 312 ( Figures 3A-3E ) and gives the voxel 312 to the voxel reconstructor 434. Figure 4B An example of includes an imaging projection data receiver 430 that receives a set of imaging projections 310 ( FIG. 3 ) and sends the set of imaging projections 310 to a pixel reconstructor 452 within the reconstructor 306 . Figure 4B Also illustrated in FIG. 4 is a pixel identifier 450 for identifying pixels 314 ( FIG. 3 ) to be reconstructed and providing the pixels 314 to a pixel reconstructor 452 . Figure 4C and Figure 4D An example of includes an imaging projection data receiver 430 that receives a set of imaging projections 310 ( FIG. 3 ) and sends the set of imaging projections 310 to a pixel reconstructor 452 within the reconstructor 306 . Figure 4C and Figure 4D A volume receiver 456 is also included to receive the volume 316 and send it to the pixel reconstructor 452 . Figure 4C and Figure 4D Also illustrated in FIG. 4 is a pixel identifier 450 for identifying the pixel 314 ( Figure 3A ) and gives pixel 314 to pixel reconstructor 452. Figure 4EAn example of includes an orientation parameter receiver 460 that receives the orientation parameter 318 and a thickness parameter receiver 462 that receives the thickness parameter 320. The orientation parameter 318 and the thickness parameter 320 are sent to the voxel reconstructor 434. Figure 4E Also illustrated in FIG. 4 is a voxel identifier 432 for identifying the voxels 312 ( FIG. 3 ) to be reconstructed and for providing the voxels 312 to a voxel reconstructor 434 .

[0073] Figures 4A-4E The reconstructor 306 includes elements that reconstruct a given voxel 312 or pixel 314 . Figure 4A The voxel reconstructor 434 includes: a voxel projector 436, which projects the voxel 312 onto each of a set of 2D projections 208, 210, 212 of Figure 2 according to the acquisition geometry; a neighboring pixel extractor 438 that extracts neighboring pixels around each projected voxel 312; and a regression model supplier 440 that supplies the regression model with the extracted neighboring pixel data. Figure 4B and Figure 4C The pixel reconstructor 452 includes: a pixel mapper 454, which maps the pixel 314 to each imaging projection of a set of projections according to the acquisition geometry; a neighboring pixel extractor 438 that extracts neighboring pixels around each mapped pixel 314; and a regression model supplier 440 that supplies the regression model with the extracted neighboring pixel data. Figure 4D The pixel reconstructor 452 includes: a pixel mapper 454, which maps the pixel 314 to each imaging projection of a set of projections according to the acquisition geometry; a neighboring pixel extractor 438 that extracts neighboring pixels around each mapped pixel 314 in the projection; a neighboring voxel extractor 458 that extracts neighboring voxels around each mapped pixel 314 in the volume; and a regression model supplier 440 that supplies the regression model with the extracted neighboring pixel data and the extracted neighboring voxel data. Figure 4E The voxel reconstructor 434 includes a voxel mapper 464 that maps voxels 312 onto voxels from the volume 316 based on the orientation parameters 318 and thickness parameters 320 of the slab from which the voxels 312 originated; a neighboring voxel extractor 458 that extracts neighboring voxels from around each mapped voxel 312; and a regression model feeder 440 that feeds the regression model with the extracted neighboring voxel data. Figures 4A-4EThe data of the voxel reconstructor 434 or the pixel reconstructor 452 is sent to the reconstruction value generator 442, which generates a reconstruction value (e.g., grayscale value, etc.) of the reconstructed pixel or the reconstructed voxel. The reconstruction value is received by the reconstructed image generator 444, which generates a reconstructed image based on the reconstruction value. The imaging projection data receiver 430, the volume receiver 456, the voxel identifier 432, the pixel identifier 450, the positioning parameter receiver 460, the thickness parameter receiver, the voxel reconstructor 434, the pixel reconstructor 452, the reconstruction value generator 442 and the reconstructed image generator 444 are part of the computer workstation 308 of Figure 3.

[0074] The computer workstation 308 also includes a feedback generator 448. The feedback generator 448 identifies whether a possible error has been generated in the reconstructed image. The reconstructed image generator 444 sends the reconstructed image to the user interface for viewing by the user. For example, if each reconstructed graphical element 312 within the reconstructed image is dark except for one outlier that is light or bright, the difference in color / intensity can be an indication of an error made by the regression model trainer 402 or the reconstructor 306. In this example, the feedback generator 448 communicates with the regression model trainer 402 to select different methods of the DAP modeler 404, the CT modeler 406, and the algorithm modifier 408 to retrain the regression model. For example, if the regression model is trained using the DAP in the DAP modeler 404, and the feedback generator 448 indicates that an error has been generated, the feedback generator 448 communicates to the regression model trainer 402 that the regression model is to be retrained on the CT modeler 406 or the algorithm modifier 408. In this example, the accuracy percentage of the DAP modeler 404 will be reduced. As a result, if the regression model trainer 402 has information on more than one way to train a regression model, the reconstruction controller 301 may be less likely to select the DAP modeler 404 to train the regression model.

[0075] The computer workstation 308 also includes a user interface 449. The reconstructed image generator 444 sends the reconstructed image to the user interface 449 for viewing by the user. In some examples, the reconstructed image may not be available to the user until the feedback generator 448 determines that there are no errors in the reconstructed image. However, in other examples, the user interface 449 may be displayed immediately after the reconstructed image is generated by the reconstructed image generator 444. In these examples, if the feedback generator 448 indicates that an error has been generated, the user interface 449 may display the first reconstructed image until a more accurate second reconstructed image has been generated.

[0076] Despite the implementation Figures 4A-4E An example embodiment of the example system 300 is shown in Figure 5-12 In the figure, but Figures 3A-4EOne or more of the elements, processes and / or devices illustrated in the drawings may be combined, separated, rearranged, omitted, eliminated and / or implemented in any other manner. Figures 4A-4E are shown separately, but in some examples, Figures 4A-4E Can be combined and implemented as a single system that accommodates multiple graphic elements such as voxels, pixels, etc. for 2D and / or 3D image generation. In addition, the following components and / or more specifically Figures 4A-4E The example system 300 may be implemented by hardware, software, firmware and / or any combination of hardware, software and / or firmware, and the components include: an example reconstruction controller 301; an example regression model trainer 402, the example regression model trainer 402 may include an example digital simulation human model modeler 404, an example CT modeler 406 and an example algorithm modifier 408 in some examples, the digital simulation human model modeler 404 may include an example DAP acquisition simulator 410, an example DAP database 412 and an example algorithm creator 414 in some examples, the example CT modeler 406 may include an example CT acquisition simulator 416, an example CT database 418 and an example CT algorithm creator 420 in some examples, the example algorithm modifier 408 may include an example algorithm database 422 and an example acquisition reconstructor 424 in some examples; An example imaging projection data receiver 430; an example volume receiver 456; an example orientation parameter receiver 460; an example thickness parameter receiver 462; an example pixel identifier 450; an example voxel identifier 432; an example reconstructor 306, which in some examples may include an example voxel reconstructor 434, an example pixel reconstructor 452, and an example reconstruction value generator 442, which in some examples may include an example voxel projector 436, an example neighboring pixel extractor 438, an example regression model provider 440, an example voxel mapper 464, and an example neighboring voxel extractor 458, which in some examples may include an example pixel mapper 454, an example neighboring pixel extractor 438, an example regression model provider 440, and an example neighboring voxel extractor 458; an example feedback generator 448. Thus, for example, any one of the following components and / or more specifically Figures 4A-4EThe example system 300 may be implemented by one or more analog or digital circuits, logic circuits, programmable processors, application specific integrated circuits (ASICs), programmable logic devices (PLDs), and / or field programmable logic devices (FPLDs), and the components include: an example reconstruction controller 301; an example regression model trainer 402, which in some examples may include: an example digital simulation human model modeler 404, an example CT modeler 406, and an example algorithm modifier 408, which in some examples may include an example DAP acquisition simulator 410, an example DAP database 412, and an example algorithm creator 414, the example CT modeler 406 in some examples may include an example CT acquisition simulator 416, an example CT database 418, and an example CT algorithm creator 420, the example algorithm modifier 408 in some examples may include an example algorithm database 42 2 and an example acquisition reconstructor 424; an example imaging projection data receiver 430; an example volume receiver 456; an example orientation parameter receiver 460; an example thickness parameter receiver 462; an example pixel identifier 450; an example voxel identifier 432; an example reconstructor 306, which in some examples may include an example voxel reconstructor 434, an example pixel reconstructor 452, an example reconstruction value generator 442 and an example reconstructed image generator 444, which in some examples may include an example voxel projector 436, an example adjacent pixel extractor 438, an example regression model feeder 440, an example voxel mapper 464 and an example adjacent voxel extractor 458, which in some examples may include an example pixel mapper 454, an example adjacent pixel extractor 438, an example regression model feeder 440 and an example adjacent voxel extractor 458; and an example feedback generator 448. When reading any of the device or system claims of this patent covering pure software and / or firmware implementations, at least one of the following components and / or more specifically Figures 4A-4EThe example system 300 is hereby expressly defined as comprising a tangible computer-readable storage device or storage disk storing software and / or firmware, such as a memory, a digital versatile disk (DVD), a compact disk (CD), a Blu-ray disk, etc., and the components include: an example reconstruction controller 301; an example regression model trainer 402, which in some examples may include: an example digital simulation human modeler 404, an example CT modeler 406 and an example algorithm modifier 408, which in some examples may include an example DAP acquisition simulator 410, an example DAP database 412 and an example algorithm creator 414, the example CT modeler 406 in some examples may include an example CT acquisition simulator 416, an example CT database 418 and an example CT algorithm creator 420, the example algorithm modifier 408 in some examples may include an example algorithm database 422 and an example an acquisition reconstructor 424; an example imaging projection data receiver 430; an example volume receiver 456; an example orientation parameter receiver 460; an example thickness parameter receiver 462; an example pixel identifier 450; an example voxel identifier 432; an example reconstructor 306, which in some examples may include an example voxel reconstructor 434, an example pixel reconstructor 452, an example reconstruction value generator 442, and an example reconstructed image generator 444, which in some examples may include an example voxel projector 436, an example neighboring pixel extractor 438, an example regression model provider 440, an example voxel mapper 464, and an example neighboring voxel extractor 458, which in some examples may include an example pixel mapper 454, an example neighboring pixel extractor 438, an example regression model provider 440, and an example neighboring voxel extractor 458; and an example feedback generator 448. Further, in addition to or in lieu of Figures 4A-4E In addition to those components shown in the figure, Figures 4A-4E The example system 300 may include one or more elements, processes, and / or devices, and / or may include more than one of any or all of the illustrated elements, processes, and devices.

[0077] exist Figure 5-12 The implementation is shown in Figures 4A-4E Flowchart of example machine readable instructions of an example system 300. In this example, the machine readable instructions include a program executed by a processor, such as the following with respect to Figure 5-12The program may be implemented in software stored on a tangible computer-readable storage medium such as a CD-ROM, floppy disk, hard drive, digital versatile disk (DVD), Blu-ray disk, or memory associated with the processor 1312, but the entire program and / or portions thereof may alternatively be executed by a device other than the processor 1312 and / or implemented in firmware or dedicated hardware. In addition, although reference is made to Figure 5-12 The flowchart illustrated in describes an example procedure, but many other methods of implementing the example system 300 may alternatively be used. For example, the order of execution of the blocks may be changed and / or some of the blocks described may be changed, eliminated, or combined.

[0078] As mentioned above, the present invention may be implemented using coded instructions (eg, computer and / or machine readable instructions). Figure 5-12 The exemplary process of the present invention is a computer-readable storage medium, wherein the coded instructions are stored on a tangible computer-readable storage medium, such as a hard drive, flash memory, read-only memory (ROM), compact disk (CD), digital versatile disk (DVD), cache, random access memory (RAM), and / or any other storage device or storage disk that can store information for any duration (e.g., an extended period of time, permanent, extremely short-term, temporary cache, and / or cache for information). As used in the present invention, the term tangible computer-readable storage medium is expressly defined to include any type of computer-readable storage device and / or storage disk, excluding propagating signals and excluding transmission media. As used in the present invention, "tangible computer-readable storage medium" and "tangible machine-readable storage medium" are used interchangeably. Additionally or alternatively, the coded instructions (e.g., computer and / or machine-readable instructions) may be used to implement Figure 5-12 An example process of the invention is provided wherein the coded instructions are stored on a non-transitory computer and / or machine readable medium, such as a hard drive, flash memory, read-only memory, compact disk, digital versatile disk, cache, random access memory, and / or any other storage device or storage disk that can store information for any duration (e.g., an extended period of time, permanent, very short time, temporary cache, and / or cache for information). As used in the present invention, the term non-transitory computer readable medium is expressly defined to include any type of computer readable storage device and / or storage disk, excluding propagating signals and excluding transmission media. As used in the present invention, the term "at least" when used as a transition word in the preamble of a claim is open ended in the same manner as the term "including" is open ended.

[0079] Figure 55 is a flow chart representing an example method 500 for generating a reconstructed image. The example method 500 begins at block 502, where the regression model trainer 402 trains a regression model. For example, to reconstruct digital breast tomosynthesis (DBT) projections, the process begins with the regression model trainer 402 training a regression model. At block 504, the regression model trainer 402 ( Figures 4A-4E ) Deploy the regression model. In this example, after the regression model has been trained, the regression model is deployed to the voxel reconstructor 434 or the pixel reconstructor 452. The DBT projection data and the voxels 312 or pixels 314 ( Figures 3A-3E ) is also sent to the voxel reconstructor 434 or the pixel reconstructor 452. At block 506, the voxel reconstructor 434 or the pixel reconstructor 452 uses the regression model for reconstruction. In the illustrated example, the voxel reconstructor 434 or the pixel reconstructor 452 reconstructs the image generated by the voxel identifier 432 or the pixel identifier 450 ( Figures 4A-4E ) sent voxels 312 or pixels 314. At block 508, the reconstruction value generator 442 provides an output of the reconstructed image value. For example, the reconstruction value generator 442 ( Figures 4A-4E ) Generate reconstructed image values ​​of the breast from the DBT projection data. Example implementations of blocks 502, 504, 506, 508 are described in more detail below.

[0080] Figure 6 It means training the regression model ( Figure 5 At block 602, the regression model trainer 402 determines whether a digitally simulated human body model (DAP) is available for the regression model trainer 402. At block 604, if the DAP is available for the regression model trainer 402, the regression model trainer 402 is commanded to train a regression model on the DAP modeler 404 using a database consisting of the DAP. Figure 7 An example of training a regression model on the DAP modeler 404 is further described. In some examples, if the regression model trainer uses the DAP for the regression model, the regression model is trained on the DAP modeler 404. In these examples, the regression model trainer 402 first checks the DAP. However, in other examples, before checking the DAP, the regression model trainer 402 can check CT reconstruction data or imaging acquisitions that have been reconstructed with a given algorithm.

[0081] After instructing the regression model trainer 402 to use the DAP modeler 404, or if the DAP is not available at block 602, the regression model trainer 402 is instructed to determine whether computed tomography (CT) data is available at blocks 606 and 608, respectively. If CT data is available, at block 610, the regression model trainer 402 is instructed to train a regression model on the CT modeler 406 using a database consisting of CT reconstruction data. Figure 8 An example of training a regression model on a database consisting of CT reconstruction data is further described. In some examples, the regression model trainer 402 may enable DAP to be used to train the regression model. In these examples, after the regression model trainer 402 instructs the DAP modeler 404 to train the regression model (block 604), the regression model trainer 402 checks whether the CT has reconstruction data available for training the regression model (block 606). If the CT reconstruction data is available, the regression model is trained on the CT modeler 406 (block 610). In some examples, the regression model trainer 402 may not enable DAP to be used to train the regression model. In these examples, the regression model trainer checks whether there is CT reconstruction data available for training the regression model (block 608). If the CT reconstruction data is available, the regression model is trained on the CT modeler 406 (block 610). In these illustrated examples, the regression model trainer 402 examines the CT reconstruction data (blocks 606, 608) after the regression model trainer 402 examines the DAP (block 602) and before the regression model trainer 402 examines the imaging acquisition that has been reconstructed with the given algorithm (block 612). However, in other examples, the regression model trainer 402 may examine the CT reconstruction data first or last.

[0082] After instructing the regression model trainer 402 to train the regression model using the CT modeler 406 (block 610); or if CT reconstruction data is not available, but DAP is available at block 602, then at block 612, the regression model trainer 402 determines whether the imaging acquisition has been reconstructed using the given algorithm. If an imaging acquisition that has been reconstructed using the given algorithm is available, and / or if neither DAP reconstruction data nor CT reconstruction data is available at blocks 602 and 608, respectively, then at block 614, the regression model trainer 402 is instructed to use a regression model that was trained at the algorithm modifier 408 using a database consisting of imaging acquisitions that have been reconstructed using the given algorithm. Reference Fig. 9Examples of training a regression model on the algorithm modifier 408 are further described. In some examples, the regression model trainer 402 has access to CT reconstruction data. In these examples, after commanding the regression model trainer 402 to train the regression model using the CT modeler 406 (block 610), the regression model trainer 402 determines whether an imaging acquisition that has been reconstructed with a particular algorithm is available (block 612). If an imaging acquisition that has been reconstructed with a particular algorithm is available, the regression model trainer 402 trains the regression model on the algorithm modifier 408 (block 614). In other examples, the regression model trainer 402 does not have access to CT reconstruction data, but has access to DAP. In these examples, after commanding the DAP modeler 404 to train the regression model (block 606), and after confirming that CT reconstruction data is not available, the regression model trainer 402 determines whether an imaging acquisition that has been reconstructed with a given algorithm is available (block 612). If an imaging acquisition that has been reconstructed using a particular algorithm is available, the regression model trainer 402 trains the regression model on the algorithm modifier 408 (block 614). In still other examples, the regression model trainer 402 may not have access to CT reconstruction data or DAP. In these examples, the regression model trainer 402 automatically commands the algorithm modifier 408 to train the regression model (block 614). In these examples, it is assumed that if a regression model is received by the regression model trainer 402, the regression model trainer 402 also receives at least one of the DAP, CT reconstruction data, or an imaging acquisition that has been reconstructed using the given algorithm. However, in other examples, the regression model trainer 402 may not receive any of the DAP, CT reconstruction data, or an imaging acquisition that has been reconstructed using the given algorithm. In these examples, because the regression model cannot be trained, an error message is displayed at Figures 4A-4E on the user interface 449.

[0083] After instructing the regression model trainer 402 to train the regression model on the algorithm modifier 408 (block 614), the process returns to Figure 5 Block 504. Alternatively, if DAP and / or CT reconstruction data is available, but imaging acquisitions that have been reconstructed using a given algorithm are not available, the process returns to Figure 5 For example, if the regression model trainer 402 instructs the algorithm modifier 408 to train the regression model, the process returns to Figure 5 In these examples, if the regression model trainer 402 instructs the DAP modeler 404 and / or the CT modeler 406 to train a regression model, but an imaging acquisition that has been reconstructed using a given algorithm is not available to the regression model trainer 402, the program returns to Figure 5 Block 504.

[0084] Figure 7 A flowchart representing an example implementation using a regression model trained with a database consisting of DAPs used to simulate imaging acquisitions. Figure 6 At block 702, the regression model trainer 402 ( Figures 4A-4E ) accesses the digital form and several alternative forms of the image. For example, the regression model trainer 402 accesses the digital 3D breast that needs to be reconstructed. In some examples, the reconstruction controller 301, the voxel reconstructor 434, or the pixel reconstructor 452 gives the acquisition of the image to the regression model trainer 402. However, in other examples, the regression model trainer 402 requests the acquisition of the image from the reconstruction controller 301, the voxel reconstructor 434, or the pixel reconstructor 452. At block 704, the acquisition simulator 410 ( Figures 4A-4E ) simulates the image while maintaining access to the original 3D image. In this example, the regression model trainer 402 maintains access to the original 3D image of the breast and also simulates the image on the DAP acquisition simulator 410. At block 706, the DAP acquisition simulator 410 performs a simulation of the imaging acquisition from the 3D digital image and uses the data as "significant evidence". In the illustrated example, the simulation of the original 3D image of the breast is used as significant evidence. At block 708, the regression model trainer 402 performs regression by training the model to obtain the original 3D human model from the simulated imaging acquisition. For example, the regression model trainer 402 performs regression to obtain the original 3D human model from the simulated DBT acquisition. At block 710, the DAP algorithm creator 414 designs an algorithm to derive the most likely mapping function from the image acquisition to the 3D reconstructed image. In this example, the most likely mapping from the DBT acquisition to the original 3D human model is used when designing the new algorithm. At block 712, the DAP modeler 404 outputs a regression model based on the designed algorithm. For example, a regression model trained based on the algorithm designed from block 710 is used to reconstruct the voxels 312 or pixels 314 in the voxel reconstructor 434 or the pixel reconstructor 452 ( Figures 3A-3E ).

[0085] Figure 8 A flowchart representing an example implementation using a regression model trained on a database of CT reconstructed data from simulated imaging acquisitions. Figure 6 At block 802, the regression model trainer 402 ( Figures 4A-4E) accesses the collection of images for which a computed tomography (CT) scan has been performed. For example, the regression model trainer 402 accesses a digital 3D breast that needs to be reconstructed. In some examples, the reconstruction controller 301, the voxel reconstructor 434, or the pixel reconstructor 452 gives the collection of images to the regression model trainer 402. However, in other examples, the regression model trainer 402 requests the collection of images from the reconstruction controller 301, the voxel reconstructor 434, or the pixel reconstructor 452. At block 804, the CT acquisition simulator 416 within the CT modeler 406 simulates the image while maintaining access to the original image in the CT scan. In this example, the regression model trainer 402 maintains access to the original image of the 3D breast and also simulates the image on the CT acquisition simulator 416. At block 806, the CT acquisition simulator 416 performs simulations on the imaging acquisition from the CT scan and uses the data as "significant evidence". In the illustrated example, the simulation of the original 3D breast is used as significant evidence. At block 808, the regression model trainer 402 performs regression by training the model to obtain the original CT scan from the simulated imaging acquisition. For example, the regression model trainer 402 performs regression to obtain the original CT scan from the simulated DBT acquisition. At block 810, the CT algorithm creator 420 within the CT modeler 406 designs an algorithm to derive the most likely mapping function from the imaging acquisition to the 3D reconstructed image. In this example, the most likely mapping from the DBT acquisition to the original CT scan is used to design a new algorithm. At block 812, the CT modeler 406 outputs a regression model based on the designed algorithm of block 810. For example, the voxel 312 or pixel 314 ( Figures 3A-3E ).

[0086] Fig. 9 Yes means Figure 6 Detailed breakdown of the example of block 614 of FIG. 614, the reconstruction controller 301 decides to use the regression model trained on the algorithm modifier 408. At block 902, the regression model trainer 402 ( Figures 4A-4E) accesses an existing imaging algorithm from the algorithm database 422, which manages noise artifacts in the image to obtain high image quality. For example, the regression model trainer 402 accesses a DBT algorithm (e.g., ASiR, MBIR, etc.). At block 904, the acquisition reconstructor 424 within the algorithm modifier 408 estimates the algorithm reconstructed with the neural network to reduce the computational complexity of the original algorithm, thereby producing the same image faster. In this example, the neural network reduces the computational power of the DBT algorithm. At block 906, the regression model trainer 402 trains the regression model using the new algorithm to reduce computing time and computational complexity. In the illustrated example, the neural network reduces the computational power of the DBT algorithm. As a result, the same reconstructed image is obtained more quickly. At block 908, the algorithm modifier 408 outputs a regression model based on the reconstructed algorithm. For example, the voxel 312 or pixel 314 in the voxel reconstructor 434 or the pixel reconstructor 452 is reconstructed using a regression model trained based on the reconstruction algorithm ( Figures 3A-3E ).

[0087] Fig.10 is the deployment regression model ( Figure 5 At block 1002, the voxel reconstructor 434 or the pixel reconstructor 452 ( Figures 4A-4E ) is given a set of data from the imaging projection data receiver 430, the volume receiver 456, the orientation parameter receiver 460 and / or the thickness parameter receiver 462 and the voxels 312 or pixels 314 reconstructed from the voxel identifier 432 or the pixel identifier 450 ( Figures 3A-3E ). For example, the voxel reconstructor 434 may be given a set of DBT projection data from the imaging projection data receiver 430 and the voxels reconstructed from the voxel identifier 432. At block 1004, the reconstruction controller 301 selects which regression model to use. In this example, the reconstruction processor 301 decides to use a regression model trained on the DAP modeler 404, the CT modeler 406, and / or the algorithm modifier 408. At block 1006, the voxel projector 436 within the voxel reconstructor 434 or the pixel mapper 454 within the pixel reconstructor 452 uses the regression model to map a set of pixel values ​​to voxels or another pixel value. Figures 1A-2B The process of mapping a set of pixel values ​​to voxels or another pixel value has been described in detail in . In the illustrated example, a set of pixel values ​​is mapped to voxel values ​​using a regression model trained in regression model trainer 402. In other examples, a set of pixel values ​​can be mapped to another pixel value to create a synthetic 2D reconstruction.

[0088] Fig.11 It means using regression model for reconstruction ( Figure 5506 of the embodiment of the present invention). At block 1102, the reconstructor 306 determines whether an imaging projection has been received. If so, at block 1104, the voxel projector 436, the pixel mapper 454, or the voxel mapper 464 reconstructs the voxel 312 or pixel 314 ( Figures 3A-3E ) onto each imaging projection. For example, based on the original breast acquisition geometry, voxels can be projected onto each 2D DBT projection. This process was previously described with Figures 1A-2B Described in association. At block 1106, the neighboring pixel extractor 438 and / or the neighboring voxel extractor 458 extracts neighboring pixels and / or neighboring voxels around each projected pixel or voxel. In this example, the neighboring pixels around each projected pixel in the voxel are extracted. At block 1108, the reconstructor 306 determines whether there is another voxel 312 or pixel 314 to be reconstructed. If yes, the process repeats starting at block 1102. In the illustrated example, if there are multiple voxels that need to be reconstructed, the process restarts at block 1102. If not, or if no imaging projection has been received, the reconstructor 306 determines whether a volume has been received at block 1110. If received, then at block 1112, based on the acquisition geometry, the voxel projector 436, the pixel mapper 454, or the voxel mapper 464 will be reconstructed The voxel 312 or pixel 314 ( Figures 3A-3E ) is projected or mapped onto each imaging projection. At block 1114, the neighboring pixel extractor 438 and / or neighboring voxel extractor 458 extracts neighboring voxels and / or neighboring pixels around each mapped pixel, voxel, or slab. At block 1116, the reconstructor 306 determines whether there is another voxel 312 or pixel 314 to be reconstructed. If yes, the process repeats starting at block 1110. If no, the process is complete and returns to Figure 5 Block 508 of .

[0089] Fig.12 is the output that provides the reconstructed image ( Figure 5 At block 1202, the voxel reconstructor 434 or the pixel reconstructor 452 reconstructs all voxels 312 and / or pixels 314 ( Figures 3A-3E) for reconstruction. For example, if there are 3 voxels that need to be reconstructed in the DBT projection data simulation, the voxel reconstructor 434 reconstructs all 3 voxels. At block 1204, the regression model supplier 440 supplies the regression model with the extracted reconstruction data to generate the reconstruction values ​​of the pixels and / or voxels at the reconstruction value generator 442. For example, the regression model supplier 440 can supply the regression model with the extracted reconstruction data through an artificial neural network (e.g., a convolutional neural network (CNN), a recurrent neural network (RNN), a feedforward neural network, etc.), a regression support vector machine (SVM), etc. In the illustrated example, the regression model is supplied with the extracted voxel reconstruction data to generate the reconstruction values ​​of the voxels in the model of the breast. At block 1206, the reconstructed image is generated by the example reconstructed image generator 444 using the reconstruction values ​​of the voxels and / or pixels (e.g., grayscale values, etc.). In this example, a 3D reconstructed image of the breast is formed by using the reconstruction values ​​of the voxels and / or pixels.

[0090] Fig.13 Is able to execute Figure 5-12 Instructions to achieve Figures 4A-4E The processor platform 1300 can be, for example, a server, a personal computer, a mobile device (e.g., a cell phone, a smart phone, an iPad, etc.). TM tablet computer), personal digital assistant (PDA), Internet tool, DVD player, CD player, digital video recorder, Blu-ray player, game console, personal video recorder, set-top box, or any other type of computing device.

[0091] The processor platform 1300 of the illustrated example includes a processor 1312. The processor 1012 of the illustrated example is hardware. For example, the processor 1312 can be implemented by one or more integrated circuits, logic circuits, microprocessors, or controllers from any desired family or manufacturer.

[0092] The processor 1312 of the illustrated example includes a local memory 1313 (e.g., a cache). The processor 1312 of the illustrated example communicates with a main memory via a bus 1318, and the main memory includes a volatile memory 1314 and a non-volatile memory 1316. The volatile memory 1314 can be implemented by a synchronous dynamic random access memory (SDRAM), a dynamic random access memory (DRAM), a RAMBUS dynamic random access memory (RDRAM), and / or any other type of random access memory device. The non-volatile memory 1316 can be implemented by a flash memory and / or any other desired type of memory device. Access to the main memory 1314, 1316 is controlled by a memory controller.

[0093] The processor platform 1300 of the illustrated example also includes an interface circuit 1320. The interface circuit 1320 may be implemented by any type of interface standard, such as an Ethernet interface, a Universal Serial Bus (USB), and / or a PCI Express interface.

[0094] In the illustrated example, one or more input devices 1322 are connected to the interface circuit 1320. The input devices 1322 allow a user to enter data and commands into the processor 1012. The input devices may be implemented by, for example, an audio sensor, a microphone, a camera (still or video), a keyboard, buttons, a mouse, a touch screen, a track pad, a track ball, an isopoint, and / or a voice recognition system.

[0095] One or more output devices 1324 are also connected to the interface circuit 1320 of the illustrated example. The output device 1024 may be implemented, for example, by a display device (e.g., a light emitting diode (LED), an organic light emitting diode (OLED), a liquid crystal display, a cathode ray tube display (CRT), a touch screen, a tactile output device, a printer, and / or a speaker). The interface circuit 1320 of the illustrated example thus generally includes a graphics driver card, a graphics driver chip, or a graphics driver processor.

[0096] The illustrated example interface circuitry 1320 also includes communication devices, such as transmitters, receivers, transceivers, modems, and / or network interface cards, to facilitate data exchange with an external machine (e.g., any type of computing device) via a network 1326 (e.g., an Ethernet connection, a digital subscriber line (DSL), a telephone line, a coaxial cable, a cellular telephone system, etc.).

[0097] The processor platform 1300 of the illustrated example also includes one or more mass storage devices 1328 for storing software and / or data. Examples of these mass storage devices 1328 include floppy disk drives, hard disks, optical disks, Blu-ray disks, RAID systems, and digital versatile disk (DVD) drives.

[0098] Figures 4A-4E The coded instructions 1332 may be stored in the mass storage device 1328, in the volatile memory 1314, in the non-volatile memory 1316, and / or on a removable tangible computer storage medium such as a CD or DVD.

[0099] By the foregoing, it is to be appreciated that the methods, apparatuses, and articles of manufacture disclosed above facilitate improved image reconstruction, and generate 2D and / or 3D images from projection data, such as DBT projection data. Certain examples facilitate improved modeling of image information to facilitate generation of synthetic 2D images from available projection information. Certain examples mitigate reliance on specific equations, instead utilizing modeling and learning to generate 2D and / or 3D images from available image projection information. Certain examples facilitate improved application of artificial intelligence techniques to image reconstruction. Certain examples provide technical improvements to processors configured to model, process, and reconstruct image data, such as 2D and / or 3D images generated from image projection data (e.g., DBT, etc.).

[0100] For example, in DBT / CE-DBT, reconstruction directly affects what the radiologist sees in the data and therefore affects the resulting diagnosis. Although today's algorithms tend to optimize the quality of the reconstructed slices (e.g., reduce noise, mitigate streak artifacts, etc.), the prior knowledge introduced in these algorithms typically only partially addresses the deficiencies that lead to incompletely reconstructed data. In addition, these sophisticated algorithms are typically complex and require considerable computing power. As a result, the design of an ideal reconstruction is limited by the proprietary technology of the algorithm designer. However, for a given voxel, the reconstruction algorithm can be viewed as just a mapping function that relates the reconstructed grayscale to a set of input grayscales extracted from the projections.

[0101] Thus, some examples use regression tools to approximate any reconstruction algorithm. Some examples bring technical improvements to processor reconstruction by learning an "ideal" reconstruction that is almost impossible to simulate. Some examples allow for the simplification of the design of reconstruction algorithms. For example, by approximating existing algorithms, computational work can be reduced. Additionally, some examples provide more relevant reconstructed volumes to radiologists by designing / learning algorithms that would otherwise be difficult to simulate (e.g., learning ground truth / perfecting data, etc.).

[0102] Although the present invention has disclosed certain example methods, apparatus, and articles of manufacture, the scope of coverage of this patent is not limited thereto. On the contrary, this patent covers all methods, apparatus, and articles of manufacture that reasonably fall within the scope of the claims of this patent.

Claims

1. A method comprising: receiving a set of imaging projection data; identifying voxels to be reconstructed; Receive the trained regression model; as well as The voxels are reconstructed by the following steps: projecting the voxel onto each imaging projection in the set of projections according to an acquisition geometry; Extract neighboring pixels around each projected voxel; supplying the regression model with the extracted neighboring pixel data to generate a reconstructed value for the voxel; as well as The reconstruction is repeated for each voxel to be reconstructed to produce a reconstructed image, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

2. The method according to claim 1, wherein: The voxels are part of the slab to be reconstructed. 3 . The method according to claim 1 , further comprising generating a reconstructed image using the reconstruction values, and displaying the reconstructed image on a user interface. 4 . The method of claim 1 , further comprising receiving a volume onto which voxels from the volume are projected according to an acquisition geometry.

5. A system comprising: A regression model trainer, wherein the regression model trainer trains the regression model; a voxel identifier that identifies voxels to be reconstructed; an imaging projection data receiver, the imaging projection data receiver receiving a set of imaging projections; as well as A voxel reconstructor, the voxel reconstructor comprising: a voxel projector that projects the voxel onto each imaging projection in the set of imaging projections based on an acquisition geometry; a neighboring pixel extractor that extracts neighboring pixels around each projected voxel; and a regression model feeder that feeds the regression model with the extracted neighboring pixel data to generate a reconstructed value for the voxel, The regression model trainer is configured to train the regression model based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

6. The system according to claim 5, wherein: The regression model trainer comprises: a digitally simulated human body modeler (DAP) comprising an acquisition simulator, an algorithm builder, and a DAP database; a computed tomography (CT) modeler comprising an acquisition simulator, an algorithm builder, and a CT database; and An algorithm modifier includes an acquisition reconstructor and an algorithm database.

7. The system according to claim 5, further comprising a feedback generator, which identifies whether an error is generated on the reconstructed image and transmits it to the regression model trainer to retrain the regression model.

8. The system according to claim 5, wherein: The voxels are part of the slab to be reconstructed.

9. The system according to claim 5, further comprising a reconstruction value generator, which generates a reconstruction value for each reconstructed pixel or voxel, and the reconstruction value is used to generate a reconstructed image.

10. The system of claim 5, further comprising a user interface that displays the reconstructed image.

11. A non-transitory computer-readable storage medium comprising instructions that, when executed, cause a machine to at least: receiving a set of imaging projection data; identifying voxels to be reconstructed; Receive the trained regression model; as well as The voxels are reconstructed by the following steps: projecting the voxel onto each imaging projection in the set of imaging projections according to an acquisition geometry; Extract neighboring pixels around each projected voxel; supplying the regression model with the extracted neighboring pixel data to generate a reconstructed value for the voxel; as well as The reconstruction is repeated for each voxel to be reconstructed to produce a reconstructed image, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

12. The non-transitory computer-readable storage medium of claim 11, further comprising instructions that, when executed, cause a machine to identify errors within the reconstructed image and communicate with the regression model trainer to retrain the regression model.

13. The non-transitory computer-readable storage medium of claim 11, wherein: The voxels are part of the slab to be reconstructed.

14. The non-transitory computer-readable storage medium of claim 11, further comprising instructions that, when executed, cause a machine to generate a reconstructed image on a user interface using the reconstruction values.

15. The non-transitory computer readable storage medium of claim 11, further comprising instructions that, when executed, cause a machine to receive a volume, acquire voxels projected onto voxels from the volume according to a geometry.

16. A method comprising: receiving a set of imaging projection data; identifying pixels to be reconstructed; Receive the trained regression model; as well as The pixel is reconstructed by the following steps: mapping the pixel to each imaging projection in the set of projections according to an acquisition geometry; Extract neighboring pixels around each mapped pixel; Supplying the regression model with the extracted adjacent pixel data to generate a reconstructed value of the pixel; as well as The reconstruction is repeated for each pixel to be reconstructed, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

17. A system comprising: A regression model trainer, wherein the regression model trainer trains the regression model; a pixel identifier that identifies pixels to be reconstructed; an imaging projection data receiver, the imaging projection data receiver receiving a set of imaging projections; as well as A pixel reconstructor, the pixel reconstructor comprising: a pixel mapper that maps the pixels onto each imaging projection in the set of imaging projections according to an acquisition geometry; a neighboring pixel extractor that extracts neighboring pixels around each mapped pixel; and a regression model feeder that feeds the regression model with the extracted neighboring pixel data to generate a reconstructed value for the pixel, The regression model trainer is configured to train the regression model based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

18. The system of claim 17, wherein: The regression model trainer comprises: a digitally simulated human body modeler (DAP) comprising an acquisition simulator, an algorithm builder, and a DAP database; a computed tomography (CT) modeler comprising an acquisition simulator, an algorithm builder, and a CT database; and An algorithm modifier includes an acquisition reconstructor and an algorithm database.

19. The system according to claim 17, further comprising a feedback generator, which identifies whether an error is generated on the reconstructed image and transmits it to the regression model trainer to retrain the regression model.

20. A non-transitory computer-readable storage medium comprising instructions that, when executed, cause a machine to at least: receiving a set of imaging projection data; identifying pixels to be reconstructed; Receive the trained regression model; as well as The pixel is reconstructed by the following steps: mapping the pixel onto each imaging projection in the set of imaging projections according to an acquisition geometry; Extract neighboring pixels around each mapped pixel; Supplying the regression model with the extracted adjacent pixel data to generate a reconstructed value of the pixel; as well as The reconstruction is repeated for each pixel to be reconstructed, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

21. The non-transitory computer-readable storage medium of claim 20, further comprising instructions that, when executed, cause a machine to identify errors within a reconstructed image and communicate with the regression model trainer to retrain the regression model.

22. A method comprising: Receiving capacity; identifying pixels to be reconstructed; Receive the trained regression model; as well as The pixel is reconstructed by the following steps: mapping the pixels onto voxels from the volume according to an acquisition geometry; extracting neighboring voxels around the mapped pixel; supplying the regression model with the extracted neighboring voxel data to generate a reconstructed value for the pixel; as well as The reconstruction is repeated for each pixel to be reconstructed, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

23. A system comprising: A regression model trainer, wherein the regression model trainer trains the regression model; a pixel identifier that identifies pixels to be reconstructed; a volume receiver, the volume receiver receiving a volume; as well as A pixel reconstructor, the pixel reconstructor comprising: a pixel mapper that maps pixels onto voxels from the volume according to an acquisition geometry; a neighboring pixel extractor that extracts neighboring voxels around each mapped pixel; and a regression model feeder that feeds the regression model with the extracted neighboring voxel data to produce a reconstructed value for the pixel, The regression model trainer is configured to train the regression model based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

24. The system of claim 23, wherein: The regression model trainer comprises: a digitally simulated human body modeler (DAP) comprising an acquisition simulator, an algorithm builder, and a DAP database; a computed tomography (CT) modeler comprising an acquisition simulator, an algorithm builder, and a CT database; and An algorithm modifier includes an acquisition reconstructor and an algorithm database.

25. The system according to claim 23, further comprising a feedback generator, which identifies whether an error is generated on the reconstructed image and transmits it to the regression model trainer to retrain the regression model.

26. A non-transitory computer-readable storage medium comprising instructions that, when executed, cause a machine to at least: Receiving capacity; identifying pixels to be reconstructed; Receive the trained regression model; as well as The pixel is reconstructed by the following steps: mapping the pixels onto voxels from the volume according to an acquisition geometry; Extract neighboring voxels around each mapped pixel; supplying the regression model with the extracted neighboring voxel data to generate a reconstructed value for the pixel; as well as The reconstruction is repeated for each pixel to be reconstructed, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

27. The non-transitory computer-readable storage medium of claim 26, further comprising instructions that, when executed, cause a machine to identify errors within a reconstructed image and communicate with the regression model trainer to retrain the regression model.

28. A method comprising: receiving a set of imaging projection data; Receiving capacity; identifying pixels to be reconstructed; Receive the trained regression model; as well as The pixel is reconstructed by the following steps: mapping the pixel to each imaging projection in the set of projections according to an acquisition geometry; mapping the pixels onto voxels from the volume according to an acquisition geometry; extracting neighboring pixels around each mapped pixel in the projection; extracting neighboring voxels surrounding each mapped pixel in the volume; supplying the regression model with the extracted neighboring pixel data and the extracted neighboring voxel data to generate a reconstructed value for the pixel; and The reconstruction is repeated for each pixel to be reconstructed, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

29. A system comprising: A regression model trainer, wherein the regression model trainer trains the regression model; a pixel identifier that identifies pixels to be reconstructed; an imaging projection data receiver, the imaging projection data receiver receiving a set of imaging projections; a volume receiver, the volume receiver receiving a volume; as well as A pixel reconstructor, the pixel reconstructor comprising: a pixel mapper that maps the pixels onto each imaging projection in the set of imaging projections according to an acquisition geometry and maps the pixels onto voxels from the volume according to the acquisition geometry; a neighboring pixel extractor that extracts neighboring pixels surrounding a mapped pixel in the imaging projection; a neighboring voxel extractor that extracts neighboring voxels surrounding a mapped pixel in the volume; and a regression model feeder, the regression model feeder feeding the regression model with the extracted neighboring pixel data and the extracted neighboring voxel data to generate a reconstructed value for the pixel, The regression model trainer is configured to train the regression model based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

30. The system of claim 29, wherein: The regression model trainer comprises: a digitally simulated human body modeler (DAP) comprising an acquisition simulator, an algorithm builder, and a DAP database; a computed tomography (CT) modeler comprising an acquisition simulator, an algorithm builder, and a CT database; and An algorithm modifier includes an acquisition reconstructor and an algorithm database.

31. The system according to claim 29, further comprising a feedback generator, which identifies whether an error is generated on the reconstructed image and transmits it to the regression model trainer to retrain the regression model.

32. A non-transitory computer-readable storage medium comprising instructions that, when executed, cause a machine to at least: receiving a set of imaging projection data; Receiving capacity; identifying pixels to be reconstructed; Receive the trained regression model; as well as The pixel is reconstructed by the following steps: mapping the pixel onto each imaging projection in the set of imaging projections according to an acquisition geometry; mapping the pixels onto voxels from the volume according to an acquisition geometry; extracting neighboring pixels around each mapped pixel in the projection; extracting neighboring voxels around a mapped pixel in the volume; supplying the regression model with the extracted neighboring pixel data and the extracted neighboring voxel data to generate a reconstructed value for the pixel; and The reconstruction is repeated for each pixel to be reconstructed, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

33. The non-transitory computer-readable storage medium of claim 32, further comprising instructions that, when executed, cause the machine to identify errors within the reconstructed image and communicate with the regression model trainer to retrain the regression model.

34. A method comprising: Receiving capacity; receiving an orientation parameter and a thickness parameter for a thick slice to be reconstructed; identifying voxels from the thick slice to be reconstructed; receiving a trained regression model; and The voxels are reconstructed by the following steps: mapping voxels from the slab onto voxels from the volume according to an orientation parameter and a thickness parameter of the slab; extracting neighboring voxels from a volume surrounding each mapped voxel from the slab; feeding the regression model with neighboring voxels extracted from the volumetric data to produce reconstructed values ​​for voxels from the slab; as well as The reconstruction is repeated for each voxel from the slab to be reconstructed, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

35. A system comprising: a volume receiver, the volume receiver receiving a volume; A regression model trainer, wherein the regression model trainer trains the regression model; an orientation parameter receiver, the orientation parameter receiver receiving orientation parameters for use in reconstructing the slab; a thickness parameter receiver, the thickness parameter receiver receiving thickness parameters for use in reconstructing the thick slice; a voxel identifier that identifies voxels from a slab to be reconstructed; as well as A voxel reconstructor, the voxel reconstructor comprising: a voxel mapper that maps voxels from the slab onto voxels from the volume based on orientation parameters and thickness parameters of the slab; a neighboring voxel extractor that extracts neighboring voxels from a volume surrounding each mapped voxel from the slab; and a regression model feeder that feeds the regression model with neighboring voxels extracted from the volume data to produce reconstructed values ​​for the voxels from the slab, The regression model trainer is configured to train the regression model based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

36. The system of claim 35, wherein: The regression model trainer comprises: a digitally simulated human body modeler (DAP) comprising an acquisition simulator, an algorithm builder, and a DAP database; a computed tomography (CT) modeler comprising an acquisition simulator, an algorithm builder, and a CT database; and An algorithm modifier includes an acquisition reconstructor and an algorithm database.

37. The system according to claim 35 further includes a feedback generator, which identifies whether an error is generated on the reconstructed image and transmits it to the regression model trainer to retrain the regression model.

38. A non-transitory computer-readable storage medium comprising instructions that, when executed, cause a machine to at least: Receiving capacity; receiving an orientation parameter and a thickness parameter for a thick slice to be reconstructed; identifying voxels from the thick slice to be reconstructed; receiving a trained regression model; and The voxels are reconstructed by the following steps: mapping voxels from the slab onto voxels from the volume according to an orientation parameter and a thickness parameter of the slab; extracting neighboring voxels from a volume surrounding each mapped voxel from the slab; feeding the regression model with neighboring voxels extracted from the volumetric data to produce reconstructed values ​​for voxels from the slab; as well as The reconstruction is repeated for each voxel from the slab to be reconstructed, Wherein, the regression model is trained based on at least one of the following: a database comprising acquired projection data and 2D mammograms acquired at the same compression, the regression model being trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; a database comprising simulated projection data and simulated 2D mammograms acquired from a digital simulated human body model at the same compression, the regression model being trained to output 2D images approximately identical to the simulated 2D mammograms when fed with the simulated projections; a database comprising a digital simulated human body model and simulated projection data obtained from the human body model for a specific acquisition geometry, the regression model being trained to output approximately the same volume as the simulated human body model when fed with simulated projections; a database comprising computed tomography (CT) reconstructed data and simulated projection data obtained from the CT reconstructed data, the regression model being trained to output approximately the same volume as the CT reconstructed data when fed with the simulated projections; or A database includes acquired projection data and data reconstructed from these projection data using a specific reconstruction algorithm, and the regression model is trained to output a volume approximately the same as the reconstructed data when fed with the acquired projections.

39. The non-transitory computer-readable storage medium of claim 38, further comprising instructions that, when executed, cause the machine to identify errors within the reconstructed image and communicate with the regression model trainer to retrain the regression model.

Images