A system and method for generating denoised spectral CT images from spectral CT image data acquired using a spectral CT imaging system.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- GE PRECISION HEALTHCARE LLC
- Filing Date
- 2023-08-10
- Publication Date
- 2026-08-03
Smart Images
Figure 0007899445000048 
Figure 0007899445000049 
Figure 0007899445000050
Abstract
Description
[Technical Field]
[0001] The embodiments of the subject matter disclosed herein relate to X-ray technology and X-ray imaging, as well as corresponding image reconstruction and imaging tasks. In particular, the embodiments of the subject matter disclosed herein relate to systems and methods for generating denoised spectral CT (computed tomography) images from spectral CT image data acquired using a spectral (energy-resolved) CT imaging system. [Background technology]
[0002] X-ray imaging systems, such as CT (computed tomography) imaging systems and more common X-ray imaging systems, have been used for many years in medical applications such as medical diagnosis and treatment.
[0003] Typically, an X-ray imaging system, such as a CT imaging system, includes an X-ray source and an X-ray detector containing multiple detector modules, each consisting of one or more detector elements for independently measuring X-ray intensity. The X-ray source emits X-rays, which pass through the subject or object being imaged and are then received by the detector. The energy spectrum of a typical medical X-ray tube is broad, ranging from zero to 160 keV. Therefore, X-ray detectors usually detect X-rays at various energy levels.
[0004] The X-ray source and X-ray detector are typically positioned to rotate around the subject or object on a rotating member of the gantry. The emitted X-rays attenuate as they pass through the subject or object, and the resulting transmitted X-rays are measured by the detector. The measured data is used to reconstruct an image of the subject or object.
[0005] The challenge for X-ray detectors is to extract as much information as possible from the detected X-rays and inject it into images of objects or subjects. [Prior art documents] [Patent Documents]
[0006] [Patent Document 1] U.S. Patent Application Publication No. 2018 / 0293762 [Overview of the project]
[0007] Referring to Figure 1A, it would be useful to briefly describe an exemplary conventional X-ray imaging system. In this exemplary embodiment, the X-ray imaging system 100 includes an X-ray source 10, an X-ray detector 20, and an associated image processing system 30. Generally, the X-ray detector 20 is optionally focused by an X-ray optical system or collimators and is configured to register radiation from the X-ray source 10 that has passed through an object, subject, or part thereof. The X-ray detector 20 is connectable to the image processing system 30 via appropriate readout electronics built into the X-ray detector 20, at least in part, to enable image processing and / or image reconstruction by the image processing system 30.
[0008] Incidentally, conventional CT imaging systems include an X-ray source and X-ray detector positioned to acquire projection images of a subject or object at different view angles covering at least 180 degrees. This is most commonly achieved by mounting the X-ray source and detector on a support that can rotate around the subject or object, such as a rotating member of a gantry. An image containing projections recorded on different detector elements for different view angles is called a sinogram. Hereafter, even if the detector is two-dimensional, a set of projections recorded on different detector elements for different view angles will be referred to as a sinogram, and the sinogram will be considered a three-dimensional image.
[0009] Figure 1B is a schematic diagram showing an example of a conventional X-ray imaging system setup, illustrating the projection lines from the X-ray source through the object to the X-ray detector.
[0010] A further development in X-ray imaging is energy-resolved X-ray imaging, also known as spectral X-ray imaging. This can be achieved by rapidly switching an X-ray source between two different emission spectra, using two or more X-ray sources that emit different X-ray spectra, or using an energy-discriminating detector which measures the incoming radiation in two or more energy levels. An example of such a detector is a multi-bin photon counting detector, in which each recorded photon generates a current pulse, and the number of photons incident in each of multiple energy bins is counted by comparing this current pulse with a set of thresholds.
[0011] Spectral X-ray projection measurement yields projection images for each energy level. The weighted sum of these projection images is described in "SNR and DQE analysis of broad spectrum X-ray imaging", Tapiovaara and Wagner, Phys. Med. Biol. 30, 519.
[0012] Another technique enabled by energy-resolved X-ray imaging is basis material resolution. This technique takes advantage of the fact that all materials composed of elements with low atomic numbers, such as human tissue, have a linear decay coefficient whose energy dependence can be approximately expressed as a linear combination of two (or more) basis functions. μ(E) = a1f1(E) + a2f2(E) Here, f1 and f2 are basis functions, and a1 and a2 are the corresponding basis coefficients. More generally, f i is a basis function, a iis the corresponding basis coefficient, for i = 1, …, N, where N is the total number of basis functions. If there is one or more elements with a sufficiently high atomic number such that a K absorption edge exists in the energy range used for imaging in the imaging volume, it is necessary to add one basis function for each such element. In the field of medical imaging, such K-edge elements can typically be iodine or gadolinium, substances used as contrast agents.
[0013] Basis material decomposition is described in ““Energy-selective reconstructions in X-ray computerized tomography”, Alvarez, Macovski, Phys Med Biol. 1976; 21(5):733-744”. In basis material decomposition, the integrated value of each basis coefficient is, for i = 1, …, N (where N is the number of basis functions),
Number
Number
[0014] Next, under the assumption that the count number of each bin is a Poisson-distributed random variable, A can be estimated using the maximum likelihood method. This is achieved by minimizing the negative log-likelihood function. See, for example, "K-edge imaging in X-ray computed tomography using multi-bin photon counting detectors", Roessl and Proksa, Phys. Med. Biol. 52 (2007), 4679-4696. i Here, m is the measured count number in energy bin i, and Mb is the number of energy bins. [Number] Here, i is the measured count number in energy bin i, and Mb is the number of energy bins.
[0015] As a result, [Number] When is arranged in the image matrix, a substance-specific projection image (also called a basis image) for each basis i is obtained. This basis image can be viewed directly (such as in projection X-ray imaging), or used as an input to a reconstruction algorithm to form a basis coefficient map a of the object interior (such as in CT imaging). In either case, the result of basis decomposition can be regarded as one or more basis image representations such as line integrals of basis coefficients or the basis coefficients themselves. i In standard data management procedures for X-ray imaging systems, various approaches for optimizing data acquisition are presented, but in some cases, costs such as spatial resolution, noise level, and / or system complexity are involved.
[0016] In standard data management procedures for X-ray imaging systems, various approaches for optimizing data acquisition are presented, but in some cases, costs such as spatial resolution, noise level, and / or system complexity are involved.
[0017] The base coefficient a inside the object i The map of is called a basis material image, a basis image, a material image, a material-specific image, a material map or a basis map.
[0018] However, a well-known limitation of this approach is that the variance of the estimated line integrals normally increases with the number of bases used in the basis decomposition. In particular, this results in an unfortunate trade-off between improved tissue quantification and increased image noise.
[0019] Furthermore, it is actually difficult to perform accurate basis decomposition using two or more basis functions, and artifacts, bias or excessive noise may occur. Also, such basis decomposition may require extensive calibration measurements and preprocessing of data to obtain accurate results.
[0020] Since many image reconstruction tasks are inherently complex, artificial intelligence (AI) and deep learning have begun to be used in general image reconstruction and satisfactory results have been obtained. It is desirable to be able to use AI and deep learning for X-ray image processing tasks including spectral CT. However, in general, improvement of noise removal methods in spectral CT is required.
[0021] Another current problem in deep learning-based image reconstruction is its limited explainability. While an image may appear to have very low noise levels, it actually contains errors due to biases in the neural network estimator. Explainable AI techniques would be able to provide some information about why the output image has certain characteristics, based on the input image and training data.
[0022] Another drawback of existing AI technologies is that they are generally not tunable, meaning they can only provide a single output image for a given input image, and there is no possibility of adjusting the characteristics of the output image without retraining the network.
[0023] Therefore, there is generally a need for improved noise reduction methods for spectral CT (Computed Tomography), and in particular, for noise reduction methods that are explainable and accommodable. overview
[0024] This abstract introduces concepts that are described in more detail in the detailed explanation. This abstract should not be used to identify the essential features of the claimed subject matter, nor should it be used to limit the scope of the claimed subject matter. [Problems that the invention aims to solve]
[0025] The inventors understand that image reconstruction in spectral CT imaging is more difficult for two main reasons: 1) In addition to increased resolution, the multiple bins and multiple materials in the analysis significantly increase the amount of data to be processed. 2) Efficient material decomposition and image reconstruction methods tend to produce noisy images that do not satisfy the expected image quality. Therefore, there is a need to denoise material images.
[0026] This disclosure presents a fast denoising tool based on deep learning and linear minimum mean square error (LMMSE), which is combined with a model-based deep learning approach. In this method, prior knowledge is incorporated into the denoising tool, but a linear estimator structure is maintained to provide interpretability of the results. The architecture with a linear estimator using matrices and vectors estimated by a deep neural network offers greater flexibility than conventional deep learning denoising tools.
[0027] This interpretability allows for the generation of images with desired characteristics by adjusting the coefficients of the estimated linear estimator. For example, one or more coefficients of the linear estimator can be increased or decreased to reduce large-area bias or to improve the preservation of fine details in the image.
[0028] This adjustment can be performed before image reconstruction, for example, during the development of the reconstruction method, or it can be done in real time while the image is displayed to the end user, allowing the user to adjust the image to obtain the desired image characteristics.
[0029] According to a first embodiment, a method is provided for denoising spectral CT image data, which includes determining a denoising linear estimate of spectral CT image data by maximizing or minimizing a first objective function, wherein at least one parameter of the denoising linear estimate is determined by at least one machine learning system.
[0030] According to a second embodiment, a CT imaging system is provided comprising an X-ray source configured to emit X-rays, an X-ray detector configured to generate spectral CT image data, and a processor configured to determine a denoising linear estimate of the generated spectral CT image data based on maximizing or minimizing a first objective function, wherein the processor is further configured to determine at least one parameter of the linear estimate by at least one machine learning system.
[0031] Various systems and methods are provided for denoising spectral CT image data, which include determining a denoising linear estimate of spectral CT image data by maximizing or minimizing a first objective function, wherein at least one parameter of the denoising linear estimate is determined by at least one machine learning system. The denoiser is based on a linear least mean squares error (LMMSE) estimator. Although LMMSE is computationally very fast, it is not commonly used for denoising CT images because it cannot adapt the denoising amount to different parts of the image and it is difficult to derive accurate statistical characteristics from CT image data. To overcome these problems, model-based deep learning models, such as deep neural networks that preserve a model-based LMMSE structure, are used. [Brief explanation of the drawing]
[0032] This embodiment, along with its further objectives and advantages, will be best understood by referring to the following description in conjunction with the accompanying drawings. [Figure 1A] This is a schematic diagram showing an example of an X-ray imaging system. [Figure 1B] This is a schematic diagram showing an example of an X-ray imaging system. [Figure 2] This is a schematic diagram showing other examples of X-ray imaging devices, such as CT imaging devices. [Figure 3] This is a schematic block diagram of a CT imaging system as an example of an X-ray imaging system. [Figure 4] This is a schematic diagram showing other examples of related components for X-ray imaging devices such as CT scanners. [Figure 5] This is a schematic diagram of a photon counting circuit and / or apparatus according to an exemplary embodiment. [Figure 6] This is a schematic diagram showing an example of a semiconductor detector submodule according to an exemplary embodiment. [Figure 7]This is a schematic diagram showing an example of a semiconductor detector submodule according to another exemplary embodiment. [Figure 8A] This is a schematic diagram showing an example of a semiconductor detector submodule according to yet another exemplary embodiment. [Figure 8B] This is a schematic diagram showing an example of a set of detector submodules arranged in a tile-like pattern, where each detector submodule is a depth-divided detector submodule, and an application-specific integrated circuit (ASIC) or corresponding circuit is located beneath the detector submodule as viewed from the direction of the incident X-rays. [Figure 9] This is a schematic diagram of a conventional denoising technique based on a convolutional neural network (CNN). In this technique, the CNN directly maps the noisy image to the clean image. [Figure 10] This is a schematic diagram of the presented denoising technique, in which a CNN is used to map linear model parameters W and b, which are used to generate a clean image. [Figure 11] An example of the interpretability of the proposed model is shown using three parameters σ, β, and λ, which are used to control various components of the output image. [Figure 12] This shows one example of how the two components of a = WX + b can be interpreted in the result. [Figure 13] This demonstrates the parallelism between this method and LMMSE. The first row represents the estimator components of a standard pre-trained LMMSE (linear least mean squared error) for a specific phantom example, and the second row represents the estimator components equivalent to our pre-trained approach. [Figure 14] This is a schematic diagram showing an example of a computer implementation according to an embodiment. [Modes for carrying out the invention]
[0033] Next, embodiments of the present disclosure will be described illustratively with reference to the figures.
[0034] To better understand the present invention, it would be useful to continue with an introductory explanation of a non-limiting example of an entire X-ray imaging system in which data processing and transfer may be carried out according to the concepts of the present invention.
[0035] Figure 2 is a schematic diagram showing an example of an X-ray imaging system 100, such as a CT imaging system, which includes an X-ray source 10 that emits X-rays, an X-ray detector 20 equipped with an X-ray detector that detects X-rays that have passed through an object, an analog processing circuit 25 that processes and digitizes the raw electrical signals from the X-ray detector, a digital processing circuit 40 that can perform further processing operations on the measurement data, such as applying corrections, temporary storage, and filtering, and a computer 50 that can store the processed data and perform further post-processing and / or image reconstruction. The digital processing circuit 40 may include a digital processor. According to the exemplary embodiment, all or part of the analog processing circuit 25 may be implemented within the X-ray detector 20. The X-ray source and X-ray detector may be coupled to a rotating member of the gantry 11 of the CT imaging system 100.
[0036] The entire X-ray detector can be considered as an X-ray detector system 20, or as an X-ray detector 20 combined with an associated analog processing circuit 25.
[0037] The image processing system 30, which communicates with and is electrically coupled to the analog processing circuit 25, may include a digital processing circuit 40 and / or a computer 50, and may be configured to perform image reconstruction based on image data from the X-ray detector. Therefore, the image processing system 30 can be viewed as the computer 50, or a system combining the digital processing circuit 40 and the computer 50, or as the digital processing circuit 40 alone if the digital processing circuit is further specialized for image processing and / or reconstruction.
[0038] A commonly used example of an X-ray imaging system is a CT imaging system, which may include an X-ray source or X-ray tube that generates a fan beam or cone beam of X-rays, and an opposing array of X-ray detectors that measure the percentage of X-rays that have passed through the patient or object. The X-ray source or X-ray tube and the X-ray detectors are mounted on a gantry 11 that can rotate around the object being imaged.
[0039] Figure 3 schematically shows a CT imaging system 100 as an example of an X-ray imaging system. This CT imaging system includes a computer 50 that receives commands and scanning parameters from an operator via a display 62 and an operator console 60 which may have some form of operator interface, such as a keyboard, mouse, joystick, touchscreen, or other input device. The commands and parameters supplied by the operator are then used by the computer 50 to supply control signals to the X-ray control device 41, the gantry control device 42, and the table control device 43. Specifically, the X-ray control device 41 supplies power and timing signals to the X-ray source 10 to control the emission of X-rays to an object or patient lying on the table 12. The gantry control device 42 controls the rotational speed and position of the gantry 11 which comprises the X-ray source 10 and the X-ray detector 20. As an example, the X-ray detector 20 may be a photon-counting type X-ray detector. The table control device 43 controls and determines the position of the patient table 12 and the scanning range of the patient. There is also a detector controller 44 configured to control the X-ray detector 20 and / or receive data from the X-ray detector 20.
[0040] In this embodiment, the computer 50 also performs post-processing and image reconstruction of the image data output from the X-ray detector 20. Thus, the computer 50 corresponds to the image processing system 30 shown in Figures 1 and 2. The operator can observe the reconstructed image and other data from the computer 50 via the associated display 62.
[0041] An X-ray source 10 positioned in the gantry 11 emits X-rays. An X-ray detector 20, which may be in the form of a photon-counting X-ray detector, detects the X-rays after they have passed through an object or patient. The X-ray detector 20 is formed by a plurality of pixels, also called sensors or detector elements, and associated processing circuits, such as an ASIC (Application Specific Integrated Circuit), positioned in the detector module. Some of the analog processing is implemented in the pixels, and the remaining processing is implemented in the ASIC, for example. In one embodiment, the processing circuit (ASIC) digitizes the analog signals from the pixels. The processing circuit (ASIC) can also constitute digital processing that can perform further processing operations on the measurement data, such as applying corrections, temporary storage, and / or filtering. During scanning to acquire X-ray projection data, the gantry and the components mounted on it rotate around the isocenter 13.
[0042] Modern X-ray detectors typically need to convert incident X-rays into electrons. This is usually done by the photoelectric effect or Compton interaction, and the resulting electrons usually produce secondary visible light until their energy is lost, which is detected by a photosensitive material. There are also semiconductor-based detectors, in which case the electrons generated by the X-rays create charge in the form of electron-hole pairs that are collected by an applied electric field.
[0043] Detectors operating in energy integration mode exist in the sense that they provide a signal integrated from a large number of X-rays. The output signal is proportional to the total energy deposited by the detected X-rays.
[0044] In medical X-ray applications, X-ray detectors with photon counting and energy resolution are becoming common. Photon counting detectors, in principle, can measure the energy of each X-ray, offering the advantage of obtaining additional information about the composition of the subject. This information can be used to improve image quality and reduce radiation dose.
[0045] Generally, photon-counting X-ray detectors determine the energy of photons by comparing the height of the electrical pulses generated by photon interactions within the detector material with a series of comparator voltages. These comparator voltages are also called energy thresholds. Typically, the analog voltages of the comparators are set by a digital-to-analog converter (DAC). The DAC converts the digital settings transmitted from the controller into analog voltages that can be used to compare the height of the photon pulses.
[0046] A photon counting detector counts the number of photons that interact within the detector during a measurement period. New photons are generally identified when the height of an electrical pulse exceeds the comparator voltage of at least one comparator. Once a photon is identified, the event is saved by incrementing a digital counter associated with the channel.
[0047] When multiple different thresholds are used, an energy-discriminating photon counting detector is obtained that can sort detected photons into energy bins corresponding to various thresholds. This type of photon counting detector is sometimes called a multi-bin detector. Generally, energy information can be used to create a new kind of image where new information is available and image artifacts inherent in the prior art can be removed. In other words, in an energy-discriminating photon counting detector, the pulse height is compared to a number N of programmable thresholds (T1-TN) in a comparator and classified according to the pulse height, which is proportional to the energy. In other words, a photon counting detector containing one or more comparators is called a multi-bin photon counting detector. In the case of a multi-bin photon counting detector, the photon count is usually stored in a set of counters, one for each energy threshold. For example, one count can be assigned to the highest energy threshold that a photon pulse exceeds. In another example, the counters record the number of times the photon pulse exceeds each energy threshold.
[0048] For example, edge-on is a special, non-restrictive design for photon counting detectors in which the X-ray sensor, such as an X-ray detector element or pixel, is oriented edge-on relative to the incident X-ray.
[0049] For example, such a photon count detector may have pixels in at least two directions, one of which of the directions of the edge-on photon count detector has a component in the direction of X-rays. Such an edge-on photon count detector is sometimes called a depth-segment photon count detector, having pixels in two or more depth segments in the direction of incident X-rays. Note that one detector element may correspond to one pixel, and / or multiple detector elements may correspond to one pixel, and / or data signals from multiple detector elements may be used for one pixel.
[0050] Alternatively, the pixels may be arranged as an array (non-depth-segmented) in a direction substantially orthogonal to the direction of the incident X-rays, and each pixel may be oriented edge-on with respect to the incident X-rays. In other words, the photon counting detector may be non-depth-segmented while remaining edge-on with respect to the incident X-rays.
[0051] By arranging the edge-on photon counting detector edge-on, absorption efficiency can be increased, the absorption depth can be selected to any length, and the edge-on photon counting detector can be fully depleted without requiring very high voltages.
[0052] The conventional mechanism for directly detecting X-ray photons through a semiconductor detector basically works as follows: The energy of the X-ray interaction in the detector material is converted into electron-hole pairs inside the semiconductor detector, and the number of electron-hole pairs is generally proportional to the photon energy. The electrons and holes drift toward the electrodes and the back surface of the detector (or vice versa). During this drift, the electrons and holes induce an electric current in the electrodes, and this current can be measured.
[0053] As shown in Figure 4, the signal is routed via a routing path 26 from the detector element 22 of the X-ray detector to the input of an analog processing circuit (e.g., an ASIC) 25. It should be understood that the term Application Specific Integrated Circuit (ASIC) is broadly interpreted as a general-purpose circuit used and configured for a specific application. An ASIC can be used to process the charge generated from each X-ray, convert it into digital data, and obtain measurement data such as the number of photons and / or estimated energy. Since the ASIC is configured to connect to a digital processing circuit, the digital data is sent to the digital processing circuit 40 and / or one or more memory circuits or memory components 45, and finally, the data becomes the input to the image processing circuit 30 or computer 50 in Figure 2 for generating a reconstructed image.
[0054] Since the number of electrons and holes generated from a single X-ray event is proportional to the energy of the X-ray photon, the total charge contained in a single induced current pulse is proportional to this energy. After the filtering step in the ASIC, the pulse amplitude is proportional to the total charge of the current pulse and therefore proportional to the X-ray energy. The pulse amplitude can be measured by comparing it to one or more thresholds (THRs) of one or more comparators (COMPs) and their values, and a counter is introduced to record the number of times the pulse is greater than the threshold. In this way, it is possible to count and / or record the number of X-ray photons with energy exceeding the energy corresponding to each threshold (THR) detected within a given time.
[0055] ASICs typically sample an analog photon pulse once per clock cycle and record the output of a comparator. The comparator (threshold) outputs either 1 or 0 depending on whether the analog signal is above or below the comparator voltage. The information available for each sample is, for example, the 1 or 0 of each comparator indicating whether the comparator was triggered (the photon pulse was above the threshold).
[0056] A photon counting detector typically has photon counting logic that determines whether a new photon has been recorded and records the photon in a counter. In the case of a multi-bin photon counting detector, there are generally multiple counters, for example, one for each comparator, and the photon count is recorded in the counter according to an estimate of the photon energy. The logic can be implemented in several different ways. Two of the most common categories of photon counting logic are non-paralyzable counting modes and paralyzable counting modes. Other photon counting logic includes, for example, local maxima detection, which counts the detected local maxima in a voltage pulse and, in some cases, also records the height of that pulse.
[0057] Photon count detectors offer many advantages, including high spatial resolution, low sensitivity to electronic noise, excellent energy resolution, and material separation capability (spectral imaging capability), but are not limited to these. However, energy-integrating detectors have the advantage of high count-rate tolerance. Count-rate tolerance stems from the fact / recognition that because the total energy of the photons is measured, adding one photon always increases the output signal (within a reasonable range), regardless of the amount of photons currently recorded in the detector. This advantage is one of the main reasons why energy-integrating detectors have become the standard in medical CT today.
[0058] Figure 5 is a schematic diagram of a photon counting circuit and / or apparatus according to an exemplary embodiment.
[0059] When photons interact in a semiconductor material, clouds of electron-hole pairs are generated. When an electric field is applied to a detector material, charge carriers are collected at electrodes attached to the detector material. The signal is sent from the detection element to a parallel processing circuit, such as an ASIC. For example, an ASIC can process the charge to generate a voltage pulse of maximum height proportional to the amount of energy of the photons accumulated in the detector material.
[0060] The ASIC may include a pair of comparators 302, each comparing the magnitude of a voltage pulse to a reference voltage. The comparator output is typically zero or one (0 / 1) depending on which of the two compared voltages is greater. Here, we assume that the comparator output is 1 if the voltage pulse is higher than the reference voltage, and 0 if the reference voltage is higher than the voltage pulse. A digital-to-analog converter (DAC) 301 may be used to convert a digital setting, which may be supplied by the user or a control program, into a reference voltage that can be used by the comparators 302. We will say that a particular comparator is triggered if the height of the voltage pulse exceeds the reference voltage of that comparator. Each comparator is generally associated with a digital counter 303, which is incremented based on the comparator output according to photon counting logic.
[0061] As mentioned above, the estimated values for each projection line
number
[0062] It will be understood that the mechanisms and arrangements described herein can be implemented, combined, and rearranged in a variety of ways.
[0063] For example, the embodiment may be implemented in hardware, at least part of it may be implemented in software for execution by a suitable processing circuit, or a combination of both.
[0064] The steps, functions, procedures, and / or blocks described herein can be implemented in hardware using prior art, including both general-purpose electronic circuits and application-specific circuits, such as discrete circuits and integrated circuit technologies.
[0065] Alternatively, or as a complement, at least some of the steps, functions, procedures, and / or blocks described herein may be implemented in software, such as a computer program, for execution by a suitable processing circuit, such as one or more processors or processing units.
[0066] The following describes non-limiting examples of specific detector modules. More specifically, these examples refer to edge-on oriented detector modules and depth-segmented detector modules. Other types of detectors and detector modules may also be feasible.
[0067] Figure 6 is a schematic diagram showing an example of a semiconductor detector submodule according to an exemplary embodiment. This is an example of a detector module 21 equipped with a semiconductor sensor having multiple detection elements or pixels 22, where each detection element (or pixel) typically has a diode with a charge collection electrode as its main component. X-rays are incident from the edge of the detector module.
[0068] Figure 7 is a schematic diagram showing an example of a semiconductor detector submodule according to another exemplary embodiment. In this example as well, the detector module 21 having a semiconductor sensor is divided into multiple depth segments or detector elements 22 in the depth direction, assuming that X-rays are incident from the edge of the detector module.
[0069] Typically, the detection element is the individual X-ray sensitive sub-element of the detector. Generally, photon interactions occur within the detector element, and the resulting charge is collected at the corresponding electrode of the detector element.
[0070] Each detection element typically measures the incident X-ray beam as a sequence of frames. A frame is a measurement data set at a specified time interval called the frame time.
[0071] Depending on the detector topology, detector elements may correspond to pixels, especially if the detector is a flat-panel detector. A depth-divided detector can be considered to have multiple detector strips, each strip having multiple depth segments. In such a depth-divided detector, each depth segment can be considered a separate detector element, particularly if each depth segment is associated with its own separate charge collection electrode.
[0072] The detector strips of a depth-resolved detector are sometimes called pixel strips because they typically correspond to pixels in a standard flat-panel detector. However, it is also possible to view a depth-resolved detector as a three-dimensional pixel array, in which case each pixel corresponds to an individual depth segment / detector element.
[0073] The semiconductor sensor can be implemented as a so-called multi-chip module (MCM) for multiple ASICs, which are mounted using the semiconductor sensor as a base substrate for electrical wiring, preferably by flip-chip technology. The wiring includes signal connections from each pixel or sensing element to the ASIC input, and connections from the ASIC to external memory and / or digital data processing. Power to the ASIC can be supplied through similar wiring, taking into account the increased cross-sectional area required for high currents in these connections, although power can also be supplied through separate connections. The ASIC can be positioned on the side of the active sensor. This means that if an absorption cover is positioned on top, it can be protected from incident X-rays, and if the absorber is also positioned in this direction, it can be protected from scattered X-rays from the side.
[0074] Figure 8A is a schematic diagram showing a detector module implemented as an MCM similar to an embodiment of U.S. Patent No. 8,183,535. In this example, it is shown that the semiconductor sensor 21 can also function as a substrate within the MCM. The signal is routed from the detection element 22 to the input of a parallel processing circuit 24 (e.g., an ASIC) located next to the active sensor area via a wiring path 23. The ASIC processes the charge generated from each X-ray and converts it into digital data that can be used to detect photons and / or estimate the energy of the photons. The ASIC may have its own digital processing circuitry and memory for small tasks. The ASIC may also be configured to connect to digital processing circuitry and / or memory circuitry or components located outside the MCM, and the data is ultimately used as input for reconstructing the image.
[0075] However, the adoption of depth segments presents two significant challenges for silicon-based photon count detectors. First, a large number of ASIC channels must be employed to process the data supplied from the relevant detector segments. In addition to the reduction in pixel size and the increase in the number of channels due to depth segmentation, multi-energy bins further increase the data size. Second, because a given X-ray input count is divided into smaller pixels, segments, and energy bins, the signal in each bin is very low, and detector calibration / correction requires several orders of magnitude more calibration data to minimize statistical uncertainty.
[0076] Naturally, as data size increases by several orders of magnitude, greater computing resources are required, such as hard disks, memory, and central processing units (CPUs) or graphics processing units (GPUs), and both data processing and preprocessing become slower. For example, when data size increases from 10 megabytes to 10 gigabytes, the time it takes to read and write the data increases by a thousandfold.
[0077] A problem with counting-type X-ray photon detectors is pile-up. When the X-ray photon flux rate is high, there can be problems distinguishing between two consecutive charge pulses. As mentioned earlier, the filtered pulse length depends on the shaping time. If this pulse length is longer than the time between two X-ray photon-induced charge pulses, the pulses grow together, and the two photons become indistinguishable and may be counted as a single pulse. This is called pile-up. Therefore, one way to avoid pile-up at high photon flux is to use a short shaping time or to use depth segmentation.
[0078] For pile-up calibration vector generation, pile-up calibration data must be preprocessed for spit correction. For material-resolved vector generation, it is desirable that material-resolved data be preprocessed for both spit and pile-up correction. For patient scan data, spit, pile-up, and material-resolved preprocessing is necessary before image reconstruction. These are simplified examples illustrating "preprocessing," as actual preprocessing steps may include several other calibration steps as needed, such as reference normalization and air calibration. The term "processing" may sometimes refer only to the final step in each calibration vector generation or patient scan, but is sometimes used interchangeably.
[0079] Figure 8B is a schematic diagram showing an example of a tiled set of detector submodules, where each detector submodule is a depth-segmented detector submodule, and the ASIC or corresponding circuit 24 is positioned below the detector element 22 as viewed from the direction of the incident X-rays, enabling routing paths 23 from the detector element 22 to the parallel processing circuit 24 (e.g., ASIC) in the space between the detector elements.
[0080] Artificial intelligence (AI) and deep learning are beginning to be used in general image reconstruction, yielding some satisfactory results. However, the current problem with deep learning-based image reconstruction is its limited explainability. Even if an image appears to have a very low noise level, it actually contains errors due to the bias of the neural network estimator.
[0081] Generally, deep learning refers to machine learning methods based on representation learning, using artificial neural networks or similar architectures. Learning can be supervised, semi-supervised, or unsupervised. Deep learning systems such as deep neural networks, deep belief networks, recurrent neural networks, and convolutional neural networks are applied to a wide range of technological fields, including computer vision, speech recognition, natural language processing, social network filtering, machine translation, and board game programs, producing results that rival, and in some cases surpass, the performance of human experts.
[0082] The adjective "deep" in deep learning derives from the use of multiple layers in the network. Early research showed that linear perceptrons could not be universal classifiers, but that a network with a non-polynomial activation function and one hidden layer of unlimited width could be a universal classifier. Deep learning is a modern variation that deals with an unlimited number of layers, enabling practical applications and optimized implementations. In deep learning, for efficiency, ease of learning, and ease of understanding, layers are heterogeneous, and are allowed to deviate significantly from biologically informed connectionist models.
[0083] The inventors recognized the need for noise reduction algorithms that improve the performance of spectral CT, particularly algorithms with improved explainability.
[0084] The proposed technique is generally applicable to providing denoised image data in spectral CT based on neural networks and / or deep learning.
[0085] To provide an exemplary framework for facilitating the understanding of the proposed technology, we then present a concrete example of deep learning-based image reconstruction in the specific context of spectral CT image reconstruction.
[0086] However, it should be understood that the proposed techniques for providing a reliability index for deep learning-based image reconstruction in spectral CT applications are generally applicable to deep learning-based image reconstruction for CT and are not limited to the following specific examples of deep learning-based image reconstruction.
[0087] The inventors disclose a novel and fast denoising tool based on a Linear Minimum Mean Square Error (LMMSE) estimator. While LMMSE is computationally very fast, it is not commonly used for denoising CT images. This is likely due to the inability to adapt the amount of denoising to different parts of the image and the difficulty in deriving accurate statistical characteristics from CT data. To overcome these problems, the inventors propose a model-based deep learning strategy, namely a deep neural network (model-based) that preserves the LMMSE structure. In this way, the solution adapts to the anatomical structure at each point in the image and the noise characteristics at that specific location.
[0088] As an exemplary and non-limiting embodiment of this disclosure, consider a linear least mean squares error (LMMSE) for denoising two material images after FPB, i.e., x = [x1, x2]. This is the following solution.
number
number
number
[0089]
number
number
[0090] Let's explain how we approach model-based deep learning in this scenario. We have a model-based solution (LMMSE denoiser), but to obtain good estimates of W and b when using only diagonal cross- and co-covariances, we need to enhance the model-based solution (LMMSE denoiser). Therefore, we want to preserve the mathematical structure (linear, fast) by using deep learning inference (estimating LMMSE parameters with a powerful statistically independent model approach). Enforcing a problem structure requires fewer training samples and aims for a more robust neural network to unknown datasets than a typical "black box" network. Of course, this result is also expected to be far better than considering diagonal cross- and covariances with an oversimplified LMMSE. Our deep learning solution is shown in Figure 10.
[0091] Further interpretations can be made of these results. The goal of this network is to obtain W and b, not a denoised image. Therefore, if we want to manipulate and understand the solution, we can consider the parameters of W and B instead of changing or accessing millions of parameters inside the CNN. These parameters are considerably fewer and also more “interpretable” (in connection to an LMMSE).
[0092] The proposed deep learning approach requires a training database. To demonstrate the proof of concept of the proposed disclosure, we trained a pre-trained LMMSE estimator using simulated photon counting data. PCCT measurements were calculated using an 8-bin silicon detector, followed by two material decomposition and FBP to acquire material images. We simulated a set of 1200 cases. We used the KiTS19 database, which mainly consists of abdominal scans. We used 1000 samples for training and 200 samples for testing. To evaluate the robustness of the technique in adapting to invisible data, we also simulated 200 additional scans from another database (NSCLS). This database also includes whole-body scans and offers greater anatomical diversity than the training database.
[0093] In this example, PyTorch and one NVIDIA GeForce RTX 2080 Ti GPU board are used to train the neural network. For comparative study, the following competing solutions are considered: (1) the original simplified LMMSE as described in the previous section, and (2) a "black box" CNN based on the UNet architecture.
[0094] Figure 9 shows an overview of a conventional CNN-based denoising technique 90, in which a black-box CNN 94 directly maps from a noisy CT image 92 to a clean CT image 96.
[0095] Figure 10 shows an overview of the presented denoising technique 1000, in which a CNN 1004 accepts a noisy CT image 1002 as input, the CNN finds a linear estimator 1004, the CNN is used to map linear model parameters W and b, the linear estimator is used to perform denoising 1006 and generate a clean CT image. Thus, a linear structure is forced during the learning process.
[0096] Figure 11 illustrates an example of this interpretability. The learned linear components W and b can be manipulated with a few parameters. ii ("Variance" of single material component of the linear model), w ij Three parameters, σ, β, and λ, are used to control b (the "cross-covariance" of the opposite material component) and b (the mean or "bias" of the single material component). When the bias is zero (λ=0), the result shows a emphasized structure. ij When the β=0 10¹² component is zero, cross-contamination between materials is not adequately corrected. Furthermore, when the "dispersion" component is zero (σ=0 10¹⁴), the variance of single-material noise may not be reduced.
[0097] Figure 12 shows a=W x Here is an example of how to interpret the effect that the two components of +b have on the result. The first term (W x) provides anatomical structural detail (finer edges, localized small structures) as shown in CT image 1016 (structure image), but does not give an accurate CT value expressed in Hounsfield units (HU). The second independent term, b, generally corrects the HU value (mean, or bias) of results where there is little anatomical structural detail (a very smoothed image), as shown in CT image 1018 (bias image).
[0098] Figure 13 shows an empirical demonstration of the parallelism between this technique and LMMSE. The first row, 1020, represents the estimator components of a normal LMMSE for a specific phantom example, and the second row, 1030, represents the estimator components equivalent to our learned approach. It can be seen that the LMMSE is not specific enough, and the results are too blurry to accurately represent the denoised example. However, our approach is more specific and accurate while maintaining similarity to the aforementioned LMMSE. Therefore, this can be interpreted as a DNN-enhanced LMMSE (which is the best linear estimator to minimize MSE).
[0099] This disclosure relates to spectral or energy-resolved image data comprising image data containing at least two spectral components. In this context, the image data can be, for example, two-dimensional, three-dimensional, or time-resolved, and refers to either a reconstructed image or an intermediate representation of the image data such as a sinogram. The different spectral components can be, for example, a composite mono-energy image, a wide-spectrum image acquired at different tube acceleration voltages, or a material-selection image such as a baseline image. The different spectral components can also be a combination of the above.
[0100] The above description should be understood as illustrative and non-restrictive, and several variations of the described methods may be conceivable. For example, several different architectures of convolutional neural networks are possible, such as UNet, ResNet, or an unrolled iterative network, e.g., an unrolled gradient descent or unrolled primal-dual network. Furthermore, batch normalization, skip connections, and different pooling layers such as maximum pooling, average pooling, or softmax pooling can be included in the network, whether desirable or undesirable to include in network training. Various loss functions can be minimized during network training, such as L1 loss, L2 loss, perceptual loss, and adversarial loss. Perceptual loss can be implemented using different classifier networks, and different layers of such networks can be used to obtain different image characteristics.
[0101] The inventors have found that if W is a complete matrix, then using a neural network to obtain all of its elements is possible. 12We recognize that this is impractical because it would require a neural network with a certain level of output. Therefore, it is desirable to impose some structure on the matrix, for example, by making W a sparse matrix, i.e., a matrix with a small number of non-zero elements. For example, matrix W can be a diagonal matrix, in which case, when the matrix is applied, each pixel value in the set of image data is multiplied by a value. Another option is that W can be a block diagonal matrix. For example, if the spectral data consists of N spectral components, W can consist of N×N block elements along the diagonal, and when W is applied to the vector, the values corresponding to different spectral components of a particular pixel are transformed by the N×N block, resulting in a new set of spectral components in the corresponding transformed set of spectral component images.
[0102] As another example, W can be applied separately to each of the different spectral components, and the entries corresponding to crosstalk between the different components can be set to zero. In both the case where W is applied separately to each of the different spectral components, and in the more general case where W includes entries for the corresponding cross components, it can be considered to correspond to a certain maximum distance between pixels in the input and output pixels. Alternatively, W can be expressed as Fourier domain W=F -1 W F It can be expressed as a transformation in F. F is a Fourier transform operator, and W corresponds to specific frequencies such as low frequency and high frequency. F Only the elements are non-zero.
[0103] Another example is that W is an element of the range of a linear or nonlinear transformation, such as in an artificial or convolutional neural network.
[0104] The vector b can also be selected as, for example, a full vector without any restrictions or a sparse vector where only certain elements are nonzero. In another exemplary embodiment of the present disclosure, b can be restricted to a range of linear or nonlinear transformations, such as artificial or convolutional neural networks, or a linear combination of Fourier components b=F -1 b F It can be restricted, where b F is a vector of Fourier components of b that can, for example, be restricted to contain high or low spatial frequencies.
[0105] In practice, such restrictions on W and B can be imposed by having the convolutional neural network output only the elements of W and b that should be non-zero, while setting the other components to zero. In another embodiment of the present disclosure, the convolutional neural network can generate feature vectors that are subsequently transformed from W and b, for example, via a linear transformation or via an artificial neural network or another convolutional neural network.
[0106] For example, a neural network can generate numerous Fourier components and transform them to form, for example, one or more diagonals of b or W. In another embodiment of this disclosure, different components of b and / or W, or feature vectors associated with b and / or W, are given different weights in the loss function used to train the neural network to generate these components, or are penalized by a penalty term so that these components are unlikely to achieve large values. For example, the components of b and / or W can be normalized such that high spatial frequencies are penalized, meaning that these components mainly consist of low frequencies. In this way, it is possible to avoid too large a variation between transformations applied to adjacent pixels and make the denoising method more robust to differences in noise characteristics and image appearance compared to the training dataset. In another example, it is possible to provide a denoising device in which low frequencies are penalized and which is particularly well suited to preserving fine details.
[0107] The inventors highly value the linear structure of this denoisinger, which provides both explainability and tuneability. The trained LMMSE denoisinger has a mathematical structure similar to that of a conventional LMMSE denoisinger based on a handcrafted noise model. By comparing the coefficients of the trained LMMSE denoisinger with those of a conventional LMMSE denoisinger, information can be obtained about how the denoisinger acts on an image. For example, such a comparison can show that the behavior of the trained LMMSE denoisinger in a limited area of an image is similar to that of a conventional LMMSE denoisinger built on a specific model of signal and noise. This information can be useful when analyzing image quality and robustness characteristics and attempting to improve the trained LMMSE denoisinger by adjusting, for example, the structure of b and / or W or the training parameters.
[0108] The structure of the linear LMMSE denoisinger also allows for model adjustment. For example,
number
number
number
[0109] Such manipulation of coefficients can be performed by multiplying the selected set of coefficients by a constant coefficient. Alternatively,
number
number
number
[0110] As an example, the present inventors,
number
number
number
number
number
number
[0111] In another embodiment of this disclosure, tunability is achieved by training a single neural network based on a tuning parameter t.
number
[0112] In another example, t can be replaced with multiple adjustment parameters to adjust several different characteristics of the image.
[0113] In yet another embodiment of this disclosure, real-time adjustability can be achieved while the image is being displayed to the end user, allowing the user to adjust the image to obtain desired image characteristics.
[0114] In exemplary embodiments of this disclosure, the convolutional neural network is trained by minimizing an L1 loss function, an L2 loss function, a perceptual loss function, an adversarial loss function, or a combination thereof.
[0115] The goal of the network is to have a=W for a=[a1,a2] x The goal is to obtain W and b such that +b, and the denoised image corresponds to the material image x=[x1,x2]. Here we describe the case with two spectral components, but this is a non-restrictive example; vectors a and x can generally have any number of components greater than or equal to two. This goal is:
number
number
[0116] One possible solution is to use a feature-based perceptual loss that compares the output with feature representations corresponding to the output and ground truth, instead of comparing the output and ground truth pixel by pixel. Feature representations are obtained by passing the target and output through a pre-trained convolutional neural network (CNN). For example, VGG16 / 19 (a CNN from the Visual Geometry Group at Oxford University) is commonly used as a feature extractor. This perceptual loss is used in various computer vision problems, such as image denoising and super-resolution. If we denote the j-th layer of the pre-trained CNN as φj, then the perceptual loss is:
number
[0117] Another possibility is to minimize the distance (some notion of distance) between the ground truth and the distribution of the output image. This can be achieved using adversarial loss based on Generative Adversarial Networks (GANs). In this setup, the network is pitted against another CNN in a minimax game, and through successive improvements, the output distribution becomes indistinguishable from the ground truth distribution. This prevents excessive denoising and oversmoothing associated with pixel-level losses such as L2 loss and L1 loss.
number
number
number
[0118] For an optimal classifier, the purpose of the generator is,
number
number
number
number
number
[0119] While WGAN-GP is not necessarily the best-performing GAN, it is one of the most stable GANs to learn. Previous presentations have demonstrated the stability of WGAN-GP across several different tasks and datasets, without experiencing common problems such as vanishing gradients or mode collapse.
[0120] To consider the trade-offs between the advantages and disadvantages of these loss functions, we can think of a weighted sum of the loss functions mentioned earlier.
[0121] In exemplary embodiments of the present disclosure, a convolutional neural network is trained as part of a pair of cycle-consistent generative adversarial networks (a pair of cycleGANs).
[0122] The data required for this disclosure consists of paired samples of noisy material images and their ground truth (low-noise) images. However, in many cases, such paired datasets are not available. Instead, there may be piles of noisy material images and piles of denoised / low-noise material images. To extend the trained LMMSE to unpaired data, a so-called cycle-consistent GAN can be applied. The key insight that makes this possible is the cycle-consistent loss. The goal is to find a map from source domain X to target domain A. G:X->A is a map that takes a pair of noisy material images x, passes it through the network of the present invention, and forms a denoised material image a=Wx+b. Using an adversarial loss, we can push the distribution induced by G(X) such that it is indistinguishable from that of A. However, this mapping is highly constrained, and the space of possible mappings is enormous. To reduce the space of possible mappings, we can consider an inverse mapping F:A->X and enforce cycle consistency with a cycle-consistent loss. This mapping is,
number
number
[0123] This is each a unique classifier D A and D X It is combined with a GAN for mapping G and inverse mapping F. Therefore, its purpose is as follows:
number
number
[0124] Similar to the original GAN, this formulation has problems with learning stability. To avoid this, the negative log likelihood loss is replaced with the L2 loss. In other words, the generator is
number
number
[0125] The method proposed by the present inventors is a step of (1) acquiring energy-resolved CT image data, and (2)
number
number
[0126] In exemplary embodiments of this disclosure,
number
[0127] In exemplary embodiments of this disclosure, the image quality measure is a mean-squared error, structural similarity, bias, fidelity of fine details, numerical observer detectability, visual grading score, or observer performance.
[0128] In exemplary embodiments of this disclosure,
number
[0129] In another exemplary embodiment of this disclosure,
number
[0130] In another exemplary embodiment of this disclosure,
number
[0131] In exemplary embodiments of this disclosure, the convolutional neural network has a ResNet architecture, a UNet architecture, an unrolled iterative architecture, or a combination thereof.
[0132] In exemplary embodiments of this disclosure, a convolutional neural network is trained by minimizing an L1 loss function, an L2 loss function, a perceptual loss function, an adversarial loss function, or a combination thereof.
[0133] In exemplary embodiments of this disclosure, a convolutional neural network is trained as a generator for a generative adversarial network.
[0134] In exemplary embodiments of the present disclosure, a convolutional neural network is trained as part of a pair of cycle-consistent generative adversarial networks (cycle GANs).
[0135] In exemplary embodiments of this disclosure, the energy-resolved image data x is a set of sinograms.
[0136] In another exemplary embodiment of the present disclosure, the energy-resolved image data x is a set of reconstructed images.
[0137] In exemplary embodiments of the present disclosure, the different components of the energy-resolved image data x consist of monoenergetic image data at different monochromatic energies, or image data corresponding to different measured energy levels or energy bins, or different basis images.
[0138] In exemplary embodiments of this disclosure, the end user is:
number
[0139] In another exemplary embodiment of the present disclosure, a convolutional neural network is trained on a dataset containing a plurality of low-noise images, each having different image characteristics for each high-noise image, and the neural network is trained to generate low-noise images having different characteristics for each setting of at least one tuning parameter.
[0140] Figure 14 is a schematic diagram showing an example of a computer implementation according to an embodiment. In this particular example, the system 200 comprises a processor 210 and a memory 220, the memory containing instructions that can be executed by the processor, thereby enabling the processor to operate to perform the steps and / or actions described herein. The instructions are typically composed of computer programs 225, 235 and may be pre-configured in the memory 220 or downloaded from an external memory device 230. Optionally, the system 200 comprises an input / output interface 240 interconnected to one or more processors 210 and / or the memory 220, which can enable the input and / or output of associated data such as one or more input parameters and / or one or more resulting output parameters.
[0141] The term "processor" should be interpreted in a general sense as any system or device capable of executing program code or computer program instructions to perform a specific processing, decision, or computational task.
[0142] Thus, a processing circuit including one or more processors is configured to perform clearly defined processing tasks, such as those described herein, when a computer program is executed.
[0143] The processing circuit does not need to be specialized solely for executing the steps, functions, procedures, and / or blocks described above; it may perform other tasks.
[0144] Furthermore, this technology also provides a computer program product consisting of computer-readable media 220 and 230 on which such computer programs are stored.
[0145] For example, software or computer programs 225, 235 can be implemented as computer program products, which are typically transported or stored on computer-readable media 220, 230, particularly non-volatile media. Computer-readable media may include, but are not limited to, one or more removable or non-removable memory devices, including read-only memory (ROM), random-access memory (RAM), compact discs (CDs), digital versatile discs (DVDs), Blu-ray discs, universal serial bus (USB) memory, hard disk drives (HDDs), flash memory, magnetic tape, or any other conventional memory devices. Thus, computer programs can be loaded into the operating memory of a computer or equivalent processing device for execution by its processing circuits.
[0146] Therefore, a computer program residing in memory can be organized as a suitable functional module configured to perform at least some of the steps and / or tasks described herein when executed by a processor.
[0147] As described above, at least some of the steps, functions, procedures, and / or blocks described herein can be implemented in software, such as a computer program, for execution by a suitable processing circuit, such as one or more processors or processing units.
[0148] A processing procedure can be considered a computer action flow if it is executed by one or more processors. Corresponding devices, systems, and / or apparatus can be defined as a group of functional modules, where each step executed by a processor corresponds to a functional module. In this case, the functional modules are implemented as computer programs executed on the processors. Therefore, devices, systems, and / or apparatus may alternatively be defined as a group of functional modules, where the functional modules are implemented as computer programs executed on at least one processor.
[0149] Therefore, a computer program residing in memory can be organized as a suitable functional module configured to perform at least some of the steps and / or tasks described herein when executed by a processor.
[0150] Alternatively, the majority of the modules can be implemented using hardware modules, or even replaced entirely with hardware. The choice between software and hardware is purely an implementation choice.
[0151] Where used herein, an element or step described in the singular and preceded by the word "a" or "an" should be understood not to exclude the plural form of the element or step unless such exclusion is expressly stated. Furthermore, references to “one embodiment” in this disclosure are not intended to be construed as excluding the existence of additional embodiments that also incorporate the mentioned features. Furthermore, unless the opposite is expressly stated, an embodiment that “includes,” “incorporates,” or “has” an element or multiple elements having a particular characteristic may include additional such elements that do not have that characteristic. The terms “includes” and “has” are used as plain equivalents to the terms “incorporates” and “possesses,” respectively. Furthermore, terms such as “first,” “second,” and “third” are used simply as labels and are not intended to impose numerical requirements or a particular positional order on their subject matter.
[0152] The embodiments of the present disclosure shown in the drawings and described above are illustrative embodiments only and are not intended to limit the scope of the appended claims, including equivalents as those included in the appended claims. It will be understood by those skilled in the art that various modifications, combinations, and changes can be made to the embodiments without departing from the scope as defined by the appended claims. Any combination of the non-exclusive features described herein is intended to be within the scope of the invention. That is, features of the described embodiments can be combined with any suitable embodiment described above, and any feature of any one embodiment can be combined with any other suitable embodiment. Similarly, features described in dependent claims can be combined with non-exclusive features of other dependent claims, particularly when the dependent claim depends on the same independent claim. In some jurisdictions, single-claim dependents are required in practice, and therefore single-claim dependents may be used, but this does not mean that the features of the dependent claims are mutually exclusive.
[0153] Furthermore, it should be noted that the concept of the present invention relates to all possible combinations of features unless otherwise explicitly stated. In particular, different component solutions in different embodiments can be combined in other configurations, where technically possible. [Explanation of symbols]
[0154] 10: X-ray source 11: Gantry 12: Table 13: Isocenter 20: X-ray detector 21, 22: X-ray detector element / pixel / detector module / semiconductor sensor 23: Path 24: Parallel processing circuit 25: Analog processing circuit 26: Path 30: Image processing system 40: Digital processing circuit 41: X-ray control device 42: Gantry control device 43: Table control device 44: Detector control device 45: Memory 50: Computer 30: Image processing system 40: Digital processing circuit 41: X-ray control device 42: Gantry control device 43: Table control device 44: Detector control device 45: Memory 50: Computer 60: Operator console 62: Display 90, 1000: Noise reduction technology 92: Noisy CT image 94: Black box CNN 96: Clean CT image 100: X-ray imaging system 200: System 210: Processor 220: Memory / Computer-readable medium 225, 235: Computer program 230: External memory device / Computer-readable medium 240: Input / output interface 301: Digital-to-analog converter (DAC) 302: Comparator 303: Digital counter 1002: Noisy CT image 1004: CNN 1006: Denoising 1010: Zero bias 1012: Cross-covariance (wij) is zero 1014: Variance component is zero 1016: Structural image 1018: Biased image
Claims
1. A method for denoising spectral CT image data, A method comprising the step of determining a denoising linear estimate of spectral CT image data by maximizing or minimizing a first objective function, wherein at least one parameter of the denoising linear estimate is determined by at least one machine learning system.
2. Determining the denoising linear estimation of spectral CT image data is: The steps include receiving spectral CT image data and The steps include processing the spectral CT image data based on at least one machine learning system so that a matrix W and a vector b are obtained, The steps include forming denoised spectral CT image data a according to linear estimation based on a = WX + b, The method according to claim 1, wherein x is a representation of spectral CT image data comprising at least two spectral components.
3. The method according to claim 2, wherein at least one of the matrix W and the vector b is adjustable to optimize at least one image quality metric of the CT image data by maximizing or minimizing the first objective function.
4. The method according to claim 2, wherein the first objective function is at least one of mean squared error, structural similarity, bias, fine detail fidelity, numerical observer detectability, visual score, and observer performance.
5. The method according to claim 2, wherein the matrix W is a diagonal matrix.
6. The method according to claim 2, wherein the matrix W is a block-diagonal matrix, and the non-zero off-diagonal entries of the matrix W correspond to the cross terms between the at least two spectral components in each pixel of the spectral CT image data.
7. The method according to claim 2, wherein the matrix W is a sparse matrix, and the non-zero elements of the matrix W correspond to pixels of the spectral CT image data that are adjacent to each other.
8. The method according to claim 2, wherein the at least one machine learning system is trained by minimizing at least one of the L1 loss function, L2 loss function, perceptual loss function, and adversarial loss function.
9. The method according to claim 2, wherein the spectral CT image data x includes at least one of a set of sinograms and a set of reconstructed CT images.
10. The method according to claim 2, wherein the at least two spectral components of the spectral CT image data x include at least one monochromatic image data at different monochromatic energies, image data corresponding to different measurement energy levels or energy bins, and different baseline images.
11. The method according to claim 2, wherein at least one of the matrix W and vector b of the denoised spectral CT image data a is adjusted by the end user.
12. The method according to claim 2, wherein the at least one machine learning system includes at least one convolutional neural network (CNN).
13. The method according to claim 12, wherein the at least one convolutional neural network includes at least one of the ResNet architecture, the UNet architecture, and the unroll iterative architecture.
14. The method according to claim 12, wherein the at least one convolutional neural network is trained as a generator of a generative adversarial network (GAN).
15. The method according to claim 12, wherein the at least one convolutional neural network is trained as part of a pair of cycle-matched generative paradoxical networks (GANs).
16. The method according to claim 12, wherein the at least one convolutional neural network is trained on a dataset comprising a plurality of low-noise images having different spatial frequency characteristics for each high-noise image, and is trained to generate low-noise images having different characteristics for each setting of at least one tuning parameter.
17. A CT imaging system, An X-ray source configured to emit X-rays, An X-ray detector configured to generate spectral CT image data, A processor configured to determine a denoising linear estimate of the generated spectral CT image data based on maximizing or minimizing a first objective function, Includes, A CT imaging system, wherein the processor is further configured to determine at least one parameter of the linear estimation by at least one machine learning system.
18. The processor is configured to process the spectral CT image data based on the at least one machine learning system so that a matrix W and a vector b are obtained, and to form denoised spectral CT image data a according to a linear estimation according to a = WX + b. The CT imaging system according to claim 17, wherein x is a representation of spectral CT image data comprising at least two spectral components.
19. The CT imaging system according to claim 18, wherein at least one of the matrix W and vector b is adjustable to enable optimization of at least one image quality index of CT image data based on maximizing or minimizing a second objective function.
20. The CT imaging system according to claim 19, wherein the second objective function is at least one of mean squared error, structural similarity, bias, fine detail fidelity, numerical observer detectability, visual score, and observer performance.
21. The CT imaging apparatus according to claim 18, wherein the matrix W is a diagonal matrix.
22. The CT imaging apparatus according to claim 18, wherein the spectral CT image data includes at least one of a set of sinograms and a set of reconstructed images.
23. The CT imaging system according to claim 18, wherein at least one of the matrix W and vector b of the denoised spectral CT image data a is adjustable by the end user.