Denoising projection data generated by CT scanners

Abstract

A mechanism for generating denoised reference projection data, in which the low-energy projection data and the high-energy projection data are processed using a neural network trained to perform the dual tasks of decomposition and denoising, whereby the neural network directly outputs the reference projection data, replacing existing decomposition and denoising techniques.

Description

[Technical field]

[0001] The present disclosure relates to the field of medical imaging, and more particularly to the field of denoising and spectral material decomposition of projection data produced by CT scanners. [Background technology]

[0002] Computed tomography (CT) scanners are well-established medical imaging devices that use a detector system or detector array to detect the interaction between X-ray radiation energy and irradiated matter to produce medical imaging data.

[0003] In the field of computed tomography imaging, the use of non-traditional imaging techniques, such as spectral computed tomography, is becoming increasingly common. CT scanners using spectral computed tomography techniques, for example, can generate images that represent specific types and / or frequencies of particles incident on the detector system of the CT scanner, depending on the modality of the CT scanner. These images are often referred to as spectral images and / or material images, and can be derived from a set of so-called reference images.

[0004] One of the recent developments in the field of CT scanners is the dual layer CT scanner. Dual layer CT scanners use a detector system with a first layer of detectors for collecting low energy data and a second layer of detectors for collecting high energy data, where the first and second layers are stacked. The low energy data and the high energy data are collected simultaneously. Projection spatial decomposition is used to process the low energy data and the high energy data to generate reference projection data. Typically, the reference projection data consists of photoelectric effect projection data and Compton scatter projection data. The reference projection data can be reconstructed to form a reference image, which can then be further processed to generate various spectral images. Summary of the Invention [Problem to be solved by the invention]

[0005] However, existing decomposition techniques operate pixel by pixel. This means that the decomposition of low and high energy data into reference projection data is very sensitive to noise. The conventional approach to reduce this noise is to perform pre-decomposition denoising (PDDN), which is often seen as an essential part of generating spectral images for using dual-layer CT scanners. Due to the sensitivity of the decomposition approach to noise, many further denoising steps are often required after the decomposition, e.g., performing further denoising of the reference projection data and / or further denoising in the image domain.

[0006] There is currently a desire to improve the quality of images produced by CT scanners. [Means for solving the problem]

[0007] The invention is defined by the claims.

[0008] According to an example according to an aspect of the present invention, a computer-implemented method for generating denoised reference projection data is provided.

[0009] The computer-implemented method comprises: acquiring a low-energy projection data set having low-energy projection data generated for a plurality of different views of an imaged subject, the low-energy projection data being generated by a first dual-layer computed tomography (CT) scanner using a first layer of detectors; acquiring a high-energy projection data set having high-energy projection data generated for a same plurality of different views of the imaged subject, the high-energy projection data being generated by the first dual layer CT scanner using a second layer of detectors; processing the low-energy projection data set and the high-energy projection data set using a neural network trained to perform denoising and material decomposition of the low-energy projection data set and the high-energy projection data set to generate a plurality of sets of denoised reference projection data, each set providing a different type of spectral reference projection data in one of a plurality of different views of the imaged object; has.

[0010] The proposed approach uses a neural network to perform both denoising and decomposition. In particular, the neural network is able to perform a decomposition that is not performed pixel-by-pixel (of the detector response). This provides a more stable material decomposition process and improves the signal-to-noise ratio of the denoised reference projection data, i.e. reduces noise.

[0011] Embodiments provide several advantages, including the advantage that pre-processing denoising with PDDN no longer needs to be performed, as the neural network decomposition approach is inherently robust to noise (i.e., it performs the denoising itself), which significantly reduces the amount of processing that needs to be performed on the projection data and preserves features that may otherwise be mistaken for noise.

[0012] Furthermore, since neural networks have a receptive field significantly larger than a single pixel, the proposed approach provides a decomposition technique that is stabilized by considering neighboring detector pixels. The use of different views of the imaged subject provides more robust noise removal, especially for non-fixed pattern noise (i.e., noise that is not due to fixed sensor noise).

[0013] The proposed approach improves the denoising of the reference projection data, which means that low-dose CT scanning techniques can be used reliably, since the additional noise inherently present from low-dose CT scanning techniques can be mitigated using the improved denoising techniques.

[0014] Thus, in some preferred examples, the low-energy projection data set and the high-energy projection data set are generated by a CT scanner operating in a low-dose mode of operation, in other words, the low-energy projection data set may be a low-dose, low-energy projection data set and the high-energy projection data set may be a low-dose, high-energy projection data set.

[0015] The multiple sets of denoised reference projection data can include a set including photoelectric effect projection data and a set including Compton scatter projection data.

[0016] The multiple sets of denoised reference projection data may include only a set including photoelectric effect projection data and a set including Compton scatter projection data.

[0017] The multiple sets of denoised reference projection data may include a first set including denoised reference projection data in a first subset of one or more views of the imaged object and a second set including denoised reference projection data in a second subset of the one or more views of the imaged object, wherein the first subset of views is different from the second subset of views.

[0018] Optionally, the first set includes photoelectric effect projection data and the second set includes Compton scatter projection data.

[0019] In a preferred example, the first subset of views or the second subset of views includes the most central view of the multiple different views. This approach ensures that the low frequency components of other parts of the low-energy or high-energy projection data correspond more closely to those of the output projection data. This improves the accuracy of the noise removal since the processed projection data corresponds more closely to the target view of the projection data. The other of the first subset of views or the second subset of views may include a view directly adjacent to the most central view of the multiple different views. This provides a similar improvement effect as the other of the first / second subset of views.

[0020] In some examples, each subset of the one or more views may include only a single view, and one of the first and second subsets of views may include only a most central view, and the other of the first and second subsets of views may include only views adjacent to the most central view.

[0021] In some examples, the neural network is trained using a minimization approach that utilizes a regularization term. The use of the regularization term can reduce the amount of anti-correlation when generating the denoised reference projection data. The regularization term can be incorporated, for example, into a loss function used to determine the error between the predicted output and the benchmark output provided by the training data set.

[0022] In some examples, the neural network is trained using a first training data set.

[0023] The first training data set is a first exemplary dual low-energy projection data set having first low-energy projection data generated for a plurality of views of a sample imaging subject by a first layer of a detector of a second dual layer CT scanner operating in a low-dose mode; a first exemplary high-energy projection data set having first high-energy projection data generated for a plurality of views of the sample imaging subject by a second layer of detectors of the second dual layer CT scanner operating in a low-dose mode; a first input training data set formed from a plurality of first input training data entries, each having a

[0024] The first training data set also includes a first output training data set formed from a plurality of first output training data entries, each first output training data entry corresponding to a respective first input training data entry; a plurality of first exemplary sets of denoised reference projection data, each of the first exemplary sets of denoised reference projection data being decomposed using a material decomposition process, a second exemplary low-energy projection data set having second low-energy projection data generated for a first subset of views of the sample imaging subject by a first layer of detectors of the second dual layer CT scanner operating in a high-dose mode; a second exemplary high-energy projection data set having second high-energy projection data generated for a second subset of views of the sample imaging subject by a second layer of detectors of the second dual layer CT scanner operating in a high-dose mode; a plurality of first example sets generated by processing the a first output training data set having has.

[0025] In another example, the neural network is trained using a second training data set.

[0026] The second training data set is a second output training data set formed from a plurality of second output training data entries, each second output training data entry having a plurality of second exemplary sets of projection data, each second exemplary set of projection data being decomposed using a material decomposition process by: a third exemplary dual low-energy projection data set having third low-energy projection data generated for a plurality of views of the sample imaging subject by a first layer of detectors of a second dual layer CT scanner; a third exemplary high-energy projection data set having third high-energy projection data generated for a plurality of views of the sample imaging subject by a second layer of a detector of the second dual layer CT scanner, each second exemplary set of projection data including a subset of the plurality of views of the sample imaging subject; A second output training dataset, generated by processing has.

[0027] The second training data set also a second input training data set formed from a plurality of second input training data entries, each second input training data entry corresponding to a respective second output training data entry; a fourth exemplary low-energy projection data set having fourth low-energy projection data generated by applying noise to the third low-energy projection data; a fourth exemplary high-energy projection data set having fourth high-energy projection data generated by applying noise to the third high-energy projection data; A second input training data set having has.

[0028] In some examples, the first subset of views and / or the second subset of views include a most central view of the multiple views, e.g., the first and / or second subset may include only a most central view of the multiple views.

[0029] The first single view may be different from the second single view. Of course, in this example, the first single view or the second single view is the most central view of the multiple views. The other of the first single view or the second single view may be directly adjacent to the most central view of the multiple different views.

[0030] It is also proposed a computer program product comprising computer program code means which, when executed on a computing device having a processing system, causes the processing system to perform all of the steps of any of the methods described herein. The computer program product is non-transitory.

[0031] A processing system configured to generate denoised reference projection data for a first dual layer CT scanner that generates low-energy projection data using a first layer of detectors and high-energy projection data using a second layer of detectors is also proposed.

[0032] 1. An apparatus configured to generate denoised reference projection data, the apparatus comprising: A processing circuit; When executed by the processing circuitry, the processing circuitry acquiring a low-energy projection data set having low-energy projection data generated for a plurality of different views of an imaged subject, the low-energy projection data being generated by a first dual-layer CT scanner using a first layer of detectors; acquiring a high-energy projection data set having high-energy projection data generated for a same plurality of different views of the imaged object, the high-energy projection data being generated by the first dual layer CT scanner using a second layer of detectors; processing the low-energy projection data set and the high-energy projection data set using a neural network trained to perform denoising and material decomposition of the low-energy projection data set and the high-energy projection data set to generate a plurality of sets of denoised reference projection data, each set providing a different type of spectral reference projection data in one of a plurality of different views of the imaged subject; A memory including instructions configured to cause the An apparatus is provided having the following:

[0033] For a better understanding of the present invention and to show more clearly how it may be carried into effect, reference will now be made, by way of example only, to the accompanying drawings in which: [Brief description of the drawings]

[0034] [Figure 1] 1 illustrates an imaging system in which embodiments may be implemented; [Diagram 2] 1 illustrates an approach taken by an embodiment. [Diagram 3] 1 is a flow chart illustrating a method according to one embodiment. [Figure 4] 1 illustrates an apparatus according to one embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0035] The present invention will now be described with reference to the drawings.

[0036] It should be understood that the detailed description and specific examples, while indicating exemplary embodiments of the devices, systems, and methods, are for purposes of illustration only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the devices, systems, and methods of the present invention will become better understood from the following description, appended claims, and accompanying drawings. It should be understood that the figures are merely schematic and are not drawn to scale. It should also be understood that the same reference numerals are used throughout the figures to denote the same or similar parts.

[0037] The present disclosure provides a mechanism for generating denoised reference projection data, where the low-energy projection data and the high-energy projection data are processed using a neural network trained to perform the dual tasks of decomposition and denoising, whereby the neural network directly outputs the reference projection data, replacing existing decomposition and denoising techniques.

[0038] The embodiments are based on the realization that neural networks can take into account information from different parts of the projection data when performing the decomposition task, compared to existing approaches that operate pixel by pixel. This means that the neural network can simultaneously perform denoising operations. Furthermore, different views of the projection data can be provided as input channels to the neural network, so that it can generate projection data at a target view with noise reduced using information from the other views.

[0039] Any of the approaches described herein can be used to process projection data generated by any suitable dual layer CT scanner.

[0040] 1 illustrates an imaging system 100, in particular a computed tomography (CT) imaging system, in which embodiments of the present invention may be used. The CT imaging system includes a dual layer CT scanner 101 and a processing interface 111 for processing and performing operations using data generated by the CT scanner 101.

[0041] The CT scanner 101 generally includes a stationary gantry 102 and a rotating gantry 104. The rotating gantry 104 is rotatably supported by the stationary gantry 102 and rotates about an examination region 106 about a longitudinal or z-axis.

[0042] A patient support 120, such as a couch, supports an object, such as a human patient, in the examination region 106. The support 120 is configured to move the object or subject for loading, scanning, and / or removing the object or subject.

[0043] A radiation source 108, such as an x-ray tube, is rotatably supported by the rotating gantry 104. The radiation source 108 rotates with the rotating gantry 104 and emits x-ray radiation that traverses an examination region 106.

[0044] A radiation sensitive detector array 110 subtends an angular arc across the examination region 106 opposite the radiation source 108. The detector array 110 includes two or more rows of detectors extending along a z-axis direction to detect radiation traversing the examination region 106 and generate projection data indicative thereof. The projection data may be, for example, cone beam projection data. The projection data is data of a projection region.

[0045] Since the CT scanner is a dual layer CT scanner, the detector array 110 comprises a first layer of detectors for collecting low energy projection data and a second layer of detectors for collecting high energy projection data, the first and second layers being stacked. The first layer of detectors is closer to the center of the CT scanner, i.e., the radiation source, than the second layer of detectors. The low energy projection data and the high energy projection data are collected simultaneously. For a more complete description of an example of this technique for a dual layer CT scanner, see Rassouli, N., Etesami, M., Dhanantwari, A. et al., "Detector-Based Spectral CT Using a Novel Dual Layer Technique," Principles and Applications, Insights Imaging 8, 589-598 (2017).

[0046] Other approaches and detector arrays for collecting low-energy and high-energy projection data will be apparent to those skilled in the art and may include any form of detector array capable of generating projection data for different energy spectra (i.e., for use in any suitable spectral CT scanner).

[0047] As the rotating gantry rotates 104, the direction or angle that the radiation source 108 makes with respect to the stationary gantry 102 changes, which in turn changes the angle that the radiation source 108 makes with respect to the examination region 106 and the imaged object disposed therein. This angle is known as the imaging angle. The detector array 110 thereby generates low and high energy projection data for each of a number of different angles that the radiation source makes with respect to the examination region. Each angle represents a different view of the imaged object in the examination region. Low and high projection data are thereby generated for a number of different (projection) views of the imaged object.

[0048] In the context of this application, sets of projection (domain) data having different views indicates that the projection data represent different regions and / or imaging angles of the object, and sets of projection (domain) data having the same view indicates that the sets of projection data represent the same region and imaging angle of the object.

[0049] The processing interface 111 processes the projection data generated by the CT scanner and may also facilitate user control over the operation of the CT scanner 102, such as controlling or defining regions or areas to undergo CT imaging and / or controlling the imaging dose or intensity.

[0050] In particular, the processing interface serves as an operator console 112 and may comprise a general-purpose computing system or computer including input devices 114, such as a mouse, keyboard, and / or the like, and output devices 116, such as a display monitor, filmer, or the like. The console 112 allows an operator to control the operation of the system 100. Data generated by the processing interface may be displayed through at least one display monitor of the output devices 116.

[0051] The processing interface processes the projection data and utilizes a reconstructor 118 to reconstruct image data, for example, for display on an output device 116 or for storage in an external server or database. It should be appreciated that the reconstructor 118 may be implemented via a microprocessor executing computer readable instructions encoded or embedded on a computer readable storage medium, such as physical memory and other non-transitory media. Additionally or alternatively, the microprocessor may execute computer readable instructions carried by carrier waves, signals, and other transitory or non-transitory media.

[0052] In a particular example, the reconstruction unit 118 may first perform a projection space decomposition to generate reference projection data, which may be alternatively labeled reference projection region data. For example, a projection space decomposition may be performed to generate photoelectric effect projection data and / or Compton scatter projection data by appropriate processing of low-energy projection data and high-energy projection data. Both photoelectric effect projection data and Compton scatter projection data thereby provide examples of reference projection data.

[0053] More specifically, the reference projection data is projection data that is responsive to a particular material type within the imaged region of interest or to the energy spectrum / level of particles passing through the scanned / imaged region of interest during the CT scanning process. Different types of reference projection data are responsive to different types of materials or energy spectra. Examples of different energy spectra include energy spectra associated with particles generated via the photoelectric effect or via Compton scattering. Thus, reference projection data may include, for example, photoelectric effect projection data or Compton scattering projection data. This is because particles generated as a result of different effects have different energies, i.e., fall into different energy spectra. Examples of different materials include water, iodine, contrast agents, etc. These can be identified and differentiated from one another because different materials have different absorption characteristics. Thus, other examples of reference projection data include water projection data, iodine projection data, contrast agent projection data, etc.

[0054] Suitable projection space decomposition techniques, i.e., basis functions, for generating such reference projection data from the low-energy and high-energy projection data will be readily apparent to those skilled in the art. Typically, but not necessarily, the reference projection data is subjected to some filtering or denoising process to generate denoised reference projection data.

[0055] The reference projection data or the denoised reference projection data may then undergo reconstruction (e.g., spectral reconstruction) to generate reference image data. In existing approaches, the reconstructor 118 uses reduced noise reconstruction algorithms such as filtered backprojection (FBP) reconstruction, iterative reconstruction algorithms that may operate in the image domain and / or the projection domain, and / or other algorithms.

[0056] Similar to the reference projection data, the different types of reference image data provide image data representing different types of materials and / or energy spectra in the scanned / imaged region of interest during the CT scanning process. By way of example, the different types of reference image data include photoelectric effect reference image data, Compton scatter reference image data, iodine image data, water image data, contrast agent image data, etc.

[0057] The reference image data may be subsequently processed to generate a number of diagnostically relevant images, such as Mono-energetic (MonoE) images, iodine maps, Z-effective images, and virtual non-contrast images, which may be performed by the reconstructor 118 or a separate processing system.

[0058] Of course, the reconstruction device may also be used to generate single energy (SECT) or combined image data, where the combined image data is generated from the combined projection data, where the combined projection data is generated by combining the high-energy and low-energy projection data together (e.g., summing or averaging the high-energy and low-energy projection data).

[0059] The present disclosure relates to the generation of reference projection data from low-energy projection data and high-energy projection data, i.e., performing a decomposition process. In particular, the proposed embodiment provides techniques for improved stability of the decomposition process, i.e., generation of reference projection data, and improved signal-to-noise of the generated reference projection data. More specifically, the proposed technique can replace existing decomposition techniques used to generate the reference projection data, as well as any denoising techniques for the reference projection data.

[0060] The embodiments thereby provide techniques for improving the reconstruction apparatus of a computed tomography system. By improving the signal-to-noise ratio of the reference projection data, a lower dose of radiation can be used to generate reference projection data of the same quality as was previously available. This advantageously facilitates a reduction in the exposure of the subject to radiation.

[0061] FIG. 2 conceptually illustrates an exemplary approach according to one embodiment.

[0062] The approach proposes to process an input data set 210, consisting of a low-energy projection data set 211 and a high-energy projection data set 212, using a neural network 220 to generate multiple sets 230 or instances 231, 232 of degraded reference projection data 230. For example, the denoised reference projection data may include a set of photoelectric effect projection data 231 and / or a set of Compton scatter projection data 232.

[0063] The low-energy projection data set 211 comprises projection data generated from a first layer of detectors of the CT scanner, i.e. low-energy projection data. The high-energy projection data set 212 comprises projection data generated from a second layer of detectors of the CT scanner, i.e. high-energy projection data. The first and second layers of detectors are stacked such that the first layer of detectors is closer to the center of the CT scanner, i.e. the radiation source, than the second layer of detectors.

[0064] The low-energy projection data includes low-energy projection data acquired from multiple views, i.e. at different orientations of the radiation source relative to the examination region, and the high-energy projection data includes high-energy projection data obtained from the same multiple views.

[0065] The neural network is suitably trained to generate reference projection data, i.e., projection region reference data. The reference projection data 230 includes at least two sets of denoised reference projection data, i.e., at least two instances of denoised reference projection data, with different instances / sets corresponding to different types of reference projection data. Thus, the neural network can generate at least two types of denoised reference projection data as output.

[0066] Suitable types include photoelectric effect projection data and Compton scatter projection data. Other suitable types include water projection data, iodine projection data, contrast agent projection data, and other material-based projection data. These latter types can use projection data at different energy levels to identify or discern specific absorption characteristics unique to particular materials, and thus provide projection data that is directed or tailored to particular materials.

[0067] Although only two sets of denoised reference projection data are shown, it will be understood that more than two sets of denoised reference projection data may be output by the neural network.

[0068] Each set provides or includes a different type of spectral reference projection data in one of a plurality of different views, with low-energy and high-energy projection data sets, in other words, each set of denoised reference projection data representing a particular target view of the object.

[0069] The denoised reference projection data 230 includes multiple sets of denoised reference projection data, where each set may include denoised reference projection data for a single view as compared to multiple views included in the projection data provided as input. In other examples, each set may include denoised reference projection data for a subset, e.g., less than all, of the views included in the input data set.

[0070] The neural network 220 performs a decomposition process and a noise removal procedure on the low-energy and high-energy projection data simultaneously. In particular, it is advantageously recognized that the decomposition performed by such a neural network is inherently robust to noise, since the decomposition is not performed pixel-by-pixel, but rather all projection data is processed simultaneously. This means that noise can be automatically compensated for using information from other parts of the low-energy and high-energy projection data.

[0071] Furthermore, since neural networks have a receptive field significantly larger than a single pixel, the proposed decomposition approach is more robust compared to the approach adopted by the prior art by taking into account neighboring detector pixels.

[0072] The neural network 220 is preferably a convolutional neural network that utilizes one or more convolution operations to process the low-energy projection data 211 and the high-energy projection data 212. This may include ensuring that the convolution is performed in the row and column dimensions of the acquired data, as well as taking into account a given set of adjacent views in different feature channels. This ensures that the receptive field for each element of the denoised reference projection data is larger than a single picture, taking advantage of the multiple dimensions of the input data set to provide more accurate and robust denoised reference projection data.

[0073] This approach utilizes a neural network to process low- and high-energy projection data to generate a set of denoised reference projection data.

[0074] The structure of a neural network is inspired by the human brain. A neural network is composed of layers, each layer containing multiple neurons. Each neuron contains a mathematical operation. In particular, each neuron may contain a single type of transformation, e.g., different weighted combinations of the same type of transformation, such as sigmoid, but with different weightings. In the process of processing input data, the mathematical operation of each neuron is performed on the input data to generate a numerical output, and the output of each layer in the neural network is sequentially fed to the next layer. The final layer provides the output.

[0075] Methods for training neural networks are well known. Typically, such methods include obtaining a training data set, including training input data entries and corresponding training output data entries. The initialized neural network is applied to each training input data entry to generate a predicted output data entry. The error generated by the loss or cost function between the predicted output data entry and the corresponding training output data entry is used to modify the machine learning algorithm. Exemplary loss functions include mean squared error (MSE) or L1 error, etc. This process can be repeated until the error converges, and the predicted output data entry is sufficiently similar to the training output data entry, e.g., ±1%. This is commonly known as a supervised learning technique.

[0076] For example, the weighting of the mathematical operations of each neuron may be modified until the error converges. Known methods for modifying neural networks include gradient descent, backpropagation algorithms, etc.

[0077] Thus, the neural networks used in the embodiments disclosed herein may be trained using a training data set, examples of which are particularly advantageous, though not required, are described below.

[0078] The training data set may comprise an input training data set formed from a plurality of input training data entries, and an output training data set formed from a plurality of output training data entries.

[0079] The output training data set may include a plurality of output training data entries. Each output training data entry may comprise a plurality of sets of exemplary reference projection data. Each set of reference projection data is generated from a first low-energy projection data and a first high-energy projection data. The first low-energy projection data is generated by a first layer of a detector of a second dual layer CT scanner operating in a high-dose mode. The first high-energy projection data is generated by the same second layer of detectors of the second dual layer CT scanner operating in a high-dose mode.

[0080] Generating denoised reference projection data for output training data entries may include processing this low-energy and high-energy projection data acquired in a high-dose mode of the CT scanner using one or more standard decomposition techniques, and optionally performing noise filtering on the input and / or output of the decomposition techniques.

[0081] Because high dose projection data inherently has relatively low noise due to the relatively large amounts of radiation used to detect / generate the projection data, using high dose projection data to generate reference projection data for output training data entries generates reference projection data having low noise.

[0082] The input training data set may include multiple input training data entries, each of which corresponds to a different output training data entry, e.g., represents the same imaging region of the same subject as the corresponding output training data entry. Different sets of input and output training entries may correspond to different regions and / or different subjects. In fact, this is preferred to avoid or reduce the risk of overfitting the neural network.

[0083] Each input training data entry includes a second low-energy projection data and a second high-energy projection data.

[0084] In a first example, the second low-energy projection data is generated by a first layer of detectors of a second dual layer CT scanner operating in a low-dose mode. In this first example, the second high-energy projection data is generated by a second layer of detectors of a second dual layer CT scanner operating in a low-dose mode. The second low-energy and high-energy projection data are spatially corresponding, i.e., represent the same region / area of ​​the same subject as the first low-energy and high-energy projection data.

[0085] In a second example, the second low-energy projection data is generated by processing the first low-energy projection data used in generating the exemplary reference projection data of the corresponding output training data entry. In this second example, the second high-energy projection data is generated by processing the first high-energy projection data used in generating the exemplary reference projection data of the corresponding output training data entry. The processing may include adding noise, e.g., Gaussian noise, to the first low-energy / high-energy projection data. This may effectively synthesize or simulate low-energy or high-energy projection data accordingly that would have been generated by a second CT scanner using a low-dose mode.

[0086] The first example is useful for generating neural networks that more closely reflect the real noise in low-dose data, providing a more accurate denoising neural network for use in denoising low-dose data in particular.

[0087] A second example is beneficial by exposing the sampled imaging subject to less radiation when collecting the training data set, because instead of two imaging processes, a high-dose and a low-dose process, only a single imaging process, a single high-dose process, needs to be performed, thereby exposing the subject to less radiation.

[0088] Furthermore, it is recognized herein that even when high dose projection data is used in generating the training output data entries, anti-correlated noise is likely to be present in the training output data entries.

[0089] One approach to reduce the amount of anti-correlated noise is to perform high intensity filtering of the high dose projection data in the training output data entries, thereby reducing the effect of noise in the training output data entries. However, this approach is not recommended as it results in undesirable training of shortcomings of the current data processing chain into the neural network. Therefore, an alternative approach that does not rely on high intensity filtering would be beneficial.

[0090] One approach is to modify the loss function used when training the neural network to include an additional regularization term. One option for this approach is to directly penalize anti-correlation in the neural network output due to scattering and absorption. Another option is to use the same regularization term as used in the anti-correlation filter (ACF) of known projection data processing algorithms, such as that proposed in European Patent No. EP 3,039,650.

[0091] The second technique utilizes a neural network setup to process multiple views of low- and high-energy projection data to generate denoised reference projection data.

[0092] It has been previously described how the neural network used in the disclosed examples is configured to receive as input low and high energy projection data generated for multiple views, whereas the output of the neural network may include two or more sets of denoised reference projection data, each set corresponding to a different type of reference projection data, e.g. photoelectric effect projection data or Compton scatter projection data, in only a single view for each set.

[0093] In the second technique, it is proposed to make the view associated with each set of denoised reference projection data different. In this way, each set includes denoised reference projection data for a different subset of views, for example different views. Thus, one of the sets of denoised reference projection data is associated with a first subset of views, which may be a single first view. The other is associated with a second different subset of views, for example a single second view. When constructing a training data set using this approach, anti-correlated noise is absent or reduced in the output training data entries.

[0094] To improve consistency, this approach should be adopted in both the training and inference stages of a neural network.

[0095] In a preferred example, when each subset of views includes only a single view, the first view or the second view may be adjacent to each other. Thus, if the first view is view i, the second view may be view i+1 or view i-1. For a typical CT scan, directly adjacent views are expected to have very similar low-frequency content, which makes the joint prediction of adjacent views advantageous, i.e. more accurate or robust, from the perspective of network training, as opposed to predicting two views with a larger shift.

[0096] In some examples, the first view or the second view may be the most central view of the available views in the projection data received as input by the neural network. Thus, if the multiple views included in the low-energy or high-energy projection data range from j-n to j+n, the first view or the second view may be j. This advantageously ensures that the content of the projection data in the low-energy or high-energy projection data is more likely to be similar to the content in the denoised reference projection data output by the neural network, at least in terms of low-frequency content.

[0097] The previously described example utilizes a neural network that processes low-energy and high-energy projection data, these two types of projection data being generated using two stacked detector layers of a dual-layer CT scanner.

[0098] In an extension of this approach, a dual layer CT scanner may be configured to operate in a kVp switching mode in which the radiation source of the CT scanner emits two distinct / different frequencies or frequency spectrums of x-ray radiation, which may be alternated or sequential.

[0099] In these scenarios, the low-energy projection data may include first low-energy projection data generated when a first frequency or frequency spectrum of X-ray radiation is output by the radiation source, and second low-energy projection data generated when a second frequency or frequency spectrum of X-ray radiation is output by the radiation source.

[0100] Similarly, the high-energy projection data may include first high-energy projection data generated when a first frequency or frequency spectrum of X-ray radiation is output by the radiation source, and second high-energy projection data generated when a second frequency or frequency spectrum of X-ray radiation is output by the radiation source.

[0101] Thus, the neural network would receive four sections of projection data for each view instead of just two, and the neural network could still be properly trained to generate a set of denoised reference projection data in the same manner.

[0102] The data corresponding to two different kVp settings contain complementary information about the absorption lengths at different energies, which can be exploited when combining the information contained in these input channels to predict the denoised reference projection data, thus improving the denoising ability of the neural network.

[0103] 3 illustrates a method 300 according to one embodiment that utilizes the techniques described above. The method may be computer implemented and executed by a processing system communicatively coupled to the CT scanner and / or a memory / storage unit.

[0104] The method 300 is configured to denoise reference projection data of a first dual-layer CT scanner that produces low-energy projection data using a first layer of detectors and high-energy projection data using a second layer of detectors.

[0105] The method 300 includes acquiring 310 a low-energy projection data set 315 that includes low-energy projection data generated for a number of different views of the imaged object. The data set can be obtained, for example, from a CT scanner or a memory / storage unit.

[0106] The method 300 further includes a step 320 of acquiring a high-energy projection dataset 325, which includes high-energy projection data generated for the same multiple different views of the imaged object. This dataset may also be obtained, for example, from a CT scanner or a memory / storage unit.

[0107] The method 300 further includes a step 330 of processing the low-energy projection data set and the high-energy projection data set using a neural network. The neural network is trained to perform denoising and material decomposition of the low-energy projection data set and the high-energy projection data set. The step 330 thereby generates a plurality of sets 335 of denoised reference projection data 331, 332, each set providing a different type of spectral reference projection data in one of a plurality of different views of the imaged subject. For example, the denoised reference projection data 330 may include a first set 331 of denoised reference projection data, e.g., photoelectric effect projection data 331, and a second set 332 of denoised reference projection data, e.g., Compton scatter projection data 332.

[0108] The method 300 may further include, for example, outputting 340 a set of denoised reference projection data for further processing or reconstruction by a reconstruction device. In particular, each output set of denoised reference projection data may be reconstructed into one or more reference images.

[0109] 4 illustrates an example of an apparatus 40, or processing system, in which one or more portions of the embodiments may be employed. The various operations described above may utilize the capabilities of the apparatus 40. For example, one or more portions of the system for generating a set of denoised reference projection data may be incorporated into any of the elements, modules, applications and / or components described herein. In this regard, it should be understood that the system functional blocks may be executed on a single computer or may be distributed across several computers and locations, e.g., connected via the Internet.

[0110] The device 40 may include, but is not limited to, a PC, a workstation, a laptop, a PDA, a palm device, a server, a storage, a cloud computing device, a distributed processing system, and the like. In general, with respect to a hardware architecture, the device 40 may include one or more processing circuits 41, a memory 42, and one or more I / O devices 43 communicatively coupled via a local interface (not shown). The local interface may be, for example, but not limited to, one or more buses or other wired or wireless connections, as known in the art. The local interface may have additional elements, such as controllers, buffers (caches), drivers, repeaters, and receivers, to enable communication. Additionally, the local interface may include address, control, and / or data connections to enable appropriate communication between the aforementioned components.

[0111] The processing circuitry 41 is a hardware device for executing software that may be stored in the memory 42. The processing circuitry 41 may be substantially any custom or commercially available processing circuitry, a central processing unit (CPU), a digital signal processor (DSP), or auxiliary processing circuitry of any of a number of processing systems associated with the device 40, and the processing circuitry 41 may be a semiconductor-based microprocessor circuit (in the form of a microchip) or a microprocessor circuit.

[0112] The memory 42 may include any one or combination of volatile memory elements, random access memory (RAM), such as dynamic random access memory (DRAM), static random access memory (SRAM), and non-volatile memory elements, ROM, erasable programmable read only memory (EPROM), electronically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), tape, compact hard disk read only memory (CD-ROM), hard disk, diskette, cartridge, cassette, etc. Additionally, the memory 42 may incorporate electronic, magnetic, optical, and / or other types of storage media. The memory 42 may have a distributed architecture in which various components are located remotely from each other, but may be accessed by the processing circuitry 41.

[0113] The software in memory 42 may include one or more separate programs, each of which comprises an ordered list of executable instructions for implementing logical functions. The software in memory 42 includes, according to an exemplary embodiment, a suitable operating system (O / S) 44, a compiler 45, source code 46, and one or more applications 47. As shown, the applications 47 comprise a number of functional components for implementing the features and operations of the exemplary embodiments. The applications 47 of device 40 may represent various applications, computational units, logic, functional units, processes, operations, virtual entities, and / or modules according to an exemplary embodiment, although application 47 is not meant to be limiting.

[0114] Operating system 44 controls the execution of other computer programs and provides scheduling, input / output control, file and data management, memory management, and communication control and related services. It is contemplated by the inventors that applications 47 for implementing exemplary embodiments may be applicable to all commercially available operating systems.

[0115] The application 47 may be a source program, an executable program (object code), a script, or any other entity comprising a set of instructions to be executed. In the case of a source program, the program is typically translated via a compiler, such as a compiler 45, assembler, interpreter, etc., which may or may not be included in the memory 42, so that the program operates properly in conjunction with the O / S 44. Furthermore, the application 47 may be written as an object-oriented programming language with classes of data and methods, with routines, subroutines, and / or functions, or a procedural programming language, such as, but not limited to, C, C++, C#, Pascal, BASIC, API calls, HTML, XHTML, XML, ASP script, JavaScript, FORTRAN, COBOL, Perl, Java, ADA, NET, etc.

[0116] The I / O devices 43 may include input devices such as, but not limited to, a mouse, keyboard, scanner, microphone, camera, etc. Additionally, the I / O devices 43 may also include output devices such as, but not limited to, a printer, display, etc. Finally, the I / O devices 43 may further include devices that communicate both input and output, such as, but not limited to, network interface cards or modulators / demodulators (for accessing remote devices, other files, devices, systems, or networks), radio frequency (RF) or other transceivers, telephone interfaces, bridges, routers, etc. The I / O devices 43 also include components for communicating over various networks, such as the Internet or an intranet.

[0117] If device 40 is a PC, workstation, intelligent device, etc., the software in memory 42 may further include a standard input / output system (BIOS). The BIOS is a set of required software routines that initializes and tests the hardware at power-up, starts the O / S 44, and supports data transfers between hardware devices. The BIOS may be stored in some type of read-only memory, such as a ROM, PROM, EPROM, EEPROM, etc., and may be executed when device 40 is powered on.

[0118] When device 40 is in operation, processing circuitry 41 is configured to execute software stored in memory 42, to communicate data to and from memory 42, and to generally control the operation of device 40 in accordance with the software. Applications 47 and O / S 44 are read, in whole or in part, by processing circuitry 41, possibly buffered within processing circuitry 41, and then executed.

[0119] It should be noted that when the application 47 is implemented in software, the application 47 may be stored on virtually any computer-readable medium for use by or in connection with any computer-related system or method. In the context of this document, a computer-readable medium may be an electronic, magnetic, optical, or other physical device or means that can contain or store a computer program for use by or in connection with a computer-related system or method.

[0120] The application 47 may be implemented on any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a system including processing circuitry, or other system that can fetch instructions from and execute instructions from an instruction execution system, apparatus, or device. In the context of this specification, a "computer-readable medium" may be any means that can store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. A computer-readable medium may be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium.

[0121] In the context of the present disclosure, the I / O device 43 may be configured to receive an input data set from a medical imaging device. In another example, the input data set is obtained from a memory structure or unit.

[0122] The I / O device 43 may be configured to provide the output dataset to a user interface. The user interface may be configured to provide a visual representation of the output dataset(s), for example, displaying image(s) corresponding to the medical imaging data included in the output dataset(s).

[0123] Those skilled in the art will be able to readily develop an apparatus having processing circuitry for carrying out the methods described herein, and each step of the flow chart may thus represent a different operation performed by the processing circuitry of the apparatus, and may be performed by a respective module of the processing circuitry of the apparatus.

[0124] Thus, the embodiments may utilize an apparatus. The apparatus may be implemented in numerous ways using software and / or hardware to perform the various functions required. A processor is one example of an apparatus that uses one or more microprocessors that may be programmed using software (e.g., microcode) to perform the necessary functions. However, the apparatus may be implemented with or without a processor, and may be implemented as a combination of dedicated hardware to perform some functions and a processor, e.g., one or more programmed microprocessors and associated circuitry, to perform other functions.

[0125] Examples of device components that may be used in various embodiments of the present disclosure include, but are not limited to, conventional microprocessors, application specific integrated circuits (ASICs), and field programmable gate arrays (FPGAs).

[0126] In various implementations, a processor or device may be associated with one or more storage media, such as volatile and non-volatile computer memories, such as RAM, PROM, EPROM, and EEPROM. The storage media may be encoded with one or more programs that, when executed on the processing circuitry, perform the necessary functions. The various storage media may be fixed within the processor or device or may be transportable, such that one or more programs stored thereon may be loaded into the processor or device.

[0127] It will be understood that the disclosed methods are preferably computer-implemented methods. Thus, the concept of a computer program comprising code means for implementing any described method when said program is executed on an apparatus including a processing circuit, such as a computer, is also proposed. Thus, different parts, lines or blocks of code of a computer program according to an embodiment may be executed by a device or computer to perform the method described herein. In some alternative implementations, the functions depicted in the block diagrams or flow charts may occur out of the order depicted in the figures. For example, two blocks shown in succession may in fact be executed substantially simultaneously, or may be executed in the reverse order depending on the functionality involved in the blocks.

[0128] Variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain means are recited in mutually different dependent claims does not indicate that a combination of these means cannot be used to advantage. Where a computer program is described above, it may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium, provided together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. It should be noted that when the term "adapted for" is used in the claims or description, the term "adapted for" is intended to be equivalent to the term "configured for". Any reference signs in the claims should not be interpreted as limiting the scope.

Claims

1. 1. A computer-implemented method for generating denoised reference projection data, the method comprising the steps of: acquiring a low-energy projection data set having low-energy projection data generated for a plurality of different views of an imaged subject, the low-energy projection data being generated by a first dual-layer computed tomography (CT) scanner using a first layer of detectors; acquiring a high-energy projection dataset having high-energy projection data generated for a same plurality of different views of the imaged subject, the high-energy projection data being generated by the first dual-layer CT scanner using a second layer of detectors; processing the low-energy projection data set and the high-energy projection data set using a neural network trained to simultaneously perform denoising and material decomposition of the low-energy projection data set and the high-energy projection data set to generate multiple sets of denoised reference projection data, each set providing a different type of spectral reference projection data in a subset of a plurality of different views of the imaged object; 10. A computer-implemented method comprising:

2. The computer-implemented method of claim 1 , wherein the plurality of sets of denoised reference projection data comprises a set having photoelectric effect projection data and a set having Compton scatter projection data.

3. The computer-implemented method of claim 2 , wherein the plurality of sets of denoised reference projection data includes only a set having the photoelectric effect projection data and a set having the Compton scatter projection data.

4. The plurality of sets of denoised reference projection data include: a first set having denoised reference projection data for a first subset of views of the imaged object; a second set having denoised reference projection data in a second subset of views of the imaged object, the first subset of views being different from the second subset of views; 10. The computer-implemented method of claim 1, comprising:

5. 5. The computer-implemented method of claim 4, wherein the first set comprises photoelectric effect projection data and the second set comprises Compton scatter projection data.

6. The computer-implemented method of claim 4 , wherein the first subset of views or the second subset of views comprises a most central view of the plurality of different views.

7. The computer-implemented method of claim 6 , wherein the other of the first subset of views or the second subset of views includes a view that is immediately adjacent to a most central view of the plurality of different views.

8. The computer-implemented method of claim 1 , wherein the neural network is trained using a minimization approach that utilizes a regularization term.

9. The neural network is trained using a first training data set, the first training data set comprising: a first exemplary dual low-energy projection data set having first low-energy projection data generated for a plurality of views of a sample imaging subject by a first layer of detectors of a second dual-layer CT scanner operating in a low-dose mode; a first exemplary high-energy projection data set having first high-energy projection data generated for a plurality of views of the sample imaging subject by a second layer of detectors of the second dual-layer CT scanner operating in a low-dose mode; a first input training data set formed from a plurality of first input training data entries, each having: a first output training data set formed from a plurality of first output training data entries, each first output training data entry corresponding to a respective first input training data entry; a plurality of first exemplary sets of denoised reference projection data, each first exemplary set of denoised reference projection data being decomposed using a material decomposition process, a second exemplary low-energy projection data set having second low-energy projection data generated for a first subset of views of the sample imaging subject by a first layer of detectors of the second dual-layer CT scanner operating in a high-dose mode; a second exemplary high-energy projection data set having second high-energy projection data generated for a second subset of views of the sample imaging subject by a second layer of detectors of the second dual-layer CT scanner operating in a high-dose mode; a plurality of first example sets generated by processing the a first output training dataset having 2. The computer-implemented method of claim 1, comprising:

10. The neural network is trained using a second training data set, the second training data set comprising: a second output training data set formed from a plurality of second output training data entries, each second output training data entry having a plurality of second exemplary sets of projection data, each second exemplary set of projection data being decomposed using a material decomposition process by: a third exemplary dual low-energy projection data set having third low-energy projection data generated for a plurality of views of the sample imaging subject by a first layer of detectors of a second dual layer CT scanner; a third exemplary high-energy projection dataset having third high-energy projection data generated for a plurality of views of the sample imaging subject by a second layer of detectors of the second dual-layer CT scanner, each second exemplary set of projection data comprising a subset of the plurality of views of the sample imaging subject; a second output training dataset generated by processing a second input training data set formed from a plurality of second input training data entries, each second input training data entry corresponding to a respective second output training data entry; a fourth exemplary low-energy projection data set having fourth low-energy projection data generated by applying noise to the third low-energy projection data; a fourth exemplary high-energy projection data set having fourth high-energy projection data generated by applying noise to the third high-energy projection data; a second input training dataset having 2. The computer-implemented method of claim 1, comprising:

11. the first dual layer CT scanner is configured to operate in a kVp switching mode, a radiation source of the first dual layer CT scanner emitting a first frequency or frequency spectrum of X-ray radiation and a second frequency or frequency spectrum of X-ray radiation; the first frequency or frequency spectrum of X-ray radiation is distinguishable from the second frequency or frequency spectrum of X-ray radiation; the low-energy projection data comprising first low-energy projection data generated when a first frequency or frequency spectrum of X-ray radiation is output by the radiation source and second low-energy projection data generated when a second frequency or frequency spectrum of X-ray radiation is output by the radiation source; and / or the high-energy projection data comprising first high-energy projection data generated when a first frequency or frequency spectrum of the X-ray radiation is output by the radiation source, and second high-energy projection data generated when a second frequency or frequency spectrum of the X-ray radiation is output by the radiation source; 10. The computer-implemented method of claim 1.

12. 12. A computer program product comprising computer program code means which, when executed on a computing device having a processing system, causes said processing system to perform all of the steps of the method according to any one of claims 1 to 11.

13. 1. An apparatus configured to generate denoised reference projection data, the apparatus comprising: a processing circuit; When executed by the processing circuitry, the processing circuitry acquiring a low-energy projection data set having low-energy projection data generated for a plurality of different views of an imaged subject, the low-energy projection data being generated by a first dual-layer CT scanner using a first layer of detectors; acquiring a high-energy projection dataset having high-energy projection data generated for a same plurality of different views of the imaged subject, the high-energy projection data being generated by the first dual-layer CT scanner using a second layer of detectors; processing the low-energy projection data set and the high-energy projection data set using a neural network trained to simultaneously perform denoising and material decomposition of the low-energy projection data set and the high-energy projection data set to generate a plurality of sets of denoised reference projection data, each set providing a different type of spectral reference projection data in one of a plurality of different views of the imaged object; a memory containing instructions configured to cause the An apparatus having:

Images