Predicting effective energy values

Abstract

The proposed concepts are thus directed to providing schemes, solutions, concepts, designs, methods and systems relating to training neural network models for predicting effective energy values for conventional CT images. In particular, embodiments are directed to utilizing single energy CT images from a spectral scanner with known effective energy values in order to train a neural network model that can then be used to predict the effective energy values of conventional, i.e., not typically single energy, CT images.

Description

Technical Field

[0001] This invention relates to the field of predicting the effective energy value of conventional CT images, and more particularly to the field of training a neural network model for predicting the effective energy value of conventional CT images. Background Technology

[0002] Dual-energy CT imaging is well-established. The simplest approach involves acquiring two separate scans using two different kVp settings, i.e., at two different photon energies. Material decomposition can then be performed in either the projection or image domain. Typically, a small but non-negligible motion exists between the two scans due to the time delay between them. Registration of both low and high datasets is required, and this registration is best performed in the image domain. After image registration, image-based material decomposition of the CT images can be applied using the now spatially matched images. Image-based material decomposition of CT images is also well-established in this field.

[0003] In the theory of image domain material decomposition, the linear attenuation coefficient of each pixel in low-CT and high-CT images is... It is approximated by a linear combination of pixel values ​​in the image of the base material; here, the photoelectric effect ( ) and Compton effect or scattering ( ): and This represents the effective energy for low and high pixel values. In the case of a single-energy image, the matrix... This holds true for all image pixels. The effective energy depends on the object's attenuation characteristics, and therefore on the variation in each CT image slice. Using... inverse matrix Calculate the image of the base material pixel by pixel based on the low-resolution and high-resolution images: Due to beam hardening, the matrix This depends on the scanner (spectrum, detector, and processing) and the composition, size, and shape of the object. It is assumed that both the low-energy and high-energy images are single-energy and effective / single-energy. and If it is known, then the matrix Given by the linear attenuation coefficient of the base material: However, a major challenge in image-based material decomposition is finding the effective energy of CT images in order to calculate the correct transfer function from a standard input CT image to a material image. In clinical practice, cumbersome system calibration processes from which the effective energy can be found should be avoided, as they slow down workflows and are susceptible to drift.

[0004] Therefore, a solution is needed for predicting effective energy values ​​in conventional CT images. Summary of the Invention

[0005] This invention is defined by the claims.

[0006] According to an example of an aspect of the present invention, a method is provided for training a neural network model for predicting the effective energy value of a conventional CT image.

[0007] The method includes: training a neural network model using a training algorithm configured to receive an array of training inputs and corresponding known outputs, wherein the training inputs include spectral scanner single-energy CT images, and the corresponding known outputs include the effective energy values ​​of the spectral scanner single-energy CT images.

[0008] Therefore, the proposed concept aims to provide schemes, solutions, concepts, designs, methods, and systems related to training neural network models for predicting effective energy values ​​of conventional CT images.

[0009] Specifically, the embodiments aim to utilize single-energy CT images from a spectral scanner with known effective energy values ​​to train a neural network model, which can then be used to predict the effective energy values ​​of conventional, i.e., typically not single-energy, CT images. In other words, the neural network model is trained using single-energy images from a spectral scanner; however, the trained neural network model can then be used to predict the effective energy values ​​of conventional CT images obtained from either a spectral or conventional scanner. By training the neural network model to predict the effective energy values ​​of CT images using only images, a method for estimating the effective energy of CT images can be provided that does not require system calibration and requires no additional system or image information. Even the actual kVp setting information of the scan is not required. The calibration process during clinical practice can then be avoided.

[0010] In other words, a training algorithm is proposed to train a neural network model that receives single-energy CT images from a spectral scanner as training input and the effective energy values ​​of the single-energy CT images as corresponding known outputs. This can ultimately provide a neural network model capable of estimating the effective energy values ​​of conventional (i.e., non-single-energy) CT images solely from the images themselves, thereby allowing for the computation of transfer functions for the images and the implementation of material decomposition. This method also avoids the CT system calibration process, which slows down the workflow and is often subject to drift. Implementing this purely data-driven solution circumvents calibration and improves the ease of obtaining estimated effective energy values ​​from conventional CT images.

[0011] By training a neural network model to predict the effective energy value based on a single input CT image, a method can ultimately be provided to estimate the effective energy value of a conventional CT image simply from the image itself.

[0012] Training a neural network model using single-energy CT images obtained from a spectral scanner with known effective energy values ​​allows the model to become proficient in estimating effective energy values, even from non-single-energy CT images, i.e., conventional CT images. In other words, it has been recognized that a neural network model can be trained using single-energy CT images available from a spectral scanner, enabling it to predict effective energy values, even from conventional CT images.

[0013] Finally, a method is provided for training a neural network model that can predict the effective energy values ​​of a conventional CT image based solely on the image itself, allowing for the estimation of the effective energy values ​​of CT images without any additional or contextual information. The effective energy values ​​of the CT images can then be used for future purposes, such as calculating transfer functions for material decomposition.

[0014] In some embodiments, the array of training inputs comprises single-energy CT images from a spectral scanner ranging from 40 keV to 140 keV with a step size of 0.1 keV or greater. This is a typical range of effective energies for conventional CT images and allows the model to estimate the effective energy of CT images with an accuracy of 0.1 keV.

[0015] In some embodiments, a spectrum scanner single-energy CT image is specifically a full-axis CT image.

[0016] In some embodiments, the array of training inputs further includes adjusted spectral scanner single-energy CT images, wherein the adjustment includes slice flipping, rotation, shifting, scaling, cropping, and using different noise levels. This increases the amount of training data for training the neural network and also allows training to be invariant to image characteristics that vary due to reconstruction parameters or anatomical variations.

[0017] In some embodiments, the neural network model includes a convolutional neural network (CNN) or a machine learning regression or classification model. CNNs have proven particularly successful in analyzing images and have been found to classify images with a much lower error rate than other types of neural networks. Regression models have been found to be particularly useful for analyzing the relationship between independent features (e.g., pixels in a CT image) and a dependent variable (e.g., effective energy). Classification models are particularly useful for assigning data to discrete categories based on a specific set of features.

[0018] In some embodiments, the array of training inputs further includes a corresponding aggregated feature vector for each corresponding single-energy CT image from a spectral scanner. It has been found that using corresponding aggregated feature vectors with single-energy CT images can improve the performance of neural network models.

[0019] In some embodiments, principal component analysis (PCA) is used to preprocess the aggregated eigenvectors. By focusing on the most important eigenvector components, it has been found that the training performance of neural network models can be further improved.

[0020] In some embodiments, the aggregated feature vector includes a histogram of Hounsfield unit (HU) values ​​within each corresponding single-energy CT image from a spectral scanner. It has been found that specifically using histograms of HU values ​​within CT images during training can significantly improve the performance of neural network models.

[0021] A method for estimating the effective energy value of a conventional CT image is also provided, the method comprising: obtaining a first CT image; and processing the first CT image using a neural network model trained according to any method disclosed herein to generate a first estimated effective energy value of the first CT image.

[0022] After being trained according to any of the methods described above, the neural network model can then be able to actually process CT images and generate estimated effective energy values ​​for the processed CT images.

[0023] In some embodiments, the method for estimating the effective energy value of a conventional CT image further includes using the first estimated effective energy value in additional image processing steps of the first CT image, such as bone beam hardening correction. Once the estimated effective energy value of the CT image is generated, this value can be used to improve the image processing steps. For example, the estimated effective energy value of the CT image can be used for bone beam hardening correction of the CT image.

[0024] In some embodiments, the method for estimating the effective energy value of a conventional CT image further includes segmenting a first CT image into multiple CT sub-images and processing the multiple CT sub-images using a neural network model to generate an estimated effective energy value for each CT sub-image. This allows for consideration of the fact that the effective energy value may vary within a conventional CT image. For example, CT images with strong beam hardening artifacts are more likely to produce images where the effective energy value varies within the image. The estimated effective energy value of each sub-image can then be used to compute a different local transfer function for each sub-image, and this local transfer function can then be used to apply material decomposition to each sub-image separately using each local transfer function.

[0025] In some embodiments, the method for estimating the effective energy value of a conventional CT image further includes obtaining a second CT image, processing the second CT image using a neural network model to generate a second estimated effective energy value for the second CT image, and calculating a transfer function based on the first estimated effective energy value of the first CT image and the second estimated effective energy value of the second CT image. This can be particularly useful in the case of dual-energy CT imaging, where two CT images are generated in each scan, one at a lower energy and one at a higher energy. Once the effective energy values ​​have been estimated for the two CT images, the transfer function can be calculated based on the estimated effective energy values.

[0026] In some embodiments, the method further includes performing material decomposition based on a first CT image and a second CT image using a calculated transfer function. Spectral imaging achieved through material decomposition is an important application of CT images.

[0027] In some embodiments, a method for estimating the effective energy value of a conventional CT image includes obtaining a second CT image, and further includes: obtaining at least a third CT image; processing the at least third CT image using a neural network model to generate a third estimated effective energy value of the at least third CT image; further calculating a transfer function based on the third estimated effective energy value of the at least third CT image; and performing material decomposition based on the first CT image, the second CT image, and the third CT image using the calculated transfer function. This can be particularly useful in the case of multi-energy material decomposition, where, for example, more than two images are generated for each scan in a photon-counting CT system.

[0028] A computer program including code modules is also provided, which, when the program is run on a processing system, are used to implement any of the methods disclosed herein.

[0029] A system for training a neural network model for predicting effective energy values ​​of conventional CT images is also provided. The system includes a processor device configured to train the neural network model using a training algorithm configured to receive an array of training inputs and corresponding known outputs, wherein the training inputs include spectral scanner single-energy CT images, and the corresponding known outputs include the effective energy values ​​of the spectral scanner single-energy CT images.

[0030] A system for estimating the effective energy value of a conventional CT image is also provided, the system comprising a processor device configured to: a first CT image; process (330) the first CT image; and generate a first estimated effective energy value of the first CT image using a neural network model trained according to any one of the methods of claims 1 to 7.

[0031] Therefore, the concept of training a neural network model for predicting the effective energy value of conventional CT images can be proposed.

[0032] These and other aspects of the invention will become apparent from and will be illustrated with reference to the embodiments described below. Attached Figure Description

[0033] To better understand the invention and to more clearly illustrate how the invention can be implemented, reference will now be made to the accompanying drawings by way of example only, in which: Figure 1 This is a schematic diagram of an imaging system such as a CT scanner; Figure 2A This is a simplified flowchart of a method for training a neural network model to predict the effective energy value of a conventional CT image; Figure 2B yes Figure 2A A more in-depth explanation of the method; Figure 3 This is a flowchart of a method for estimating the effective energy value of a conventional CT image; Figure 4 This is a flowchart of a method for performing material decomposition based on multiple CT images; Figure 5 This is a simplified block diagram of a system for training a neural network model to predict the effective energy values ​​of conventional CT images; and Figure 6 An example of a computer in which one or more parts of an embodiment may be employed is shown. Detailed Implementation

[0034] The invention will be described with reference to the accompanying drawings.

[0035] It should be understood that while the detailed description and specific examples indicate exemplary embodiments of the apparatus, system, and method, they are intended for illustrative purposes only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the apparatus, system, and method of the present invention will be better understood from the following description, claims, and drawings. It should be understood that the drawings are merely schematic and not drawn to scale. It should also be understood that the same reference numerals are used in all drawings to indicate the same or similar parts.

[0036] By studying the accompanying drawings, the disclosure, and the claims, those skilled in the art can understand and implement variations of the disclosed embodiments in practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude a plurality.

[0037] It should be understood that the accompanying drawings are merely schematic and not drawn to scale. It should also be understood that the same reference numerals are used in all the accompanying drawings to indicate the same or similar parts.

[0038] Embodiments of this disclosure relate to various techniques, methods, schemes, and / or solutions related to training a neural network model for predicting effective energy values ​​in conventional CT images. Based on the proposed concepts, multiple possible solutions can be implemented individually or in combination. That is, although these possible solutions may be described individually below, two or more of these possible solutions may be implemented in one combination or another.

[0039] Embodiments of the present invention aim to utilize single-energy CT images from a spectral scanner with known effective energy values ​​to train a neural network model, which can then be used to predict the effective energy values ​​of conventional (i.e., non-single-energy) CT images. By training the neural network model to predict the effective energy values ​​of CT images using only images, a method for estimating the effective energy of CT images can be provided that does not require system calibration and requires no additional system or image information. Even the actual kVp setting information of the scan is not required. The calibration process during clinical practice can then be avoided.

[0040] Therefore, the proposed concept aims to provide schemes, solutions, concepts, designs, methods, and systems related to training neural network models for predicting effective energy values ​​of conventional CT images.

[0041] Figure 1An imaging system 100, such as a CT scanner configured for spectral (multi-energy) imaging, is schematically shown, which may be used to acquire training data. The imaging system 100 includes a generally fixed gantry 102 and a rotatable gantry 104, which is rotatably supported by the fixed gantry and rotates about the z-axis around an examination area 106. A subject support 120 (such as a sofa) supports an object or subject in the examination area.

[0042] A radiation source 108 (such as an X-ray tube) is rotatably supported by a rotatable gantry 104, rotates with the gantry, and emits radiation through the examination area 106. In one example, the radiation source comprises a single broad-spectrum X-ray tube. In another example, the radiation source comprises a single X-ray tube configured to switch between at least two different emission voltages (e.g., 80 kVp and 140 kVp) during a scan. In yet another example, the radiation source comprises two or more X-ray tubes configured to emit radiation with different average spectra. In yet another example, the radiation source comprises a combination thereof.

[0043] A radiation-sensitive detector array 110 spans an inspection region 106 facing an angular arc opposite a radiation source 108. The array detects radiation passing through the inspection region and generates one or more electrical signals (projection data) indicative of it. In cases where the radiation source comprises a single broadband X-ray tube, the array includes energy-resolved detectors (e.g., direct-conversion photon-counting detectors, at least two sets of scintillators with different spectral sensitivities (multilayers), etc.). Utilizing kVp switching and multi-tube configurations, the detector array can include single-layer detectors, direct-conversion photon-counting detectors, and / or multilayer detectors. Direct-conversion photon-counting detectors can include conversion materials such as CdTe, CdZnTe, Si, Ge, GaAs, or other direct-conversion materials. Examples of multilayer detectors include dual-layer detectors, such as those described in US 7968853 B2.

[0044] The reconstruction device 118 receives spectral projection data from the detector array 110 and reconstructs spectral volumetric image data, such as sCCTA image data, high-energy images, low-energy images, photoelectric images, Compton scattering images, iodine images, calcium images, virtual non-contrast images, bone images, soft tissue images, and / or other base material images. The reconstruction device is also capable of reconstructing non-spectral volumetric image data, for example, by combining spectral projection data and / or spectral volumetric image data. Typically, the spectral projection data and / or spectral volumetric image data will include data from at least two different energy and / or energy ranges.

[0045] The console 112 serves as an operator console. The console is operatively connected to an output device 116, such as a monitor, and an input device 114, such as a keyboard, mouse, etc. Software residing on the console 112 allows the operator to interact with and / or operate the imaging system 100 via a graphical user interface (GUI) or other means.

[0046] Now for reference Figure 2A and Figure 2B The flowchart depicts a method 200 for training a neural network model 250 for predicting the effective energy value of conventional CT images according to the proposed embodiment.

[0047] The method includes a step 210 of training a neural network model 250 using a training algorithm 240 configured to receive an array of training inputs 220 and corresponding known outputs 230, wherein the training inputs include single-energy CT images from a spectral scanner, and the corresponding known outputs include the effective energy values ​​of the single-energy CT images from the spectral scanner. The effective energy of a CT image depends on the attenuation characteristics of the scanned object, and therefore varies for each CT image; even for the same patient, an axial scan of the abdomen will have a different effective energy than an axial scan of the legs. The trained neural network model will be able to process and estimate the effective energy values ​​of CT images, even from different CT systems to the CT system from which its training data was obtained.

[0048] In some embodiments, the neural network model 250 includes a convolutional neural network (CNN) or a machine learning regression or classification model. CNNs have proven particularly successful in analyzing images and have been found to classify images with a much lower error rate than other types of neural networks. Regression models have been found to be particularly useful for analyzing the relationship between independent features (e.g., pixels in a CT image) and a dependent variable (e.g., effective energy). Classification models are particularly useful for assigning data to discrete categories (e.g., effective energy values) based on a specific set of features. Example CNN models that can be used are DenseNet, ResNet-like, or EfficientNet. A machine learning regression model that can be used is a random forest.

[0049] Artificial neural networks (or simply neural networks) are inspired by the human brain. A neural network consists of layers, each containing multiple neurons. Each neuron performs a mathematical operation. Specifically, each neuron can include different weighted combinations of a single type of transformation (e.g., the same type of transformation, sigmoid, etc., but with different weights). In processing input data, the mathematical operation of each neuron is performed on the input data to produce a numerical output, and the output of each layer in the neural network is sequentially fed into the next layer. The final layer provides the output.

[0050] Several types of neural networks exist, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). This exemplary embodiment of the invention employs a CNN-based learning algorithm because CNNs have proven particularly successful in analyzing videos and are able to classify frames within a video with a much lower error rate than other types of neural networks.

[0051] CNNs typically consist of several layers, including convolutional layers, pooling layers, and fully connected layers. Convolutional layers consist of a set of learnable filters that extract features from the input. Pooling layers are a form of non-linear downsampling that reduces data size by combining the outputs of multiple neurons in one layer into a single neuron in the next layer. Fully connected layers connect each neuron in one layer to all neurons in the next layer.

[0052] Methods for training machine learning algorithms are well-known. Typically, such methods involve obtaining a training dataset that includes training input data entries and corresponding training output data entries. An initialized machine learning algorithm is applied to each input data entry to generate a predicted output data entry. The error between the predicted output data entry and its corresponding training output data entry is used to modify the machine learning algorithm. 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 often referred to as supervised learning.

[0053] For example, the weights of the mathematical operations for each neuron can be modified until the error converges. Known methods for modifying neural networks include gradient descent, backpropagation, and other algorithms.

[0054] The training input data entries 220 for the CNN that can be used in method 200 correspond to spectral scanner single-energy CT images. The training output data entries 230 correspond to the effective energy values ​​of each corresponding spectral scanner single-energy CT image. Furthermore, several preprocessing methods can be employed to improve the training samples. In other words, a first CNN can be trained using training algorithm 240, which is configured to receive an array of training inputs and corresponding known outputs, wherein the training inputs include spectral scanner single-energy CT images, and the corresponding known outputs include the effective energy values ​​of the spectral scanner single-energy CT images. In this way, the CNN is trained to output the effective energy values ​​of the spectral scanner single-energy CT images when provided with them.

[0055] By creating corresponding single-energy images for various single energy levels, single-energy CT images of a spectral scanner can be easily created using a spectral CT system. Note that real human patient data is required, not just phantom data.

[0056] In some embodiments, the array of training inputs 220 comprises spectral scanner single-energy CT images ranging from 40 keV to 140 keV with a step size of 0.1 keV or greater. This is a typical range of effective energies for conventional CT images and enables the model to estimate the effective energy of CT images with an accuracy of 0.1 keV.

[0057] In some embodiments, a spectrum scanner single-energy CT image is specifically a full-axis CT image.

[0058] In some embodiments, the array of training inputs 220 also includes adjusted spectral scanner single-energy CT images, wherein the adjustments include slice flipping, rotation, shifting, scaling, cropping, and using different noise levels. For example, a low-pass filter can be applied to the single-energy CT images. This increases the amount of training data for training the neural network model 250 and also allows training to be invariant to image characteristics that vary due to reconstruction parameters or anatomical variations. In other words, providing multiple versions of CT images at a given target energy makes training somewhat agnostic. This technique is also known as data augmentation.

[0059] In some embodiments, the array of training inputs 220 also includes a corresponding aggregated feature vector for each corresponding spectral scanner single-energy CT image. It has been found that using the corresponding aggregated feature vector with the single-energy CT images can improve the performance of the neural network model 250. For example, the corresponding aggregated feature vector for each corresponding spectral scanner single-energy CT image can be used during training in a multilayer perceptron (MLP), a random decision forest, or adaptive augmentation. In some embodiments, the aggregated feature vector itself is sufficient, essentially replacing the original CT image.

[0060] In some embodiments, principal component analysis (PCA) is used to preprocess the aggregated eigenvectors. By focusing on the most important eigenvector components, it has been found that the training performance of the neural network model 250 can be further improved.

[0061] In some embodiments, the aggregated feature vector includes a histogram of Hounsfield unit (HU) values ​​within each corresponding single-energy CT image from a spectral scanner. It has been found that using a histogram of HU values ​​within CT images specifically during training can significantly improve the performance of neural network model 250.

[0062] Now for reference Figure 3 A figure 300 is provided for a method 300 for estimating the effective energy value of conventional CT images.

[0063] In step 310, a first CT image is obtained. The first CT image does not have to be single-energy and can be any conventional CT image. Conventional CT images may include scatter correction and beam hardening correction.

[0064] In step 320, the first CT image is segmented into multiple CT sub-images. This allows for the fact that the effective energy may vary within the first CT image. For example, a CT image with strong beam hardening artifacts is more likely to produce an image where the effective energy varies within the image. However, step 320 is optional, and the method can continue with a single complete first CT image, i.e., without performing step 320. In step 330, the neural network trained in method 200 is used to process the multiple CT sub-images (or a single complete first CT image) to generate an estimated effective energy value for each CT sub-image (or to generate an estimated effective energy value for the entire first CT image). In some cases, sub-images may each have the same estimated effective energy value, in which case a single effective energy value can be used as the value for the entire image.

[0065] In step 340, the estimated effective energy value of the first CT image (or sub-image) can be used in another image processing step of the first CT image. For example, the estimated effective energy value of the CT image can be used for bone beam sclerosis correction of the CT image.

[0066] Now for reference Figure 4 A figure is provided for a method 400 for performing material decomposition based on multiple (i.e., at least two) CT images.

[0067] In step 310, a first CT image is obtained. In step 410, a second CT image is obtained. In step 420, at least a third CT image is obtained.

[0068] In step 430, the neural network trained in method 200 is used to process the first CT image, the second CT image, and at least the third CT image to generate a first effective energy value, a second effective energy value, and a third effective energy value for the first CT image, the second CT image, and at least the third CT image, respectively.

[0069] In step 440, a transfer function is calculated based on a first effective energy value, a second effective energy value, and a third estimated effective energy value, respectively, of the first CT image, the second CT image, and at least the third CT image. Estimating the effective energy values ​​of two CT images and calculating the transfer function based on the two estimated effective energy values ​​can be particularly useful in the case of dual-energy CT imaging, where two conventional CT images are generated in each scan, one at a lower energy and one at a higher energy. Examples of dual-energy CT systems are fast kVp switching systems, dual-layer detector systems, and dual-source systems. The addition of at least a third CT image, along with its subsequent estimation of effective energy value and its use in the calculation of the transfer function, can be particularly useful in the case of multi-energy material decomposition, where, for example, more than two images are generated for each scan in a photon-counting CT system.

[0070] In step 450, material decomposition based on the first CT image, the second CT image, and at least the third CT image is performed using the transfer function calculated in step 440. When only two CT images are available, material decomposition is performed based solely on the first and second CT images using the transfer function calculated using the first and second estimated effective energy values, respectively. Spectral imaging achieved through material decomposition is an important application of CT images.

[0071] More specifically, for each energy channel e Conventional CT images must be reconstructed, and the effective energy for each energy channel (i.e., for each CT image) needs to be estimated, for example, using any of the methods disclosed herein. E e When the number of materials in a CT image (e.g., 2) is less than the number of energy channels (e.g., 3), the forward model can be given by the following equation: Then you can use pseudo-inverse. A T A Or weighted pseudo-inverse A T C -1 A Inverse equation (4) is obtained, where C is the covariance matrix between the input CT images. The weighted pseudo-inverse is an embodiment of the above transfer function.

[0072] Taking the acquisition and processing of two CT images as an example, the effective energy value can be estimated for each image pair at a given z-position. The matrix given in Equation 3 can then be formalized, inverted, and applied to the image pair to calculate the material image at the given z-position. ph and sc .

[0073] Now for reference Figure 5 A simplified block diagram of a system 500 for training a neural network model for predicting effective energy values ​​of conventional CT images, according to an embodiment, is depicted. The system configured to train a neural network model for predicting effective energy values ​​of conventional CT images includes a processor device 510.

[0074] The processor device 510 is configured to train a neural network model using a training algorithm configured to receive an array of training inputs and corresponding known outputs, wherein the training inputs include spectral scanner single-energy CT images and the corresponding known outputs include the effective energy values ​​of the spectral scanner single-energy CT images.

[0075] Further reference Figure 5 A simplified block diagram of a system 520 for estimating the effective energy value of conventional CT images using a neural network model is also depicted. This system uses a trained neural network model to predict the effective energy value of conventional CT images according to the present invention. System 520 includes a processor device 530.

[0076] Now for reference Figure 6 An example of a computer 600 in which one or more portions of an embodiment may be employed is shown. The various operations discussed above can utilize the capabilities of computer 600. In this regard, it should be understood that system functional blocks can run on a single computer or can be distributed across several computers and locations (e.g., via an Internet connection).

[0077] Computer 600 includes, but is not limited to, PCs, workstations, laptops, PDAs, handheld devices, servers, storage devices, etc. Typically, in terms of hardware architecture, computer 600 may include one or more processors 610, memory 620, and one or more I / O devices 630 communicatively coupled via a local interface (not shown). As known in the art, the local interface can be, for example, but not limited to, one or more buses or other wired or wireless connections. The local interface may have additional elements such as controllers, buffers (caches), drivers, repeaters, and receivers to enable communication. Furthermore, the local interface may include address, control, and / or data connections to enable appropriate communication between the aforementioned components.

[0078] Processor 610 is a hardware device for executing software that can be stored in memory 620. Processor 610 can actually be any custom or commercially available processor, central processing unit (CPU), digital signal processor (DSP), or auxiliary processor among several processors associated with computer 600, and processor 610 can be a semiconductor-based microprocessor (in the form of a microchip) or microprocessor.

[0079] Memory 620 may include any one or a combination of volatile memory elements (e.g., random access memory (RAM), such as dynamic random access memory (DRAM), static random access memory (SRAM), etc.) and non-volatile memory elements (e.g., ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic tape, optical disc read-only memory (CD-ROM), magnetic disk, floppy disk, cassette tape, cassette tape, etc.). Furthermore, memory 620 may contain electronic, magnetic, optical, and / or other types of storage media. Note that memory 620 may have a distributed architecture, in which various components are geographically separated but accessible by processor 610.

[0080] The software in memory 620 may include one or more individual programs, each including an ordered list of executable instructions for implementing logical functions. According to an exemplary embodiment, the software in memory 620 includes a suitable operating system (O / S) 650, a compiler 660, source code 670, and one or more applications 680. As shown, application 680 includes numerous functional components for implementing features and operations of the exemplary embodiment. Application 680 of computer 600 may represent various applications, computing units, logic, functional units, processes, operations, virtual entities, and / or modules according to the exemplary embodiment, but application 680 is not intended to be limiting.

[0081] Operating system 650 controls the execution of other computer programs and provides scheduling, input-output control, file and data management, memory management, communication control, and related services. The inventors envision that application 680, used to implement exemplary embodiments, can be applied to all commercially available operating systems.

[0082] Application 680 can be a source program, an executable program (object code), a script, or any other entity including a set of instructions to be executed. When it is a source program, the program is typically translated by a compiler (such as compiler 660), assembler, interpreter, etc., which may or may not be included in memory 620 to operate appropriately in conjunction with O / S 650. Furthermore, application 680 can be written in an object-oriented programming language with classes of data and methods, or a procedural programming language with routines, subroutines, and / or functions, such as, but not limited to, C, C++, C#, Pascal, BASIC, API calls, HTML, XHTML, XML, ASP scripts, JavaScript, FORTRAN, COBOL, Perl, Java, ADA, .NET, etc.

[0083] I / O device 630 may include input devices, such as, but not limited to, a mouse, keyboard, scanner, microphone, camera, etc. Furthermore, I / O device 630 may also include output devices, such as, but not limited to, a printer, monitor, etc. Finally, I / O device 630 may also include devices for transmitting both input and output, such as, but not limited to, a NIC or modulator / demodulator (for accessing remote devices, other files, devices, systems, or networks), radio frequency (RF) or other transceivers, telephone interfaces, bridges, routers, etc. I / O device 630 also includes components for communication over various networks, such as the Internet or intranets.

[0084] If the computer 600 is a PC, workstation, intelligent device, etc., the software in the memory 620 may also include a Basic Input / Output System (BIOS) (omitted for simplicity). The BIOS is a set of basic software routines that initialize and test the hardware at startup, boot the O / S 650, and support data transfer between hardware devices. The BIOS is stored in some type of read-only memory, such as ROM, PROM, EPROM, EEPROM, etc., so that the BIOS can be executed when the computer 800 is activated.

[0085] When the computer 600 is operating, the processor 610 is configured to execute software stored in the memory 620 to transfer data to and from the memory 620, and typically controls the operation of the computer 600 according to the software. Applications 680 and O / S 650 are read, possibly buffered within the processor 610, and then executed.

[0086] When application 680 is implemented as software, it should be noted that application 680 can 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 can be an electronic, magnetic, optical, or other physical device or apparatus capable of containing or storing computer programs for use by or in connection with a computer-related system or method.

[0087] Application 680 can be embodied in 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 processor-containing system, or other system capable of acquiring and executing instructions from and from an instruction execution system, apparatus, or device. In the context of this document, "computer-readable medium" can be any device capable of storing, transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device. Computer-readable media can be, for example, but not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, devices, or propagation media.

[0088] Figure 1-3 Methods and Figure 4The system can be implemented in hardware, software, or a combination of both (e.g., as firmware running on a hardware device). Where embodiments are implemented partially or entirely in software, the functional steps shown in the process flowchart can be executed by appropriately programmed physical computing devices, such as one or more central processing units (CPUs) or graphics processing units (GPUs). Each process as shown in the flowchart and its individual component steps can be executed by the same or different computing devices. According to embodiments, a computer-readable storage medium stores a computer program including computer program code configured to cause one or more physical computing devices to perform the encoding or decoding methods described above when the program is run on one or more physical computing devices.

[0089] Storage media can include volatile and non-volatile computer memories, such as RAM, PROM, EPROM and EEPROM, optical discs (such as CD, DVD, BD), and magnetic storage media (such as hard disks and magnetic tapes). Various storage media can be fixed within a computing device or can be portable, allowing one or more programs stored thereon to be loaded into the processor.

[0090] With regard to the implementation of some or all of the embodiments in hardware, Figure 4 The blocks shown in the block diagram can be individual physical components or logical subdivisions of a single physical component, or they can all be implemented in an integrated manner within a single physical component. The functionality of one block shown in the figures can be divided among multiple components in an implementation, or the functionality of multiple blocks shown in the figures can be combined in a single component in an implementation. Hardware components suitable for embodiments of the present invention include, but are not limited to, conventional microprocessors, application-specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs). One or more blocks can be implemented as a combination of dedicated hardware for performing some functions and one or more programmable microprocessors and associated circuitry for performing other functions.

[0091] By studying the accompanying drawings, disclosure, and claims, those skilled in the art can understand and implement variations of the disclosed embodiments in practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude multiple. A single processor or other unit can perform the functions of several items recited in the claims. The fact that certain measures are recited in mutually different dependent claims does not mean that a combination of these measures cannot be used advantageously. If a computer program has been discussed above, it can be stored / distributed on a suitable medium, such as an optical storage medium or solid-state medium provided with or as part of other hardware, but it can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems. If the term "suitable" is used in the claims or description, it should be noted that the term "suitable" is intended to be equivalent to the term "configured as." Any reference numerals in the claims should not be construed as limiting the scope.

[0092] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of instructions, which includes one or more executable instructions for implementing one or more specified logical functions. In some alternative implementations, the functions marked in the blocks may occur in a non-linear order. For example, two blocks shown consecutively may actually be executed substantially simultaneously, or these blocks may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart illustrations, and combinations of blocks in the block diagram and / or flowchart illustrations, can be implemented by a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions. The architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of instructions, which includes one or more executable instructions for implementing one or more specified logical functions. In some alternative implementations, the functions marked in the blocks may occur in a non-linear order. For example, two boxes shown consecutively may actually be executed substantially simultaneously, or these boxes may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart illustration, as well as combinations of boxes in the block diagram and / or flowchart illustration, can be implemented by a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions.

Claims

1. A method (200) for training a neural network model (250) for predicting effective energy values ​​of conventional CT images, the method comprising: A neural network model is trained using a training algorithm (240), which is configured to receive an array of training inputs (220) and corresponding known outputs (230), wherein the training inputs include spectral scanner single-energy CT images, and the corresponding known outputs include the effective energy values ​​of the spectral scanner single-energy CT images.

2. The method according to claim 1, wherein, The array of training inputs comprises single-energy CT images from a spectral scanner ranging from 40 keV to 140 keV with step sizes of 0.1 keV or larger.

3. The method according to claim 1 or 2, wherein, The array of training inputs also includes adjusted spectral scanner single-energy CT images, wherein the adjustments include slice flipping, rotation, shifting, scaling, cropping, and using different noise levels.

4. The method according to any one of claims 1-3, wherein, The neural network model includes convolutional neural networks, machine learning regression models, or classification models.

5. The method according to any one of claims 2-4, wherein, The array of training inputs also includes the corresponding aggregated feature vector for each single-energy CT image from the corresponding spectral scanner.

6. The method according to claim 5, wherein, Principal component analysis (PCA) was used to preprocess the aggregated feature vectors.

7. The method according to claim 5 or 6, wherein, The aggregated feature vector includes a histogram of HU values ​​within each corresponding single-energy CT image from a spectral scanner.

8. A method for estimating the effective energy value of a conventional CT image, the method comprising: Obtain (310) the first CT image; as well as The first CT image is processed (330) using the neural network model trained according to any one of the methods of claims 1 to 7 to generate a first estimated effective energy value of the first CT image.

9. The method according to claim 8, wherein, The method further includes: The first estimated effective energy value is used in another image processing step of the first CT image, such as bone beam hardening correction.

10. The method according to claim 8 or 9, wherein, The method further includes: The first CT image is segmented (320) into multiple CT sub-images; and The neural network model is used to process the multiple CT sub-images to generate an estimated effective energy value for each CT sub-image.

11. The method according to any one of claims 8-10, further comprising: Obtain (410) the second CT image; The second CT image is processed (430) using the neural network model to generate a second estimated effective energy value for the second CT image; as well as The transfer function (440) is calculated based on the first estimated effective energy value of the first CT image and the second estimated effective energy value of the second CT image.

12. The method according to claim 11, wherein, The method further includes: (450) Material decomposition is performed using the calculated transfer function based on the first CT image and the second CT image.

13. A computer program including code modules, wherein when the program is run on a processing system, the code modules are configured to implement the method according to any of the preceding claims.

14. A system (500) for training a neural network model (250) for predicting effective energy values ​​of conventional CT images, said system comprising: Processor device (510), which is configured to: A training algorithm (240) is used to train a neural network model, the training algorithm being configured to receive an array of training inputs (220) and corresponding known outputs (230), wherein the training inputs include spectral scanner single-energy CT images, and the corresponding known outputs include the effective energy values ​​of the spectral scanner single-energy CT images.

15. A system (520) for estimating the effective energy value of a conventional CT image, the system comprising: Processor device (530), which is configured as follows: Obtain the first CT image. Process the first CT image, The neural network model trained according to any one of the methods of claims 1 to 7 is used to generate a first estimated effective energy value for the first CT image.

Images