reconstructed image data

Abstract

A method for reconstructing image data is disclosed, specifically a method that includes techniques for determining an optimal weighted execution sequence of available reconstruction algorithms using a multi-processor unit. The disclosed method includes executing a series of optimal weighted execution sequence candidates on a representative slice of image data and comparing their results in order to select one of the candidates as the optimal weighted execution sequence.

Description

[0001] Cross Reference to Related Applications

[0002] This application claims the benefit of U.S. Provisional Application No. 62 / 721,517, entitled "Algorithm Weighting System for Image Reconstruction," filed August 22, 2018, by Shekhar Dwivedi, commonly assigned with the present application and incorporated herein by reference.

[0003] Reconstruction," filed August 22, 2018, by Shekhar Dwivedi, commonly assigned with the present application and incorporated herein by reference. TECHNICAL FIELD

[0004] The present application relates generally to data imaging. More particularly, the present application relates to a system and method for weighting and ordering reconstruction algorithms for image reconstruction. BACKGROUND

[0005] Data imaging is an important technology utilized by a wide variety of industries. Depending on the field, the custom of any particular industry in imaging operations can be to produce one or more multi-dimensional images using one or more imaging modalities. In medical imaging, for example, images can be produced using a range of modalities including X-ray, MRI, CT scans, and ultrasound. X-ray based imaging is also used in security, oil and gas industries. SUMMARY

[0006] In one aspect, a system for reconstructing image data is described. The system includes a multi-processor unit configured to: receive a request to visualize image data; select one or more reconstruction algorithms to be performed on the image data based on the request; and determine an optimal weighted execution sequence of the reconstruction algorithms. The optimal weighted execution sequence includes a weight for each reconstruction algorithm and an execution sequence for the reconstruction algorithms.

[0007] In another aspect, a method for reconstructing image data is described. The method includes: receiving a request to visualize image data; selecting one or more reconstruction algorithms to be performed to process the image data based on the request; and determining an optimal weighted execution sequence of the reconstruction algorithms, wherein the optimal weighted execution sequence includes a weight for each reconstruction algorithm and an execution sequence for the reconstruction algorithms. BRIEF DESCRIPTION OF DRAWINGS

[0008] Reference will now be made to the following descriptions taken in conjunction with the accompanying drawings, in which:

[0009] Figure 1 An embodiment of a system for reconstructing image data constructed in accordance with the principles of the present disclosure is illustrated;

[0010] An embodiment of a system for reconstructing image data constructed in accordance with the principles of the present disclosure is illustrated;Figure 2 An embodiment of a multi-processor unit constructed in accordance with the principles of the present disclosure is shown;

[0011] Figure 3 A flowchart showing an embodiment of a method for reconstructing image data implemented in accordance with the principles of the present disclosure; and

[0012] Figure 4 An example of a reconstructed image is shown. DETAILED DESCRIPTION

[0013] Image reconstruction is a technique for generating an image from data from an imaging device, such as a sensor. There are many different reconstruction algorithms in the literature and in practice, and each algorithm provides different benefits to the quality of the final image. The ordering and weighting of these reconstruction algorithms can have a significant impact on the final image quality, and a unique "recipe" of algorithm sequence and weights can be necessary to produce an optimal image.

[0014] The weighting and ordering is highly dependent on the problem at hand, the application, and several other factors. It is therefore almost impossible to test every combination with different weights and sequences. For this reason, image reconstruction is traditionally performed either by using only one algorithm, which is often applied over several iterations, sometimes hundreds of iterations, on the same image data, or by applying a different algorithm during a pre-processing or post-processing phase to enhance the final image quality from the algorithm used during reconstruction.

[0015] However, the number of iterations that the algorithms are applied during reconstruction is often static, regardless of the dataset reconstruction or specific task for which / over which it is performed. It is often the case that the number of iterations applied far exceeds the point of diminishing returns, i.e., the point at which the benefit of each additional iteration no longer outweighs the cost of applying that iteration. Also, applying a different algorithm during pre-processing or post-processing requires extending the time of pre-processing and post-processing, which is often difficult to accommodate due to lack of time and computational resources. For this reason, conventional approaches result in insignificant or difficult to detect benefits while suffering from significantly reduced returns.

[0016] The present disclosure introduces a technique for generating an optimal weighted execution sequence of reconstruction algorithms using a multi-processor unit. The introduced technique can be parameterized to generate a weighted sequence that provides the best reconstructed image given time and computational resources. It should be understood that "optimal weighted execution sequence" means an execution sequence of weighted reconstruction algorithms that the processing unit executes against a given image data to reconstruct an image that exceeds one or more quality preference benchmarks / thresholds under a given time constraint. The optimal weighted execution sequence includes the sequence / order in which the reconstruction algorithms should be run and the respective weights (or iterations or percentage of time) that each reconstruction algorithm should utilize to run against the image data in order to reconstruct the image.

[0017] In one embodiment, the introduced technique dynamically determines the optimal weighted sequence for two or more algorithms by testing a plurality of weighted sequences during reconstruction and selecting the best sequence. Rather than testing the weighted sequences against the entire image data, the introduced technique tests the weighted sequences against a representative portion of the image data. The introduced technique tests the plurality of weighted sequences in parallel using the parallel processing capabilities of the multi-processor unit. As such, the introduced technique can test the plurality of weighted sequences using a much smaller amount of time and computational resources compared to conventional methods by utilizing the parallel processing capabilities of advanced hardware processors, such as one or more GPUs. The introduced technique can also significantly reduce the time and computational resources spent because less pre-processing and post-processing is required. Furthermore, because the weighted sequence determination process can be separate from the actual reconstruction, the introduced technique can provide the weighted sequence to the user for further modification prior to reconstruction if desired.

[0018] Figure 1 An embodiment of a system 100 for visualizing image data constructed in accordance with the principles of the present disclosure is shown. In the illustrated embodiment, the system 100 includes a multi-processor unit 110, a local imaging / computing device 120, and a remote imaging / computing device 130. The multi-processor unit 110 can be co-located with the local device 120 in a data center or server 150 for local computing, or located remotely from the remote imaging / computing device 130 for cloud computing. The multi-processor unit 110 can be communicatively connected to the devices 120, 130 through a local area network, a wide area network, a data bus, and / or a wireless data protocol based on the location of the devices 120, 130.

[0019] In the illustrated embodiment, the multi-processor unit 110 is configured to determine an optimal weighted execution sequence of reconstruction algorithms for processing image data from one or more of the devices, such as 120 and 130. The multi-processor unit 110 can be any processing unit capable of parallel processing, such as a graphics processing unit (GPU), a central processing unit (CPU), or a combination of both a CPU and a GPU.

[0020] In the illustrated embodiment, each of the devices 120, 130 can be an imaging device configured to generate image data of an object, such as an organ of a patient. The devices 120, 130 can be a radiographic imaging device, a magnetic resonance imaging (MRI) device, a nuclear medicine imaging device, an ultrasound imaging device, an ultrasonography imaging device, an elastography imaging device, an optoacoustic imaging device, a tomographic imaging device, an echocardiography imaging device, a functional near-infrared spectroscopy imaging device, and a magnetic particle imaging device. Each of the devices 120, 130 can also be a computing device configured to issue a request to a connected multi-processor unit, such as 110, to determine an optimal weighted execution sequence and to optimally visualize the image data using the optimal weighted execution sequence.

[0021] Figure 2 An embodiment of a multi-processor unit 200 constructed in accordance with the principles of the present disclosure is shown, such as 110 in FIG. 1. The multi-processor 200 includes a plurality of processors 210, a memory 220, a network interface card (NIC) 230, and an input / output (I / O) interface 235 interconnected to each other using conventional means. Figure 1

[0022] In the illustrated embodiment, the NIC 230 and the I / O interface 235 are configured to receive a request to visualize image data. The request can be received from any communicatively connected computing / imaging device, such as 120, 130 in FIG. 1, or from a directly connected input device, such as but not limited to a keyboard, a mouse, or a touch screen. The request can include: the image data to be visualized; a reconstruction algorithm to be executed to process the image data; and one or more preferred processing parameters of the requester. Figure 1

[0023] The NIC 230 and the I / O interface 235 are also configured to receive the processing results from the memory 220 and transmit them to an output device, such as but not limited to a monitor or a display, or to other computing devices. The processing results can include the visualized image and / or the selected weighted sequence of the reconstruction algorithm.

[0024] In the illustrated embodiment, the memory 220 is a non-transitory medium configured to store various data. The memory 220 can store the received requests and inputs and the processing results of these requests and inputs. The memory 220 can also store a series of instructions that, when executed, configure the processors 210 to perform a method for processing image data, an embodiment of which is described below with reference to FIG. 2. Figure 3

[0025] ​​​The processor 210 is configured to select the reconstruction algorithms to be executed with the optimal weighting execution sequence. Using all available algorithms for ranking can increase the workload and can not be necessary for all applications and incoming data sets. A guided or curated selection of algorithms for ranking is essential. The selection of reconstruction algorithms can be specified in an input from a user or in a request. When these algorithms are not specified, the processor 210 can select them from the available reconstruction algorithms based on several factors such as: characteristics of the image data, for example, the modality the image data relates to and the resolution of the image data; characteristics of the available reconstruction algorithms, for example, the processing parameters the algorithm enhances / weakens, the number of iterations of the algorithm, the point of diminishing returns of the algorithm (how well each reconstruction algorithm works after a certain number of iterations) and the coverage of the algorithm, for example, local or global (whether the given algorithm affects only a certain region of the image or the entire image); characteristics of the patient, for example, the disease / organ being treated and the patient’s treatment history; and preferred processing parameters. It should be appreciated that the listed factors are not exclusive and other factors related or derivable from the listed factors can be included.

[0026] In one embodiment, the selection of reconstruction algorithms can be performed using a machine learning model. A regression or classification algorithm can be designed. All available algorithms are used as a label set. The ground truth is created by determining the figure of merit for each training set using each algorithm, then, all algorithms that create a figure of merit above / below a certain threshold are considered available for that training set. A feature set can be extracted from the image data characteristics, patient / data characteristics, figure of merit characteristics, and time and computational constraints. For any new incoming data, the features are extracted and the machine learning model is evaluated for each available algorithm in order to propose the optimal set of algorithms for the new data.

[0027] Similar to the reconstruction algorithms, the preferred processing parameters can be specified in the request. These preferred processing parameters refer to the parameters the requester wants to emphasize during the reconstruction. The preferred processing parameters can include at least one of: preferred performance parameters such as preferred amount of time or computational resources for the reconstruction, or preferred visual parameters such as signal-to-noise ratio (SNR), organ contrast, image sharpness at the border, peak SNR, SSIM (structural similarity index), or SUV (standard uptake value) in functional imaging such as SPECT and PET imaging. It should be appreciated that the listed parameters are not exclusive and other parameters related or derivable from the listed parameters can be included.

[0028] The processor 210 is further configured to determine an optimal weighted execution sequence of the selected reconstruction algorithms. The optimal weighted execution sequence includes weights for each of the selected reconstruction algorithms and an order of execution of the selected reconstruction algorithms.

[0029] In one embodiment, the processor 210 determines the optimal weighted execution sequence using a representative slice technique. Using the slice technique, the processor 210 first selects a representative portion of the image data, such as a slice that contains the most information about a region of interest or an organ. The portion can be selected based on patient characteristics such as, but not limited to, the disease / condition being treated, the organ or organs being treated, and the patient's treatment history. For example, for image data related to a lung cancer patient, a portion of the image data, such as one or more slices, centered on the lung and thus containing the most information about the lung would be selected.

[0030] The processor 210 then determines a plurality of weighted execution sequences for the representative slice. The weighted execution sequences represent candidates for the optimal weighted execution sequence. The weighted execution sequences are determined based on various factors such as, but not limited to, the preferred processing parameters, the characteristics of the selected reconstruction algorithms, the characteristics of the image data, and time and computational constraints. The time and computational constraints represent the resources available to complete the request, the computational time requirements, and the available processing power. It should be understood that the listed constraints are not exclusive and that other constraints related or derivable from the listed constraints can be included.

[0031] When the weighted execution sequences are determined, the selected algorithms are executed on the representative data in accordance with the weighted execution sequences. The weighted execution sequences are executed in parallel using the plurality of processors 210 of the multiprocessor unit 200 for maximum efficiency and speed. Each of the weighted execution sequences can be executed on a different processor 210 of the multiprocessor unit 200, or the weighted execution sequences can be executed one at a time using all of the plurality of processors 210.

[0032] Based on the comparison of the reconstructed images, one of the weighted sequences that provides the best figure of merit (FOM) for the preferred processing parameters is selected as the optimal weighted execution sequence. The FOM of the weighted execution sequences can be compared using several mechanisms such as, but not limited to, artificial / automatic ROI detection, which artificially or automatically detects a region of interest in the reconstructed images and compares certain processing parameters / values in the region; model-based evaluation, which uses a predefined model to evaluate a certain region of the reconstructed images; and a multiple user voting mechanism, in which multiple users vote on which is the best reconstructed image.

[0033] In one embodiment, the optimal weighted execution sequence can be determined using a predetermined template. The predetermined template is a fixed template that is customized for determining the optimal weighted execution sequence for a certain data image and certain patient characteristics. For example, there can be a predetermined template for image data of a certain resolution corresponding to a certain modality and a certain disease / condition. The predetermined template can be a transfer function generated from previous reconstruction results.

[0034] In another embodiment, the optimal weighted execution sequence can be determined using a machine learning model. With the optimal weighted execution sequence for a given application as a label set, a machine learning algorithm (e.g., classification, decision tree / random forest, neural network, etc.) can be designed to compute the "best" weighted sequence for a given input data set. The parameters for the learning algorithm can be extracted from the application at hand, such as (but not limited to) image data characteristics, patient / object characteristics, time and computational constraints, characteristics of the algorithms involved, and quality factor characteristics. The ground truth is generated by labeling each training set with the optimal weighted sequence from a list of available weighted sequences. The ground truth is generated by an expert after examining the quality factor for each training set against the label set for all weighted sequences. Machine learning models such as (but not limited to) artificial neural networks and / or regression models can be used.

[0035] Once the optimal weighted execution sequence is determined, the processor 210 can execute the selected reconstruction algorithms in accordance with the optimal weighted sequence. Each selected reconstruction algorithm is executed iteratively on the image data based on the respective weights and also sequentially based on the execution order.

[0036] Instead of executing locally, the optimal weighted sequence can also be exported to a different computing system. The optimal weighted sequence can first be stored in the memory 220 and then transmitted to another computing system via the NIC 230.

[0037] Figure 3 A flowchart illustrating an embodiment of a method 300 for visualizing image data in accordance with the principles of the present disclosure is shown. The method 300 can be performed by a multi-processor unit such as 200 in Figure 2 The method 300 begins at step 305.

[0038] At step 310, a request to optimally process image data or to determine an optimal weighted execution sequence is received. The request can be received from a local or remote device, such as 120 and 130 in Figure 1 The request can be received from an input device such as (but not limited to) a keyboard, mouse, or touch screen that is communicatively connected to the multi-processor unit that is performing the method 300. As above with respect to Figure 2The request discussed may include: image data to be processed; a reconstruction algorithm to be executed to process the image data; and one or more preferred processing parameters.

[0039] In one example, a request to reconstruct the original sinogram data is received at the multiprocessor unit performing image reconstruction. This request includes the original sinogram data and specifies minimum noise and maximum organ contrast as preferred processing parameters. The sinogram data may be received from a remote imaging device via a wireless data protocol.

[0040] At step 320, a reconstruction algorithm to be executed is selected from the available reconstruction algorithms. The request received at step 310 can indicate which reconstruction algorithms to execute. Some commonly available reconstruction algorithms include: Simultaneous Iterative Reconstruction Technique (SIRT); Simultaneous Algebraic Reconstruction Technique (SART); Conjugate Gradient Least Squares (CGLS); Ordered Subset Expectation Maximization (OSEM), an optimized version of MLEM in which a subset of data is used during iteration; and Maximum Likelihood Expectation Maximization (MLEM), which performs computation for each pixel in the image. Non-iterative analysis algorithms such as (but not limited to) Filtered Back Projection (FBP) can also be used for selection.

[0041] When no algorithm is specified, they can be selected based on several factors, such as: the characteristics of the image data, such as the pattern involved in the image data and the resolution of the image data; the characteristics of the available reconstruction algorithm, such as the processing parameters for algorithm enhancement and the number of iterations of the algorithm; the characteristics of the patient, such as the disease / organ being treated and the patient's treatment history; and the preferred processing parameters. It should be understood that the listed factors are not exclusive and may include other factors not listed but relevant or that can be derived from the listed factors.

[0042] Continuing with the example above, the request did not specify the reconstruction algorithm to be used. Therefore, considering 1) the patient has a history of lung cancer treatment (patient characteristics), 2) the image data is raw ultrasound data (image data characteristics), and 3) SIRT, SART, and CGLS address SNR and organ contrast issues (characteristics of available algorithms), the multiprocessor unit selects SART, CGLS, and SIRT for reconstruction.

[0043] In steps 330-360, the optimal weighted execution sequence for the selected algorithms is determined. This optimal weighted execution sequence includes weights for each selected reconstruction algorithm and the execution order of these selected reconstruction algorithms. This optimal weighted execution sequence represents the best FOM (Form of Memory) that provides preferred processing parameters for a given time period when each selected algorithm is executed based on its respective weights and the order of this weighted execution sequence.

[0044] At step 330, a portion of the image data representative of the image data (representative slice) is determined. The representative portion can be selected based on patient characteristics such as (but not limited to) the disease / condition being treated, the organ or organs being treated, and the patient's treatment history. In the example above, the slice of image data capturing the lung (centered thereon) is selected as the representative portion based on the patient characteristic that the patient is a lung cancer patient with a history of lung cancer treatment.

[0045] At step 340, a plurality of weighted execution sequences of the selected reconstruction algorithms is determined. Since each of the weighted execution sequences represents a candidate of the optimal weighted execution sequence, each weighted execution sequence includes an execution order and a weight (number of iterations) for each reconstruction algorithm. The weight and sequence for each of the weighted execution sequences is determined based on various factors.

[0046] The factors include preferred processing parameters, characteristics of the selected reconstruction algorithms, time and computational constraints, and characteristics of the image data. The preferred processing parameters can be specified in an input from a user or in the request. As disclosed above, the preferred processing parameters can include at least one of: a preferred performance parameter such as a preferred amount of time or computational resources for reconstruction, or a preferred visual parameter such as SNR, organ contrast, image sharpness at boundaries, peak SNR, SSIM, or SUV in functional imaging. It should be appreciated that the listed parameters are not exclusive and other parameters not listed but can be relevant or derivable from the listed parameters can be included.

[0047] The characteristics of the selected algorithms represent performance characteristics of the algorithms, for example (but not limited to) preferred processing parameters involved by the algorithms, number of iterations of the algorithms, point of diminishing returns of the algorithms, and coverage of the algorithms (whether a given algorithm affects only a certain region of the image or the entire image). The time and computational constraints represent resources available to complete the request, for example (but not limited to) available time and processing power. The characteristics of the image data can include a modality corresponding to the image data and resolution of the image data. It should be appreciated that the listed characteristics / constraints are not exclusive and other characteristics / constraints not listed but can be relevant or derivable from the listed characteristics / constraints can be included.

[0048] According to the example above, the multi-processor unit determines that the selected algorithm can be ordered in six different sequences: SART→CGLS→SIRT, SART→SIRT→CGLS, CGLS→SART→SIRT, CGLS→SIRT→SART, SIRT→CGLS→SART, SIRT→SART→CGLS. Of these six sequences, the multi-processor selects three sequences SART→CGLS→SIRT, SART→SIRT→CGLS, CGLS→SART→SIRT based on the coverage of the algorithm and the time and computational constraints. For example, the selected sequences perform the SART, which is a global work, before the SIRT, which is a local work.

[0049] The multi-processor also determines that four different weight sets for the selected sequences SART, CGLS, SIRT can be used. For example, a first weight set has respective weights of 100 iterations, 25 iterations, and 75 iterations for SART, CGLS, SIRT; a second weight set has respective iterations of 95, 20, 85; a third weight set has respective iterations of 95, 35, 70; and a fourth weight set has respective iterations of 110, 20, 70. These weights are selected based on the fact that the point of diminishing returns for the algorithm is 110, 40, 80, and the time constraint is 200 iterations.

[0050] When the weighted execution sequences are determined, at step 350, the selected algorithm is executed in parallel on the representative data according to the weighted execution sequences. To achieve maximum efficiency, each sequence can be executed on each different processor of the available processors, or the weighted execution sequences can be executed one at a time using all of the available processors. At step 360, based on a comparison of the reconstructed images, one of the weighted sequences that provides the best figure of merit (FOM) for the preferred processing parameters is selected as the optimal weighted execution sequence. The FOM of the weighted execution sequences can be compared using mechanisms such as, but not limited to, human / automatic ROI detection, model-based evaluation, and a voting mechanism based on multiple users.

[0051] Continuing the example, since there are three sequences with four different weight sets, there are 12 weighted sequence combinations. These 12 combinations are executed in parallel by the four processors of the multi-processor unit in three stages. After execution, using automatic ROI detection focused on SNR and organ contrast of the lungs, the sequence SART→CGLS→SIRT with respective weights of 100, 25, 75 iterations is calculated to provide the lowest SNR with the highest organ contrast, and is therefore selected as the optimal weighted execution sequence.

[0052] It should be appreciated that instead of steps 330-360, a predetermined template or machine learning model can also be used to determine the optimal weighted execution sequence. The predetermined template can be a fixed template that is optimal for a certain data image and patient characteristics. For example, there can be a predetermined template that can be used for image data of a certain resolution corresponding to a certain modality and / or a certain disease / condition. The predetermined template can be a transfer function generated from previous reconstruction / processing results. It should be appreciated that the predetermined template can be generated at "off-peak" times, such as during the night or on weekends, and / or by an "offline" server where more computing resources are available that are not performing the reconstruction.

[0053] For embodiments that employ a machine learning model for deriving the optimal weighted and ordered sequence, the optimal weighted execution sequence for a given application is treated as a set of labels. A machine learning algorithm (e.g., classification, decision tree / random forest, neural network, etc.) can be designed to compute the "best" weighted sequence for a given input data set. Parameters for the learning algorithm can be extracted from the application at hand, such as (but not limited to) image data characteristics, patient / subject characteristics, temporal and computational constraints, characteristics of the algorithms involved, and quality factor characteristics. The true values are generated by labeling each training set with the optimal weighted sequence from the list of available weighted sequences. The true values can be generated by an expert after examining the quality factors generated for each training set against the set of labels. Machine learning models such as artificial neural networks and / or regression models can be used.

[0054] At step 370, the selected algorithms are executed in accordance with the optimal weighted execution sequence. Each selected reconstruction algorithm is iteratively executed on the image data based on the respective weights and also sequentially based on the execution order. Instead of step 370, the method 300 can send the optimal weighted sequence to a different computing system for execution. The optimal weighted sequence can first be stored in a memory such as 220 in FIG. 2 and then transmitted to another computing system using a NIC such as 230 in FIG. 2. Figure 2 Figure 2

[0055] In this example, the request specifies the reconstruction of the image data. As such, the selected algorithms are executed in accordance with the optimal weighted execution sequence on the entire imaging data set, and the resulting image is provided to the requester. If the request specifies the determination of the optimal weighted execution sequence, the selected sequence will be provided to the requester. The method 300 ends at step 375.

[0056] Figure 4 ​​Three differently reconstructed images 410, 420, 430 of the same image data are depicted to illustrate the differences between images reconstructed using a single reconstruction algorithm versus multiple reconstruction algorithms. The image data was reconstructed from abdominal CT sinogram data. The images 410, 420, 430 illustrate the abdominal region near the spine and kidneys.

[0057] The first image 410 depicts an image reconstructed by executing a first algorithm (Algorithm A) for 200 iterations. This image was reconstructed in approximately 130 seconds. Although the first algorithm can be suitable for removing background noise, the level of detail in the first image 410 is low compared to the second and third images 420, 430. For example, regions 415 and 425 are not as clearly marked as in the other two images 420, 430. Upon closer inspection, one can observe that certain details within regions 415 and 425 are absent in the first image 410.

[0058] The second image 420 depicts an image reconstructed by executing a second algorithm (Algorithm B) for 200 iterations. This image was reconstructed in approximately 60 seconds. In this example, the second algorithm can be suitable for preserving detail, but compromises on background noise, as regions 415 and 425 appear noisier than in the first and third images 410, 430. Upon closer inspection, one can observe that regions 415 and 425 provide better internal structural detail of very small organs and details.

[0059] The third image 430 depicts an image reconstructed by executing the first algorithm for 100 iterations and then executing the second algorithm for 100 iterations. The third image 430 removes background noise to a large extent while preserving a relatively high level of detail. Its run time is approximately 75 seconds. These depicted images illustrate how the execution sequence A→B of 100 iterations of each algorithm (equal weight) can improve the image reconstruction process for even a simple weighted execution sequence.

[0060] In explaining the present disclosure, all terms should be interpreted in the broadest possible way consistent with the context. In particular, the terms "comprises", "comprising", "includes", "including" or "contains", "containing" should be read expansively and without limitation. The terms "comprises" and "comprising" should be interpreted as including but not limited to.

[0061] Those skilled in the art to which the application pertains will appreciate that other and further additions, deletions, substitutions, and modifications can be made to the described embodiments. It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only, and is not intended to be limiting, as the scope of the present disclosure will only be limited by the claims. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. Although any methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present disclosure, a limited number of exemplary methods and materials are described herein.

[0062] It should be noted that, as used in herein and in the appended claims, the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise.

[0063] The apparatus, system or method described above, or at least a portion of it, can be implemented in or by various different processors, such as digital data processors or computers, which are programmed or store executable program or software instruction sequences to perform one or more of the steps of the described methods or the functions of the described apparatus or system. The software instructions of such programs can represent algorithms and be encoded in machine-executable form on non-transitory digital data storage media, such as magnetic or optical disks, random access memories (RAM), magnetic hard disks, flash memories and / or read only memories (ROM) to enable various different types of digital data processors or computers to perform one, more or all of the steps of the methods described above or the functions of the systems described herein.

[0064] Certain embodiments disclosed herein, or certain features thereof, can further relate to computer storage products with a non-transitory computer-readable medium having program code thereon for performing various different computer-implemented operations that implement at least portions of the described apparatus, system, or that implement or direct at least some of the steps of the methods set forth herein. Non-transitory medium, as used herein, refers to all computer-readable media that exist solely in a physical form that is not a transitory, propagating signal. Examples of non-transitory computer-readable media include, but are not limited to: magnetic media, such as hard disks, floppy disks, and magnetic tapes; optical media such as CD-ROM disks; magneto-optical media, such as optical disks; and specially configured hardware devices, such as ROM and RAM devices. Examples of program code include machine code produced by a compiler and files containing higher level code, such as an interpreted executable, that has been written in a higher order language.

Claims

1. A processor comprising: one or more circuits to use one or more neural networks to determine an optimal one of a plurality of weighted execution sequences of one or more image reconstruction algorithms based on at least a portion of image data and to use the optimal weighted execution sequence to execute the one or more image reconstruction algorithms on the image data, wherein the optimal one of the one or more reconstruction algorithms is determined by: executing the one or more image reconstruction algorithms on the at least a portion of the image data according to the weighted execution sequences; and selecting one of the weighted execution sequences that provides a best figure of merit (FOM) as the optimal weighted execution sequence.

2. The processor of claim 1, wherein two or more image reconstruction algorithms are executed using weighted execution sequences in parallel processing.

3. The processor of claim 2, wherein the weights of the weighted execution sequences correspond to a number of iterations and the execution sequences correspond to an order of execution of the two or more image reconstruction algorithms.

4. The processor of claim 1, wherein the two or more image reconstruction algorithms are selected from a plurality of image reconstruction algorithms based at least in part on one or more of an image data characteristic, a reconstruction algorithm characteristic, a patient characteristic, or a processing parameter.

5. The processor of claim 1, wherein the one or more circuits cause the neural networks to select the two or more image reconstruction algorithms based at least in part on a determination that respective figures of merit of the two or more image reconstruction algorithms exceed a threshold value.

6. The processor of claim 1, wherein the two or more image reconstruction algorithms are selected based at least in part on a comparison between a plurality of execution sequences performed for a representative slice of an image.

7. The processor of claim 6, wherein at least a portion of the plurality of execution sequences are executed in parallel.

8. A method comprising: causing a processor to use one or more neural networks to determine an optimal one of a plurality of weighted execution sequences of one or more image reconstruction algorithms based on at least a portion of image data and to use the optimal weighted execution sequence to execute the one or more image reconstruction algorithms on the image data, wherein the optimal one of the one or more reconstruction algorithms is determined by: executing the one or more image reconstruction algorithms on the at least a portion of the image data according to the weighted execution sequences; and selecting one of the weighted execution sequences that provides a best figure of merit (FOM) as the optimal weighted execution sequence.

9. The method of claim 8, wherein two or more image reconstruction algorithms are executed using weighted execution sequences in parallel processing.

10. The method of claim 9, wherein the weights of the weighted execution sequences correspond to a number of iterations and the execution sequences correspond to an order of execution of the two or more image reconstruction algorithms.

11. The method of claim 8, wherein the two or more image reconstruction algorithms are selected from a plurality of image reconstruction algorithms based at least in part on one or more of image data characteristics, reconstruction algorithm characteristics, patient characteristics, or processing parameters.

12. The method of claim 8, further comprising: causing the processor to select the two or more image reconstruction algorithms based at least in part on a determination that respective figure of merits of the two or more image reconstruction algorithms exceed a threshold.

13. The method of claim 8, wherein the two or more image reconstruction algorithms are selected based at least in part on a comparison between a plurality of execution sequences performed for representative slices of an image.

14. The method of claim 13, wherein at least a portion of the plurality of execution sequences are performed in parallel.

15. A system comprising: one or more processors that use one or more neural networks to determine an optimal weighted execution sequence of a weighted execution sequence of one or more image reconstruction algorithms based on at least a portion of image data, and use the optimal weighted execution sequence to perform the one or more image reconstruction algorithms on the image data; and a memory to store parameters of the one or more processors, wherein the optimal weighted execution sequence of the one or more reconstruction algorithms is determined by: performing the one or more image reconstruction algorithms on the at least a portion of the image data according to the weighted execution sequence; and selecting one of the weighted execution sequences that provides a best figure of merit (FOM) as the optimal weighted execution sequence.

16. The system of claim 15, wherein two or more image reconstruction algorithms are performed using weighted execution sequences by parallel processing.

17. The system of claim 16, wherein weights of the weighted execution sequences correspond to a number of iterations, and an execution sequence corresponds to an order of performance of the two or more image reconstruction algorithms.

18. The system of claim 15, wherein the two or more image reconstruction algorithms are selected from a plurality of image reconstruction algorithms based at least in part on one or more of image data characteristics, reconstruction algorithm characteristics, patient characteristics, or processing parameters.

19. The system of claim 15, wherein the one or more processors cause a neural network to select the two or more image reconstruction algorithms based at least in part on a determination that respective figure of merits of the two or more image reconstruction algorithms exceed a threshold.

20. The system of claim 15, wherein the two or more image reconstruction algorithms are selected based at least in part on a comparison between a plurality of execution sequences performed for representative slices of an image.

Images