System and method for image conversion

Abstract

The invention relates to a method and a CT imaging system, the method is implemented on a computing device, and the method comprises the following steps: acquiring low-dose image data; obtaining a first neural network model; processing the low dose image data based on a first neural network model to generate virtual high dose image data corresponding to the low dose image data; the first neural network model is obtained by the following steps: obtaining high-dose projection data, wherein the high-dose projection data is generated by scanning an examined object by using a scanner; acquiring low-dose projection data corresponding to the high-dose projection data, wherein the low-dose projection data is acquired from a scanner or is simulated and determined based on the high-dose projection data; generating a high-dose image through a first reconstruction technique based on the high-dose projection data; generating a low-dose image by a second reconstruction technique based on the low-dose projection data; and training a neural network model based on a neural network training algorithm, the high-dose image and the corresponding low-dose image to determine a first neural network model.

Description

[0001] Divisional Statement

[0002] This application is a divisional application of the Chinese application with the application number 201710770592.9, the application date of August 31, 2017, and the title of "System and method of image conversion". TECHNICAL FIELD

[0003] The present disclosure relates generally to an imaging system, and more particularly to a method and system of converting a low-dose image to a high-dose image. BACKGROUND

[0004] Computed tomography (CT) technology is a technology that uses a computer to process a combination of X-ray data taken from different angles to produce 2D or 3D images. CT technology has been widely used in medical diagnosis. In the process of reconstruction of a CT image based on low-dose projection data, noise and / or artifacts (e.g., stair-step artifacts) can appear in the reconstructed CT image. Artifacts can reduce image quality and affect the diagnostic results based on such images. A high-dose CT scan can at least partially eliminate such problems, but at the cost of exposing the scanned subject to too much radiation. It is desirable to provide some systems and methods to produce a high-dose CT image of improved quality based on a low-dose CT scan. SUMMARY

[0005] According to one aspect of the present disclosure, a method of converting a low-dose image to a high-dose image is provided. The method can be implemented on at least one machine, each machine having at least one processor and storage. The method can include: obtaining a first set of projection data associated with a first dose level; reconstructing a first image based on the first set of projection data; determining a second set of projection data based on the first set of projection data, the first set of projection data being associated with a second dose level, the second dose level being lower than the first dose level; reconstructing a second image based on the second set of projection data; training a first neural network model based on the first image and the second image, the trained first neural network model being configured to convert a third image to a fourth image, the fourth image exhibiting a lower noise level than the third image and corresponding to a higher dose level.

[0006] In some embodiments, the first neural network model can be constructed based on at least one of a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory (LSTM), or a generative adversarial network (GAN).

[0007] In some embodiments, the first image can be reconstructed based on an iterative reconstruction algorithm using first reconstruction parameters.

[0008] In some embodiments, the second image can be reconstructed using the second reconstruction parameters based on an analytical reconstruction algorithm or an iterative reconstruction algorithm. In some embodiments, the second reconstruction parameters can be at least partially different from the first parameters.

[0009] In some embodiments, the first image can be reconstructed by applying at least one of a larger slice thickness, a larger reconstruction matrix, or a smaller field of view relative to reconstructing the second image.

[0010] In some embodiments, the second set of projection data can be determined based on at least one of a scan parameter of a scanner that acquired the first set of projection data, an attenuation coefficient related to the object, a noise corresponding to the scanner, a response of a tube, a response of a detector of the scanner, a size of a focus of the scanner, a fly focus of the scanner, an integration time of a detector of the scanner, or a scatter coefficient of the object.

[0011] In some embodiments, determining the second set of projection data can comprise: determining a first distribution of radiation with respect to the second dose level before the radiation traverses the object; determining a second distribution of radiation based on the first distribution of radiation and the first set of projection data after the radiation traverses the object; determining a noise estimate of the scanner; determining the second set of projection data based on the second distribution of radiation and the noise estimate. In some embodiments, determining the noise estimate can comprise detecting a response of a detector in the scanner when the scanner does not emit radiation.

[0012] In some embodiments, training the first neural network model based on the first image and the second image can comprise: extracting a first region from the first image; extracting a second sub-region from the second image corresponding to the first region in the first image, the first region of the first image having the same size as the second region; training the first neural network model based on the first region of the first image and the second region of the second image.

[0013] In some embodiments, training the first neural network model based on the first region of the first image and the second region of the second image can comprise: initializing parameter values of the first neural network model; iteratively determining a value of a cost function related to the parameter values of the first neural network model in each iteration based at least in part on the first region of the first image and the second region of the second image, including updating at least some of the parameter values of the first neural network model based on an updated value of the cost function in a most recent iteration process after each iteration; and determining the trained first neural network model until a condition is satisfied.

[0014] In some embodiments, the condition can comprise a change in the cost function values among a plurality of iterations being below a threshold value, or a threshold number of iterations having been performed.

[0015] In some embodiments, the method may further include training a second neural network model based on the sixth and seventh images. In some embodiments, the sixth and seventh images may be reconstructed based on a third set of projection data. In some embodiments, the image quality of the seventh image may be better than that of the sixth image. Image quality may be related to at least one of contrast and spatial resolution.

[0016] In some embodiments, the third set of projection data may include the first set of projection data.

[0017] In some embodiments, the first image or the first neural network model is at least two-dimensional.

[0018] According to another aspect of this disclosure, a method for converting a low-dose image into a high-dose image is provided. The method can be implemented on at least one machine, each machine having at least one processor and storage. The method may include: acquiring a first set of projection data regarding a first dose level; determining a second set of projection data based on a first neural network model and the first set of projection data, the second set of projection data being correlated with a second dose level higher than the first dose level; generating a first image based on the second set of projection data; and generating a second image based on a second neural network model and the first image.

[0019] In some embodiments, a first neural network model may be generated by: acquiring a third set of projection data about a third dose level; simulating a fourth set of projection data based on the third set of projection data, the fourth set of projection data being related to a fourth dose level below the third dose level; and training the first neural network model based on the third set of projection data and the fourth set of projection data.

[0020] In some embodiments, simulating a fourth set of projection data may include: determining a first distribution of radiation with respect to a fourth dose level before the radiation passes through the object; determining a second distribution of radiation based on the first distribution of radiation and the third set of projection data after the radiation passes through the object; determining a noise estimate of the scanner; and determining the fourth set of projection data based on the second distribution of radiation and the noise estimate.

[0021] In some embodiments, the second neural network can be generated by: acquiring a third image, which is reconstructed based on a fifth set of projection data; acquiring a fourth image, which is reconstructed based on the fifth set of projection data; and training the second neural network model based on the third and fourth images. In some embodiments, the image quality of the fourth image may be superior to that of the third image, and the image quality is related to at least one of contrast and spatial resolution.

[0022] In some embodiments, the fifth set of projection data may include the first set of projection data.

[0023] In some embodiments, the first image or the first neural network model may be at least two-dimensional.

[0024] In some embodiments, the first dose level may be 5 millisieverts (mSv) or higher.

[0025] In some embodiments, the first dose level may be 15 millisieverts (mSv) or higher.

[0026] In some embodiments, the second dose level may be 10% or lower than the first dose level.

[0027] In some embodiments, the second dose level may be 40% or lower of the first dose level.

[0028] According to one aspect of this disclosure, a system for converting a low-dose image into a high-dose image is provided. The system may include at least one processor and executable instructions. When the at least one processor executes the executable instructions, the instructions cause the at least one processor to implement a method. The method may include: acquiring a first set of projection data regarding a first dose level; reconstructing a first image based on the first set of projection data; determining a second set of projection data based on the first set of projection data, the second set of projection data being related to a second dose level lower than the first dose level; reconstructing a second image based on the second set of projection data; training a first neural network model based on the first and second images, the trained first neural network model being configured to convert a third image into a fourth image, the fourth image exhibiting a lower noise level and corresponding to a higher dose level than the third image.

[0029] According to another aspect of this disclosure, a non-transitory computer-readable medium is provided. The non-transitory computer-readable medium may include executable instructions. When executed by at least one processor, the instructions cause the at least one processor to implement a method. The method may include: acquiring a first set of projection data regarding a first dose level; reconstructing a first image based on the first set of projection data; determining a second set of projection data based on the first set of projection data, the second set of projection data relating to a second dose level lower than the first dose level; reconstructing a second image based on the second set of projection data; training a first neural network model based on the first and second images, the trained first neural network model being configured to convert a third image into a fourth image, the fourth image exhibiting a lower noise level and corresponding to a higher dose level than the third image.

[0030] According to one aspect of this disclosure, a system for converting a low-dose image into a high-dose image is provided. The system may include an image data simulation unit. The image data simulation unit may be configured to determine a second set of projection data based on a first set of projection data, wherein the first set of projection data is related to a first dose level, and the second set of projection data is related to a second dose level, the second dose level being lower than the first dose level. The system may further include an image reconstruction unit configured to reconstruct a first image based on the first set of projection data and a second image based on the second set of projection data. The system may further include a neural network training unit configured to train a first neural network model based on the first and second images, the trained first neural network model being configured to convert a third image into a fourth image, the fourth image exhibiting a lower noise level and corresponding to a higher dose level than the third image.

[0031] According to one aspect of this disclosure, a system for converting a low-dose image into a high-dose image is provided. The system may include at least one processor and executable instructions. When the at least one processor executes the executable instructions, the instructions cause the at least one processor to implement a method. The method may include: acquiring a first set of projection data regarding a first dose level; determining a second set of projection data regarding a second dose level higher than the first dose level based on a first neural network model and the first set of projection data; generating a first image based on the second set of projection data; and generating a second image based on a second neural network model and the first image.

[0032] According to another aspect of this disclosure, a non-transitory computer-readable medium is provided. The non-transitory computer-readable medium may include executable instructions. When at least one processor executes these instructions, the instructions cause the at least one processor to implement a method. The method may include: acquiring a first set of projection data regarding a first dose level; determining a second set of projection data based on a first neural network model and the first set of projection data, the second set of projection data being correlated with a second dose level higher than the first dose level; generating a first image based on the second set of projection data; and generating a second image based on a second neural network model and the first image.

[0033] According to one aspect of this disclosure, a system for converting a low-dose image into a high-dose image is provided. The system may include an acquisition module. The acquisition module may be configured to acquire a first set of projection data with respect to a first dose level. The system may further include an image data processing module. The image data processing module may be configured to determine a second set of projection data based on a first neural network model and the first set of projection data, the second set of projection data being correlated with a second dose level higher than the first dose level; generate a first image based on the second set of projection data; and generate a second image based on a second neural network and the first image.

[0034] According to another aspect of this disclosure, a method for training a neural network is provided. The method can be implemented on at least one machine, each machine having at least one processor and storage. The method may include: acquiring a first set of projection data regarding a first dose level; determining a second set of projection data based on the first set of projection data, the second set of projection data relating to a second dose level lower than the first dose level; training a neural network model based on the first set of projection data and the second set of projection data, the trained neural network model being configured to convert a third set of projection data into a fourth set of projection data, the fourth set of projection data having a lower noise level than the third set of projection data.

[0035] According to another aspect of this disclosure, a method for training a neural network is provided. The method can be implemented on at least one machine, each machine having at least one processor and storage. The method may include: acquiring projection data regarding a dose level; reconstructing a first image based on the projection data using first reconstruction parameters; reconstructing a second image based on the projection data using second reconstruction parameters, the second reconstruction parameters being different from the first reconstruction parameters; training a neural network model based on the first and second images, the neural network model being configured to convert a third image into a fourth image, the fourth image exhibiting better image quality compared to the third image, the image quality being related to at least one of contrast and spatial resolution.

[0036] Additional features will be set forth in portions of the following specification, and will become apparent in part to those skilled in the art upon review of the following description and the accompanying drawings, or may be learned by example from production or operation. The features of this disclosure may be realized and achieved through the practice or use of various aspects of the methods, means, and combinations set forth in the detailed examples discussed below. Attached Figure Description

[0037] This disclosure is further described with reference to exemplary embodiments. These exemplary embodiments will be detailed with reference to the accompanying drawings. These embodiments are non-limiting exemplary embodiments, wherein similar reference numerals in various views of the drawings represent similar structures, and wherein:

[0038] Figure 1 This is a schematic diagram of an exemplary CT imaging system described according to some embodiments of the present disclosure.

[0039] Figure 2 This is a schematic diagram of exemplary hardware and / or software components of an exemplary computing device described according to some embodiments of the present disclosure.

[0040] Figure 3This is a schematic diagram of exemplary hardware and / or software components of an exemplary mobile device described according to some embodiments of the present disclosure.

[0041] Figure 4 This is a block diagram of an exemplary processing engine described according to some embodiments of the present disclosure;

[0042] Figure 5 This is a block diagram of an exemplary neural network determination module described according to some embodiments of the present disclosure;

[0043] Figure 6 This is a flowchart illustrating an exemplary process for processing image data according to some embodiments of the present disclosure;

[0044] Figure 7 This is a flowchart illustrating an exemplary process for determining a first neural network model according to some embodiments of the present disclosure;

[0045] Figure 8 This is a flowchart illustrating an exemplary process for simulating low-dose projection data according to some embodiments of the present disclosure;

[0046] Figure 9 This is a flowchart illustrating an exemplary process for determining a second neural network model according to some embodiments of the present disclosure;

[0047] Figure 10 This is a flowchart of an exemplary process 1000 for training a neural network model, as described in some embodiments of this disclosure;

[0048] Figure 11 This is a schematic diagram of an exemplary neural network model described according to some embodiments of the present disclosure. Detailed Implementation

[0049] In the following detailed description, numerous specific details are set forth by way of example to provide a thorough understanding of the relevant disclosure. However, it will be apparent to those skilled in the art that this disclosure may be practiced without these details. In other instances, well-known methods, processes, systems, components, and / or circuits have been described at a relatively high level without detail to avoid unnecessarily obscuring aspects of this disclosure. Various modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the invention. Thus, the invention is not limited to the illustrated embodiments but is accorded the broadest scope consistent with the claims.

[0050] The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not restrictive. As used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should be further understood that the terms “comprising,” “including,” “including,” “containing,” “comprise,” “include,” “include,” and “comprising” in this specification indicate the presence of the claimed features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or components and / or any combination thereof.

[0051] It should be understood that the terms “system,” “engine,” “unit,” “module,” and / or “block” used in this document are a way of distinguishing different levels of components, elements, parts, sections, or assemblies in ascending order. However, these terms may be replaced with other expressions if they can achieve the same purpose.

[0052] Generally, the terms "module," "unit," or "block" refer to logic embodied in a set of hardware, firmware, or software instructions. The modules, units, or blocks described herein may be implemented as hardware and / or software and may be stored on any form of non-transitory computer-readable medium or other storage device. In some embodiments, the software module / unit / block may be compiled and linked into an executable program. It should be understood that a software module can be called from other modules / units / blocks or by itself, and / or may be called in response to a detected event or interrupt. It is configured to be used in a computing device (e.g., Figure 2Software modules / units / blocks executing on the processor 210 shown may be provided on computer-readable media, such as compressed optical discs, digital video discs, flash drives, hard disks, or any other tangible media, or provided as digital downloads (and may be stored raw in compressed or installable formats that require installation, decompression, or decryption before execution). Such software code may be stored, in part or in whole, in the storage device of the executing computing device for execution by the computing device. Software instructions may be embedded in firmware such as erasable programmable read-only memory (EPROM). It should be further understood that hardware modules / units / blocks may be included in connected logical components such as gates and flip-flops, and / or in programmable units such as programmable gate arrays or processors. The modules / units / blocks or computing device functions described herein may be implemented as software modules / units / blocks, but may also be presented as hardware or firmware. Generally, the modules / units / blocks described herein refer to logical modules / units / blocks that can be combined with other modules / units / blocks or divided into submodules / subunits / subblocks, regardless of their physical structure or storage. This description may apply to a system, an engine, or a part thereof.

[0053] It should be understood that when a unit, engine, module, or block is referred to as being "located," "connected to," or "coupled to" other units, engines, modules, or blocks, it can mean being directly located, connected to, or coupled to, or communicating with other units, engines, modules, or blocks, or it can mean that there are intermediate units, engines, modules, or blocks present, unless the context clearly indicates otherwise. As used herein, the term "and / or" includes any and all combinations of one or more of the related listed items.

[0054] These or other features, characteristics, methods of operation, functions of structural components, parts and economic combinations of the product will be more readily understood with reference to the following description and the corresponding drawings, all of which form part of this disclosure. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this disclosure. It should be understood that the drawings are not to scale.

[0055] This document provides systems and components for image processing. In some embodiments, the imaging system may include a single-mode imaging system, such as a computed tomography (CT) system, emission computed tomography (ECT), ultrasound imaging system, X-ray optical imaging system, positron emission tomography (PET) system, or the like, or any combination thereof. In some embodiments, the imaging system may include a multi-mode imaging system, such as a computed tomography-magnetic resonance imaging (CT-MRI) system, a positron emission tomography-magnetic resonance imaging (PET-MRI) system, a single-photon emission computed tomography (SPECT-CT) system, a digital subtraction angiography-computed tomography (DSA-CT) system, etc. It should be noted that the CT imaging system 100 described below is for illustrative purposes only and is not intended to limit the scope of this disclosure.

[0056] For illustrative purposes, this disclosure describes systems and methods for CT image processing. The system can generate CT images based on a neural network model. For example, a neural network model can process low-dose CT image data to generate high-dose CT image data. High-dose CT image data can exhibit better quality compared to low-dose CT image data. The neural network model can be obtained through training based on multiple low-dose images or image data and high-dose images, the high-dose images being reconstructed separately using different reconstruction techniques.

[0057] The following description is provided to aid in a better understanding of CT imaging reconstruction methods and / or systems. It is not intended to limit the scope of this disclosure. A number of variations, alterations, and / or modifications can be made to those skilled in the art based on the teachings of this disclosure. Such variations, alterations, and / or modifications do not depart from the scope of this disclosure.

[0058] Figure 1 This is a schematic diagram of an exemplary CT imaging system 100 described according to some embodiments of the present disclosure. As shown, the CT imaging system 100 may include a scanner 110, a processing engine 120, storage 130, one or more terminals 140, and a network 150. In some embodiments, the scanner 110, processing engine 120, storage 130, and / or terminal 140 may be interconnected and / or communicate with each other via wireless connections (e.g., network 150), wired connections, or combinations thereof. The connections between the components of the CT imaging system 100 are variable. By way of example only, the scanner 110 may be connected to the processing engine 120 via network 150, such as... Figure 1 As shown. As another example, scanner 110 can be directly connected to processing engine 120. As a further example, storage 130 can be connected to processing engine 120 via network 150, as... Figure 1As shown, or directly connected to processing engine 120. As a further example, terminal 140 can connect to processing engine 120 via network 150, such as... Figure 1 As shown, or directly connected to the processing engine 120.

[0059] Scanner 110 can generate or provide image data by scanning an object or a portion thereof. In some embodiments, scanner 110 may include a single-mode scanner and / or a multi-mode scanner. Single-mode scanners may include, for example, computed tomography (CT) scanners, positron emission tomography (PET) scanners, etc. Multi-mode scanners include single-photon emission computed tomography-computed tomography (SPECT-CT) scanners, positron emission tomography-computed tomography (CT-PET) scanners, computed tomography-ultrasound (CT-US) scanners, digital subtraction angiography-computed tomography (DSA-CT) scanners, or the like, or combinations thereof. In some embodiments, image data may include projection data, images of the object, etc. Projection data may be raw data generated by scanner 110 scanning the object, or data generated by forward projection of an image related to the object. In some embodiments, the object may include a body, substance, object, or the like, or combinations thereof. In some embodiments, the object may include specific parts of the body, such as the head, chest, abdomen, or the like, or combinations thereof. In some embodiments, the object may include a specific organ or region of interest, such as the esophagus, trachea, bronchi, stomach, gallbladder, small intestine, colon, bladder, ureter, uterus, fallopian tubes, etc.

[0060] In some embodiments, scanner 110 may include a tube, detector, etc. The tube may generate and / or emit one or more radiation beams directed toward an object, according to one or more scanning parameters. Radiation as used herein may include particle rays, photon rays, or the like, or any combination thereof. In some embodiments, radiation may include multiple radiating particles (e.g., neutrons, atoms, electrons, muons, heavy ions, etc.), multiple radiating photons (e.g., X-rays, gamma rays, ultraviolet light, lasers, etc.), or the like, or combinations thereof. Exemplary scanning parameters may include tube current / voltage, detector integration time, size of the tube's focal spot, detector response, tube response, collimation width, slice thickness, slice gap, field of view (FOV), etc. In some embodiments, scanning parameters may be related to the dose level of radiation emitted from the tube. As used herein, the dose level of radiation may be defined by the CT dose index (CTDI), effective dose, dose-length product, etc. The CT dose index (CTDI) may refer to the radiant energy of radiation associated with a single slice along the long axis (e.g., the axial direction) of scanner 110. The dose-length product can refer to the total radiant energy received by the object during the integral scan. The effective dose can refer to the radiant energy received by a specific area of ​​the object during the integral scan.

[0061] The detector of scanner 110 can detect one or more radiation beams emitted from the tube. In some embodiments, the detector of scanner 110 may include one or more detector units that can detect the distribution of the radiation beams emitted from the tube. In some embodiments, the detector of scanner 110 may be connected to data conversion circuitry configured to convert the distribution of the detected radiation beams into image data (e.g., projection data). The image data may correspond to the dose level of the detected radiation beams. In some embodiments, the dose level of the detected radiation beams may include noise presented in the image data. For example, the higher the radiation dose level, the lower the noise level presented in the image data may be relative to the true signal (reflecting the actual anatomical structure). The lower the radiation dose level, the higher the noise level presented in the image data may be.

[0062] Processing engine 120 can process data and / or information acquired by scanner 110, storage 130, and / or terminal 140. For example, processing engine 120 can reconstruct an image based on projection data generated by scanner 110. As another example, processing engine 120 can determine one or more neural network models configured to process and / or transform images. In some embodiments, processing engine 120 can be a single server or a cluster of servers. The server cluster can be centralized or distributed. In some embodiments, processing engine 120 can be local or remote. For example, processing engine 120 can access information and / or data from scanner 110, storage 130, and / or terminal 140 via network 150. As another example, processing engine 120 can be directly connected to scanner 110, terminal 140, and / or storage 130 to access information and / or data. In some embodiments, processing engine 120 can be implemented on a cloud platform. For example, the cloud platform can include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, interconnected cloud, multi-cloud, or the like, or a combination thereof. In some embodiments, the processing engine 120 may be executed by a computing device 200, the computing device 200 having one or more combinations Figure 2 The component being described.

[0063] Storage 130 may store data, instructions, and / or any other information. In some embodiments, storage 130 may store data from processing engine 120, terminal 140, and / or interaction device 150. In some embodiments, storage 130 may store data and / or instructions that processing engine 120 may execute or use to perform the exemplary methods described herein. In some embodiments, storage may include mass storage, erasable storage, volatile read-write memory, read-only memory (ROM), or the like, or any combination thereof. Exemplary mass storage may include a hard disk, optical disk, solid-state drive, etc. Exemplary erasable storage may include a flash drive, floppy disk, optical disk, memory card, compact disk, magnetic tape, etc. Exemplary volatile read-write memory may include random access memory (RAM). Exemplary RAM may include dynamic RAM (DRAM), double-rate synchronous dynamic RAM (DDR SDRAM), static RAM (SRAM), thyristor RAM (T-RAM), and zero-capacitance RAM (Z-RAM), etc. Exemplary ROMs may include mask ROMs (MROMs), programmable ROMs (PROMs), erasable programmable ROMs (EPROMs), electrically erasable programmable ROMs (EEPROMs), compressed optical disc ROMs (CD-ROMs), and digital universal disk ROMs, etc. In some embodiments, storage 130 may be implemented on a cloud platform, as described in other parts of this disclosure.

[0064] In some embodiments, storage 130 may be connected to network 150 to communicate with one or more other components of CT imaging system 100 (e.g., processing engine 120, terminal 140, etc.). One or more components of CT imaging system 100 may access data or instructions stored in storage 130 via network 150. In some embodiments, storage 130 may be part of processing engine 120.

[0065] Terminal 140 may connect to and / or communicate with scanner 110, processing engine 120, and / or memory 130. For example, terminal 140 may obtain processed images from processing engine 120. As another example, terminal 140 may obtain image data acquired via scanner 110 and transmit the image data to processing engine 120 for processing. In some embodiments, terminal 140 may include mobile device 140-1, tablet computer 140-2, laptop computer 140-3, or the like, or combinations thereof. For example, mobile device 140-1 may include mobile phone, personal data assistant (PDA), gaming device, navigation device, point-of-sale (POS) device, laptop computer, tablet computer, desktop, or the like, or any combination thereof. In some examples, terminal 140 may include input devices, output devices, etc. Input devices may include alphanumeric and other keys that can be input via keyboard, touchscreen (e.g., haptic or haptic feedback), voice input, eye-tracking input, brain monitoring system, or any other comparable input mechanism. Input information received via an input device may be transmitted to the processing engine 120 for further processing, for example, via a bus. Other types of input devices may include cursor control devices, such as a mouse, trackball, or arrow keys. Output devices may include a display, speakers, printer, or the like, or combinations thereof. In some embodiments, the terminal 140 may be part of the processing engine 120.

[0066] Network 150 may include any suitable network that facilitates the exchange of information and / or data between the CT imaging system 100 and the CT imaging system 100. In some embodiments, one or more components of the CT imaging system 100 (e.g., scanner 110, processing engine 120, storage 130, terminal 140, etc.) may exchange information and / or data with other components of the CT imaging system 100 via network 150. For example, processing engine 120 may acquire image data from scanner 110 via network 150. As another example, processing engine 120 may acquire user instructions from terminal 140 via network 150. Network 150 may be or include public networks (e.g., the Internet), private networks (e.g., local area networks (LANs), wide area networks (WANs), etc.), wired networks (e.g., Ethernet), wireless networks (e.g., 802.11 networks, Wi-Fi networks, etc.), cellular networks (e.g., LTE networks), frame relay networks, virtual private networks (VPNs), satellite networks, telephone networks, routers, hubs, switches, server computers, and / or any combination thereof. For example, network 150 may include a cable network, wired network, fiber optic network, telecommunications network, intranet, wireless local area network (WLAN), metropolitan area network (MAN), public switched telephone network (PSTN), Bluetooth network, ZigBee network, near field communication (NFC) network, or the like, or any combination thereof. In some embodiments, network 150 may include one or more network access points. For example, network 150 may include wired and / or wireless network access points such as base stations and / or internet switching points, through which one or more components of the CT imaging system 100 may connect to network 150 to exchange data and / or information.

[0067] These descriptions are for illustrative purposes only and do not limit the scope of this disclosure. Many alternatives, modifications, and variations will be apparent to those skilled in the art. The features, structures, methods, and other characteristics of the exemplary embodiments described herein may be combined in different ways to obtain other and / or alternative exemplary embodiments. For example, storage 130 may be data storage for a cloud computing platform including public cloud, private cloud, community cloud, hybrid cloud, etc. However, these variations and modifications do not depart from the scope of this disclosure.

[0068] Figure 2 These are exemplary hardware and / or software components of an exemplary computing device 200 that implements an exemplary processing engine 120 according to some embodiments described in this disclosure. Figure 2 As shown, the computing device 200 may include a processor 210, storage 220, input / output devices (I / O) 230, and communication ports 240.

[0069] Processor 210 can execute computer instructions (e.g., program code) and implement the functions of processing engine 120 according to the techniques described herein. Computer instructions may include, for example, routines, programs, objects, components, data structures, procedures, modules, and functions that perform the specific functions described herein. For example, processor 210 can process image data obtained from CT scanner 110, terminal 140, storage 130, and / or any other component of CT imaging system 100. In some embodiments, processor 210 may include one or more hardware processors, such as microcontrollers, microprocessors, reduced instruction set computers (RISC), application-specific integrated circuits (ASICs), application-specific instruction set processors (ASIPs), central processing units (CPUs), graphics processing units (GPUs), physical processing units (PPUs), microcontroller units, digital signal processors (DSPs), field-programmable gate arrays (FPGAs), advanced RISC machines (ARMs), programmable logic devices (PLDs), any circuit or processor capable of performing one or more functions, or the like, or any combination thereof.

[0070] For illustrative purposes only, the computing device 200 describes only one processor. However, it should be noted that the computing device 200 of this disclosure may also include multiple processors, and therefore the operations and / or methods performed by one processor described in this disclosure may also be performed jointly or separately by multiple processors. For example, if the processor of the computing device 200 of this disclosure performs operations A and B, it should be understood that operations A and B may also be performed jointly or separately by two or more different processors in the computing device 200 (e.g., a first processor performs operation A, a second processor performs operation B, or the first and second processors jointly perform operations A and B).

[0071] Storage 220 may store data / information from CT scanner 110, terminal 140, storage 130, and / or any other component of CT imaging system 100. In some embodiments, storage 220 may include mass storage, erasable storage, volatile read-write memory, read-only memory (ROM), or the like, or any combination thereof. For example, mass storage may include disks, optical disks, solid-state drives, etc. Erasable storage may include flash drives, floppy disks, optical disks, memory cards, compressed optical disks, magnetic tapes, etc. Volatile read-write memory may include random access memory (RAM). RAM may include dynamic RAM (DRAM), double-rate synchronous dynamic RAM (DDR SDRAM), static RAM (SRAM), thyristor RAM (T-RAM), and zero-capacitance RAM (Z-RAM), etc. ROM may include mask ROM (MROM), programmable ROM (PROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), optical disc ROM (CD-ROM), and digital universal disk ROM, etc. In some embodiments, storage 220 may store one or more programs and / or instructions to perform the exemplary methods described herein. For example, storage 220 may store a program for determining a regularization item for processing engine 120.

[0072] Input / output device 230 can input and / or output signals, data, information, etc. In some embodiments, input / output device 230 allows a user to interact with processing engine 120. In some embodiments, input / output device 230 may include an input device and an output device. Examples of input devices may include a keyboard, mouse, touchscreen, microphone, or the like, or combinations thereof. Examples of output devices may include a display, speaker, printer, projector, or the like, or combinations thereof. Examples of displays may include a liquid crystal display (LCD), a light-emitting diode (LED) based display, a flat panel display, a curved screen, a television device, a cathode ray tube (CRT), a touchscreen, or the like, or combinations thereof.

[0073] Communication port 240 can be connected to a network (e.g., network 150) to facilitate data communication. Communication port 240 can establish a connection between processing engine 120, CT scanner 110, terminal 140, and / or storage 130. The connection can be a wired connection, a wireless connection, any other communication connection that enables data transmission and / or data reception, and / or any combination of these connections. Wired connections may include, for example, cables, optical fibers, telephone lines, or the like, or any combination thereof. Wireless connections may include, for example, Bluetooth links, Wi-Fi links, Global System for Microwave Access (WiMax) links, Wireless Local Area Network (WLAN) links, ZigBee links, mobile network links (e.g., 3G, 4G, 5G, etc.), or the like, or combinations thereof. In some embodiments, communication port 240 may be and / or include standard communication ports, such as RS232, RS485, etc. In some embodiments, communication port 240 may be a specially designed communication port. For example, communication port 240 may be designed according to the Medical Digital Imaging and Communication (DICOM) protocol.

[0074] Figure 3 This is a schematic diagram of exemplary hardware and / or software of an exemplary mobile device 300 implementing terminal 140, as described in some embodiments of this disclosure. Figure 3 As shown, the mobile device 300 may include a communication platform 310, a display 320, a graphics processing unit (GPU) 330, a central processing unit (CPU) 340, an input / output device 350, a memory 360, and a storage 390. In some embodiments, the mobile device 300 may also include any other suitable components, including but not limited to a system bus or controller (not shown). In some embodiments, a mobile operating system 370 (e.g., iOS, Android, Windows Phone, etc.) and one or more applications 380 may be loaded from storage 390 into storage 360 ​​for execution by the CPU 340. Applications 380 may include a browser or any other suitable mobile application for receiving and presenting information about image processing or other information from processing engine 120. User interaction with the information flow may be achieved through input / output device 350 and provided via network 150 to processing engine 120 and / or other components of CT imaging system 100.

[0075] To implement the various modules, units, and functions of this disclosure, a computer hardware platform may be used as the hardware platform for one or more elements described herein. A computer with a user interface element may be used to implement a personal computer (PC) or any other type of workstation or terminal device. If properly programmed, the computer may also act as a server.

[0076] Figure 4This is a block diagram of an exemplary processing engine 120 described according to some embodiments of the present disclosure. The processing engine 120 may include an acquisition module 410, a control module 420, a neural network determination module 430, an image data processing module 440, and a storage module 450. The processing engine 120 may be implemented on various components (e.g., Figure 2 The processor 210 of the computing device 200 shown. For example, at least a portion of the processing engine 120 may be implemented in Figure 2 The computing device shown or Figure 3 The mobile device shown.

[0077] The acquisition module 410 can acquire image data. The acquisition module 410 can acquire image data from the scanner 110 or a storage device (e.g., storage 130, storage 220, storage 390, memory 360, storage module 450, or the like, or combinations thereof). The image data may include projection data, images, etc. In some embodiments, the acquisition module 410 can send the acquired image data to other modules or units of the processing engine 120 for further processing. For example, the acquired image data may be sent to the storage module 450 for storage. As another example, the acquisition module 410 can send image data (e.g., projection data) to the image data processing module 440 to reconstruct the image.

[0078] Control module 420 can control the operation of acquisition module 410, neural network determination module 430, image processing module 440, and / or storage module 450 (e.g., by generating one or more control parameters). For example, control module 420 can control acquisition module 410 to acquire image data. As another example, control module 420 can control image data processing module 440 to process the image data acquired by acquisition module 410. As yet another example, control module 420 can control neural network determination module 430 to train a neural network model. In some embodiments, control module 420 can receive real-time commands or retrieve predetermined commands provided by, for example, a user (e.g., a doctor) or system 100 to control one or more operations of acquisition module 410, neural network determination module 430, and / or image data processing module 440. For example, control module 420 can adjust image data processing module 440 to generate object images according to real-time instructions and / or predetermined instructions. In some embodiments, control module 420 can communicate with one or more other modules of processing engine 120 to exchange information and / or data.

[0079] The neural network determination module 430 can determine one or more neural network models. For example, the neural network determination module 430 can determine a first neural network model, which is configured to, for example, reduce the noise level of an image. As another example, the neural network determination module 430 can determine a second neural network model, which is configured to, for example, increase the contrast of an image by performing an image enhancement operation on the image. In some embodiments, the neural network determination module 430 can send the determined neural network model to one or more other modules for further processing or application. For example, the neural network determination module 430 can send the neural network model to the storage module 450 for storage. As another example, the neural network determination module 430 can send the neural network model to the image data processing module 440 for image processing.

[0080] Image data processing module 440 can process information provided by various modules of processing engine 120. Processing module 440 can process image data acquired by acquisition module 410, image data retrieved from storage module 450, etc. In some embodiments, image data processing module 440 can reconstruct images based on image data according to reconstruction techniques, generate reports including one or more images and / or other relevant information, and / or perform any other image reconstruction functions according to various embodiments of this disclosure.

[0081] Storage module 450 may store image data, models, control parameters, processed image data, or the like, or combinations thereof. In some embodiments, storage module 450 may store one or more programs and / or instructions executable by the processor of processing engine 120 to perform the exemplary methods described in this disclosure. For example, storage module 450 may store programs and / or instructions executable by the processor of processing engine 120 to acquire image data, reconstruct images based on image data, train neural network models, and / or display any intermediate results or result images.

[0082] In some embodiments, Figure 4 One or more modules shown can be Figure 1 This is implemented in at least a portion of the exemplary CT imaging system 100 shown. For example, the acquisition module 410, control module 420, storage module 450, neural network determination module 430, and / or image data processing module 440 may be integrated into a console (not shown). Through the console, a user can set parameters for the scanned object, control the imaging process, control parameters for image reconstruction, observe the reconstructed image, etc. In some embodiments, the console may be implemented via processing engine 120 and / or terminal 140. In some embodiments, neural network determination module 430 may be integrated into terminal 140.

[0083] In some embodiments, the processing engine 120 does not include a neural network determination module 430. One or more neural network models determined by another device may be stored in the system 100 (e.g., storage 130, storage 220, storage 390, memory 360, storage module 450, etc.) or on an external device accessible through the processing engine 120 via, for example, network 150. In some embodiments, such a device may include the same or similar components as the neural network determination module 430. In some embodiments, the neural network determination module 430 may store one or more neural network models determined by another device and accessible to one or more components of the system 100 (e.g., image reconstruction unit 520, image data simulation unit 540, etc.). In some embodiments, neural network models applicable to this disclosure may be determined by the system 100 (or including, for example, a portion of the processing engine 120) or by an external device accessible to the system 100 (or including, for example, a portion of the processing engine 120). See, for example, [link to relevant documentation]. Figure 7 , Figure 9 and Figure 10 And its description.

[0084] Figure 5 This is a block diagram of an exemplary neural network determination module 430 described according to some embodiments of the present disclosure. As shown, the neural network determination module 430 may include an image reconstruction unit 520, an image data simulation unit 540, a neural network training unit 560, and a storage unit 580. The neural network determination module 430 can be configured in various components (e.g., such as...) Figure 2 This is implemented on the processor 210 of the computing device 200 shown. For example, at least a portion of the neural network determination module 430 may be implemented on... Figure 2 The computing device shown or Figure 3 The mobile device shown.

[0085] Image reconstruction unit 520 can reconstruct one or more images based on one or more reconstruction techniques. In some embodiments, image reconstruction unit 520 can reconstruct a first image (e.g., a high-dose image) based on a first reconstruction technique. Image reconstruction unit 520 can reconstruct a second image (e.g., a low-dose image) based on a second reconstruction technique. The first and second reconstruction techniques can be the same or different. In some embodiments, image reconstruction unit 520 can send the reconstructed image to other units or blocks of neural network determination module 430 for further processing. For example, image reconstruction unit 520 can send the reconstructed image to neural network training unit 560 to train a neural network model. As another example, image reconstruction unit 520 can send the reconstructed image to storage unit 580 for storage.

[0086] Image data simulation unit 540 can simulate image data. In some embodiments, image data simulation unit 540 can simulate virtual low-dose image data based on high-dose image data acquired by CT scan. As used herein, virtual low-dose image data may correspond to a lower dose level than real high-dose image data. In some embodiments, image data simulation unit 540 can send simulated image data to other units and / or blocks of neural network determination module 430 for further processing. For example, simulated image data can be sent to image reconstruction unit 520 to generate an image. As another example, simulated transmission data can be sent to neural network training unit 560 for training a neural network model.

[0087] The neural network training unit 560 can train a neural network model. In some embodiments, the neural network training unit 560 can train a first neural network model, which is configured, for example, to reduce the noise level in an image. Such a neural network model can be obtained using multiple high-dose images and corresponding low-dose images. In some embodiments, the neural network training unit 560 can train a second neural network model, which is configured, for example, to increase the contrast of an image. Such a neural network model can be obtained using multiple images with higher contrast and corresponding images with lower contrast. As used herein, two images can be considered to correspond to each other when they involve the same region of the subject. By way of example only, two corresponding images may differ in one or more aspects, including, for example, a high-dose image and a low-dose image, an image with high contrast and an image with low contrast, or the like, or combinations thereof.

[0088] In some embodiments, the neural network training unit 560 may further include an initialization block 562, an extraction block 564, a computation block 566, and a decision block 568. The initialization block 562 may initialize the neural network model. For example, the initialization block 562 may construct an initial neural network model. As another example, the initialization block 562 may initialize one or more parameter values ​​of the initial neural network model. The extraction block 564 may extract information from one or more training images (e.g., high-dose images and low-dose images). For example, the extraction block 564 may extract features about one or more regions from these training images. The computation block 566 may perform computational functions, for example, during the training of the neural network model. For example, the computation block 566 may compute one or more parameter values ​​of the neural network model updated during iterative training. The decision block 568 may perform decision functions, for example, during the training of the neural network model. For example, the decision block 568 may determine whether conditions are met during the training of the neural network model.

[0089] Storage unit 580 can store information, for example, about training a neural network model. In some embodiments, information related to training a neural network model may include images used to train the neural network model, algorithms used to train the neural network model, parameters of the neural network model, etc. For example, storage unit 580 can store training images (e.g., high-dose images and low-dose images) according to certain criteria. Training images can be stored or uploaded to storage unit 580 based on the dimension of the training images. For illustrative purposes, two-dimensional (2D) images or three-dimensional (3D) images can be stored as 2D or 3D matrices comprising multiple elements (e.g., pixels or voxels). The elements of a 2D matrix are arranged in storage unit 580 such that each row of elements is stored sequentially in storage unit 580, with each row corresponding to the length of the 2D image, so that elements in the same row are adjacent to each other in storage unit 580. The elements of a 3D matrix are arranged in storage unit 580 such that multiple 2D matrices constituting the 3D matrix are stored sequentially in storage unit 580, and then the rows and / or columns of each 2D matrix are stored sequentially in storage unit 580. Storage unit 580 may be a memory that stores data to be processed by a processing device such as a CPU or GPU. In some embodiments, storage unit 580 may be a memory accessed by one or more GPUs, or a memory accessed only by a specific GPU.

[0090] It should be noted that the above description of the processing module 430 is provided for illustrative purposes only and is not intended to limit the scope of this disclosure. Various modifications or variations can be made by those skilled in the art based on the teachings of this disclosure. However, such modifications and variations do not depart from the scope of this disclosure. For example, the image reconstruction unit 520 and the image data simulation unit 540 may be integrated into a single unit.

[0091] Figure 6 This is a flowchart of an exemplary process 600 for processing image data, described according to some embodiments of the present disclosure. In some embodiments, Figure 6 One or more processing operations of the process 600 shown for processing image data can be performed in... Figure 1 Implemented on the CT imaging system 100 shown. For example, Figure 6 The process 600 shown can be stored in storage 130 as instructions and processed by processing engine 120 (e.g., as...). Figure 2 The processor 210 of the computing device 200 shown is, for example Figure 3 The GPU 330 or CPU 340 of the mobile device 300 shown is invoked and / or executed.

[0092] In step 602, low-dose image data can be acquired. Step 602 can be performed by acquisition module 410. As used herein, low-dose image data can refer to image data (e.g., projection data, images, etc.) corresponding to a first dose level. In some embodiments, low-dose image data may include low-dose projection data. In some embodiments, low-dose image data may include low-dose images. In some embodiments, low-dose image data may include two-dimensional (2D) image data, three-dimensional (3D) image data, four-dimensional (4D) image data, or image data of other dimensions. In some embodiments, low-dose image data may be real image data obtained from an object scanner (e.g., scanner 110) by scanning an object at a low dose level (e.g., a first dose level). In some embodiments, low-dose image data may be virtual image data obtained by simulating other image data, such as high-dose image data. In some embodiments, low-dose image data may be acquired from storage 130, terminal 140, storage module 450, and / or any other external storage device.

[0093] In step 604, a first neural network model is obtained. Step 604 can be performed by the neural network determination module 430. In some embodiments, the first neural network model may be predefined (e.g., provided by a CT scanner manufacturer, an entity specializing in image processing, an entity accessing training data, etc.). In some embodiments, the first neural network model may be configured to process image data (e.g., low-dose image data obtained in step 602). Exemplary image data processing may include transformations, modifications, and / or conversions. For example, the first neural network model may be configured to convert low-dose image data into high-dose image data corresponding to the low-dose image data. As another example, the first neural network model may be configured to reduce the noise level in image data (e.g., low-dose image data obtained in step 602). In some embodiments, the first neural network model may be constructed based on convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory (LSTMs), generative adversarial networks (GANs), or the like, or combinations thereof. See, for example, [link to relevant documentation]. Figure 11 And its description. In some embodiments, the first neural network model can be configured as a two-dimensional (2D) model, a three-dimensional (3D) model, a four-dimensional (4D) model, or a model of any other dimension. In some embodiments, it can be based on Figure 7 The process shown in 700 is used to determine the first neural network model.

[0094] In 606, low-dose image data can be processed based on a first neural network model to generate (virtual) high-dose image data corresponding to the low-dose image data. Operation 606 can be performed by image data processing module 440. In some embodiments, the (virtual) high-dose image data corresponding to the low-dose image data may exhibit a lower noise level than the low-dose image data. As used herein, (virtual) high-dose image data corresponding to low-dose image data refers to image data (e.g., projection data, images, etc.) corresponding to a second dose level. The second dose level of the (virtual) high-dose image data may be greater than the first dose level of the low-dose image data. Corresponding (virtual) high-dose image data and low-dose image data refer to representations of the same object or the same part or region of the object being examined (e.g., patient, tissue, etc.). In some embodiments, the (virtual) high-dose image data may include high-dose projection data. In some embodiments, the (virtual) high-dose image data may include high-dose images. In some embodiments, the high-dose image data may include two-dimensional (2D) image data, three-dimensional (3D) image data, four-dimensional (4D) image data, or image data of other dimensions.

[0095] In operation 608, a second neural network model may be obtained. Operation 608 may be performed by the neural network model determination module 430. In some embodiments, the second neural network model may be predefined (e.g., provided by a CT scanner manufacturer, an entity specializing in image processing, an entity accessing training data, etc.). In some embodiments, the second neural network model may be configured to process image data (e.g., (virtual) high-dose image data generated in 606). Exemplary image data processing may include transformations, modifications, and / or conversions, etc. For example, the second neural network model may be configured to perform image data augmentation operation 606 on image data (e.g., (virtual) high-dose image data generated in operation 606). In some embodiments, the second neural network model may be constructed based on convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory (LSTMs), generative adversarial networks (GANs), or the like or combinations thereof. See, for example, [link to relevant documentation]. Figure 11 And its description. In some embodiments, the second neural network model can be configured as a two-dimensional (2D) model, a three-dimensional (3D) model, a four-dimensional (4D) model, or a model of any other dimension. In some embodiments, it can be based on Figure 9 The process shown in 900 is used to determine the second neural network model.

[0096] In step 610, the (virtual) high-dose image data can be post-processed based on a second neural network model. Operation 610 can be performed by the image data processing module 440. In some embodiments, the post-processed high-dose image data can exhibit higher quality than the high-dose image data obtained in step 608. For example, the post-processed high-dose image data corresponding to the high-dose image data can exhibit higher contrast than the high-dose image data obtained in step 608.

[0097] In step 612, post-processed high-dose image data can be output. Operation 612 can be performed by image data processing module 440. In some embodiments, the post-processed high-dose image data can be output to terminal 140 for display, for example, in the form of an image. In some embodiments, the post-processed high-dose image data can be output to storage 130 and / or storage module 508 for storage.

[0098] It should be noted that the above description is provided for illustrative purposes only and is not intended to limit the scope of this disclosure. Various modifications and variations can be made to the teachings of this disclosure by those skilled in the art. However, such modifications and variations do not depart from the scope of this disclosure. For example, process 600 may include preprocessing operations, such as noise reduction, on the low-dose image data prior to operation 604. As another example, operations 606 and / or 608 may be unnecessary and omitted. In some embodiments, process 600 may also include outputting the high-dose image data generated in 606.

[0099] Figure 7 This is a flowchart of an exemplary process 700 for determining a first neural network model, as described in some embodiments of this disclosure. Figure 6 The illustrated operation 604 can be performed according to process 700. In some embodiments, the first neural network model can be configured to convert low-dose image data into high-dose image data. The first neural network model can be determined by training the neural network model using multiple low-dose images and multiple corresponding high-dose images. The low-dose images and corresponding high-dose images can be reconstructed based on different reconstruction techniques, respectively. In some embodiments, the method used to determine the first neural network model... Figure 7 One or more operations of the process 700 shown can be performed in Figure 1 This is implemented in the CT imaging system 100 shown. For example, Figure 7 The process 700 shown can be stored in storage 130 as instructions and processed by processing engine 120 (e.g., as...). Figure 2 The processor 210 of the computing device 200 shown is, for example Figure 3 The GPU 330 or CPU 340 of the mobile device 300 shown is invoked and / or executed.

[0100] In step 702, high-dose projection data can be acquired. Operation 702 can be performed by the image data simulation unit 540. In some embodiments, the high-dose projection data may include 2D projection data, 3D projection data, etc. In some embodiments, the high-dose projection data can be obtained from a scanner (e.g., scanner 110) and generated by scanning the object being examined. In some embodiments, the high-dose projection data can be generated by forward projection of an image. In some embodiments, the high-dose projection data can be acquired from storage 130, terminal 140, storage module 450, and / or any other external storage device.

[0101] In step 704, low-dose projection data corresponding to the high-dose projection data can be acquired. Operation 704 can be performed by the image data simulation unit 540. As used herein, the corresponding low-dose projection data and high-dose projection data refer to representations of the same object or the same portion of an object (e.g., patient, tissue, etc.). In some embodiments, the high-dose projection data may correspond to a first dose level, and the low-dose projection may correspond to a second dose level. The first dose level may be greater than the second dose level. In some embodiments, the first and second dose levels may vary depending on clinical needs (e.g., tissue type). For example, in a liver scan, the first dose level may be equal to or greater than 5 mSv, or 10 mSv, or 15 mSv, etc. The second dose level may be less than 15 mSv, or 10 mSv, or 5 mSv, etc. The ratio of the second dose level to the first dose level may range from 5% to 40%, such as 10%, 15%, 20%, 25%, 30%, etc. As another example, in a chest scan, the first dose level may be equal to or greater than 2 mSv, or 7 mSv, etc. The second dose level can be below 7 mSv or 2 mSv, etc. In some embodiments, the ratio of the first dose level to the estimated effective dose can be equal to or greater than 1%, or 5%, or 10%, or 25%, or 50%, or 100%, or 150%, etc. The ratio of the second dose level to the estimated effective dose can be equal to or less than 1%, or 5%, or 10%, or 25%, etc. The estimated effective dose can be the dose level received in the region of interest of CT imaging in the integral scan schedule. The dose level of the estimated effective dose can be in the range of, for example, 0.1 mSv to 1.5 mSv.

[0102] In some embodiments, low-dose projection data can be obtained from a scanner (e.g., scanner 110). In some embodiments, low-dose projection data can be obtained from storage 130, terminal 140, storage module 450, and / or any other external storage device. In some embodiments, low-dose projection data can be determined based on high-dose projection data. For example, low-dose projection data can be determined through simulation based on high-dose projection data.

[0103] It should be noted that the projection data may relate to the distribution of radiation emitted from a scanner (e.g., scanner 110) after the radiation has passed through the object being inspected. The projection data may include noise associated with the scanner (e.g., electronic noise from the detector in scanner 110). The distribution of the radiation beam emitted from the scanner may be related to scanning conditions including one or more scanning parameters, such as tube current / voltage, detector integration time, tube focal spot size, detector response, tube response, collimation width, etc. Different scanning conditions can be configured to produce radiation beams with different dose levels. For example, a higher tube current / voltage may result in a higher dose level of the produced radiation beam. In some embodiments, such as in combination... Figure 8 The method described above can acquire low-dose projection data corresponding to a second dose level based on high-dose projection data corresponding to a first dose level. In some embodiments, both the high-dose projection data and the corresponding low-dose projection data can be acquired by a scanner (e.g., scanner 110) by scanning the object being inspected.

[0104] In step 706, a high-dose image is generated based on high-dose projection data using a first reconstruction technique. Operation 706 can be performed by the image reconstruction unit 520. In some embodiments, the first reconstruction technique may include iterative reconstruction techniques, analytical reconstruction techniques, or the like, or combinations thereof. Exemplary iterative reconstruction techniques may include Algebraic Reconstruction Technique (ART), Simultaneous Iterative Reconstruction Technique (SIRT), Simultaneous Algebraic Reconstruction Technique (SART), Adaptive Statistical Iterative Reconstruction Technique (ASIR), Model-Based Iterative Reconstruction Technique (MAIR), Sinogram Confirmed Iterative Reconstruction Technique (SAFIR), or the like, or combinations thereof. Exemplary analytical reconstruction techniques may include applying the FDK algorithm, the Katsevich algorithm, etc., or combinations thereof. In some embodiments, one or more reconstruction parameters may be determined prior to the high-dose image reconstruction process. Exemplary reconstruction parameters may include field of view (FOV), slice thickness, reconstruction matrix, slice gap, convolution kernel, or the like, or combinations thereof. For example, compared to the reconstruction of a low-dose image, a high-dose image can be reconstructed by applying a larger slice thickness, a larger reconstruction matrix, a smaller FOV, etc.

[0105] In some implementations, a high-dose image may present a first image quality. As used herein, the first image quality may be defined by the high dose or the first noise level of the first image. In some embodiments, the first noise level of a high-dose image reconstructed based on the same image data but with different reconstruction techniques may differ. For example, the first noise level of a high-dose image reconstructed using an iterative reconstruction technique may be lower than the first noise level of a high-dose image reconstructed using an analytical reconstruction technique. In some embodiments, the first noise level of a high-dose image reconstructed based on the same image data and the same reconstruction technique but with different reconstruction parameters may differ. For example, the first noise level of a high-dose image reconstructed using a larger slice thickness, a larger reconstruction matrix, a smoother reconstruction kernel, and / or a smaller field of view may be lower than the first noise level of a high-dose image reconstructed based on the same reconstruction technique using a smaller slice thickness, a smaller reconstruction matrix, a sharper reconstruction kernel, and / or a larger field of view.

[0106] In some embodiments, during the reconstruction of high-dose images, denoising techniques or filtering kernel functions used to perform image smoothing can be used to reduce the initial noise level of the high-dose images. Exemplary denoising techniques may include adaptive filtering algorithms, Kal filtering algorithms, or the like, or combinations thereof. Exemplary adaptive filtering algorithms may include least mean square (LMS) adaptive filtering algorithms, recursive least squares (RLS) adaptive filtering algorithms, transform domain adaptive filtering algorithms, affine projection algorithms, conjugate gradient algorithms, subband decomposition-based adaptive filtering algorithms, QR decomposition-based adaptive filtering algorithms, etc. In some embodiments, denoising techniques may include applying a denoising model. Exemplary denoising models may include spatial domain filter models, transform domain filter models, morphological noise filter models, or the like, or combinations thereof. Exemplary spatial domain filter models may include field-average filter models, median filter models, Gaussian filter models, or the like, or combinations thereof. Exemplary transform domain filter models may include Fourier transform models, Walsh-Hadamard transform models, cosine transform models, KL transform models, wavelet transform models, or the like, or combinations thereof. In some embodiments, the denoising model may include a partial differential model or a variational model, such as the Perona-Malik (PM) model, the Total Variation (TV) model, or the like, or a combination thereof. Exemplary filtering kernel techniques for performing image smoothing functions may include applying, for example, linear smoothing filters (e.g., block filters, mean filters, Gaussian filters, etc.), nonlinear smoothing filters (e.g., median filters, sequential statistical filters, etc.).

[0107] In step 708, a low-dose image is generated based on low-dose projection data using a second reconstruction technique. Operation 708 may be performed by image reconstruction unit 520. Exemplary second reconstruction techniques may include iterative reconstruction techniques, analytical reconstruction techniques, or the like, or combinations thereof, as described in other parts of this disclosure. In some embodiments, the second reconstruction technique may be different from or the same as the first reconstruction technique. For example, the second reconstruction technique may include analytical reconstruction techniques, and the first reconstruction technique may include iterative reconstruction techniques. As another example, the second reconstruction technique and the first reconstruction technique may include the same iterative reconstruction technique. In some embodiments, one or more reconstruction parameters for reconstructing the low-dose image may be determined. Exemplary reconstruction parameters may include field of view (FOV), slice thickness, reconstruction matrix, slice gap, convolution kernel, or the like, or combinations thereof. For example, compared to the reconstruction of a high-dose image, a low-dose image can be reconstructed by applying a smaller slice thickness, a smaller reconstruction matrix, a larger FOV, a sharper reconstruction kernel, etc.

[0108] In some embodiments, the low-dose image may exhibit a second image quality. As used herein, the second image quality may be defined by the low dose or the second noise level of the second image. The second noise level of the low-dose image may be higher than the first noise level of the corresponding high-dose image. In some embodiments, the second noise levels of multiple low-dose images reconstructed from the same image data using different reconstruction techniques may differ. For example, the second noise level of a low-dose image reconstructed using analytical reconstruction techniques may be higher than the second noise level of a low-dose image reconstructed using iterative reconstruction techniques. In some embodiments, the second noise levels of multiple low-dose images reconstructed from the same image data using the same reconstruction technique but with different reconstruction parameters may differ. For example, the second noise level of a low-dose image reconstructed using a smaller slice thickness, a smaller reconstruction matrix, a larger field of view, etc., may be higher than the second noise level of a low-dose image based on the same reconstruction technique but using a larger slice thickness, a larger reconstruction matrix, a smaller field of view, etc.

[0109] In some embodiments, the second noise level of a low-dose image can be increased by using a filtering kernel technique for performing image sharpening functions. Exemplary filtering kernel techniques for performing image sharpening functions may include applying, for example, linear sharpening filters (e.g., Laplacian operators, high-frequency boost filters, etc.), nonlinear sharpening filters (e.g., gradient-based sharpening filters, max-min sharpening transforms, etc.). In some embodiments, the second noise level of a low-dose image can be reduced by using filtering kernel techniques for performing image smoothing functions and / or by using denoising techniques, as described in other parts of this disclosure.

[0110] In 710, a first neural network model can be determined based on high-dose and low-dose images. In some embodiments, operation 710 can be performed by a neural network training unit 560. In some embodiments, the first neural network model can be configured to improve image quality by, for example, reducing the noise level of the image, increasing the contrast of the image, or the like, or a combination thereof. In some embodiments, the effectiveness of the first neural network model for improving image quality (e.g., the function of reducing noise level) may be related to the difference between the high-dose and low-dose images. As used herein, the difference between the high-dose and low-dose images may refer to the difference between a first noise level in the high-dose image and a second noise level in the low-dose image. The greater the difference between the first noise level in the high-dose image and the second noise level in the low-dose image, the more effective the first neural network model may be in improving image quality by reducing the noise level of the image generated based on the first neural network model. As another example, the lower the first noise level in the high-dose image, the more effective the first neural network may be in improving image quality by reducing the noise level of the low-dose image generated based on the first neural network. The higher the second noise level in the low-dose image, the more effective the first neural network may be in improving image quality by reducing the noise level of the low-dose image generated based on the first neural network. In some embodiments, a first neural network model can be determined by training a neural network model based on a neural network training algorithm, high-dose images, and corresponding low-dose images. Exemplary neural network training algorithms may include gradient descent, Newton's algorithm, quasi-Newton's algorithm, Levenberg-Marquardt algorithm, conjugate gradient algorithm, or the like, or combinations thereof.

[0111] In some embodiments, process 700 can be repeated for multiple training datasets to improve or optimize the first neural network model. The multiple training datasets include high-dose and low-dose projection data and images corresponding to different groups. In different rounds of process 700 performed based on different pairs of high-dose and low-dose images, high-dose images can be obtained using the same or different reconstruction techniques. Similarly, in different rounds of process 700 performed based on different pairs of high-dose and low-dose images, low-dose images can be obtained using the same or different reconstruction techniques.

[0112] It should be noted that the above description is provided for illustrative purposes only and is not intended to limit the scope of this disclosure. Various modifications or variations can be made by those skilled in the art based on the teachings of this disclosure. However, such modifications and variations do not depart from the scope of this disclosure. For example, process 700 may include operations for preprocessing high-dose projection data and / or low-dose projection data. As another example, operations 702 and 704 may be performed simultaneously or in conjunction with... Figure 7The operations are executed in the reverse order shown, and / or operations 706 and 708 can be executed simultaneously or in conjunction with... Figure 7 The process is performed in the reverse order shown. In some embodiments, process 700 may further include storing high-dose and low-dose images in storage 130, terminal 140, storage module 450, storage unit 580, and / or other external storage devices. In some embodiments, operations 706 and 708 may be omitted. The first neural network model can be determined directly based on the high-dose projection data and the low-dose projection data. Therefore, the first neural network model can be configured to transform the original projection dataset into a different projection dataset. This different projection dataset may exhibit a lower noise level than the original projection dataset.

[0113] Figure 8 This is a flowchart of an exemplary process 800 for generating simulated low-dose projection data, as described in some embodiments of this disclosure. Figure 7 The operation 704 shown can be performed according to process 800. In some embodiments, the generation of simulated projection data may be related to multiple factors, including, for example, scanning parameters of scanner 110, attenuation or absorption coefficient of the object, scattering coefficient of the object, noise corresponding to scanner 110, or the like, or combinations thereof. Scanning parameters may include, for example, detector response, tube response, filtering of antiscattering grid, tube current value, tube voltage value, collimation width, exposure time (e.g., scan time), focal spot size, flyfocus mode, detector integration time, etc. In some embodiments, for determining the first neural network model Figure 8 One or more operations of the process 800 shown can be performed in Figure 1 This is implemented in the CT imaging system 100 shown. For example, Figure 8 The process 800 shown can be stored in storage 130 as instructions and processed by processing engine 120 (e.g., as...). Figure 2 The processor 210 of the computing device 200 shown is, for example Figure 3 The CPU 340 of the mobile device 300 shown calls and / or executes.

[0114] In step 802, a first distribution of the first radiation can be determined. The first radiation can be generated by scanning using a scanner (e.g., scanner 110) under first scanning conditions. Operation 802 can be performed by acquisition module 410. The first distribution of the first radiation (e.g., a radiation beam including X-ray photons) can refer to the incident intensity distribution of the first radiation before passing through the object being inspected. In some embodiments, the first distribution of the first radiation can be determined using the detector unit of the scanner. For example, when scanner 110 performs an air scan without an object placed between the X-ray generator and the detector unit, the first distribution of the first radiation can be detected by the detector unit of scanner 110.

[0115] In some embodiments, the first distribution of the first radiation may be related to a first dose level of the first radiation. The first dose level of the first radiation can be determined based on first scanning conditions. The first scanning conditions can be defined by values ​​of a plurality of first scanning parameters, including, for example, tube current value, tube voltage value, collimation width, exposure time (e.g., scan time), filtering of the antiscattering grid, detector response, tube (or radiation source) response, tube focal spot size, fly-focus mode, detector integration time, etc. The first dose level of the first radiation can be determined based on the values ​​of one or more first scanning parameters. For example, a higher tube current may result in a higher first dose level.

[0116] In step 804, a second distribution of the second radiation from the scanner can be determined based on a first distribution of the first radiation. The second radiation can be a virtual radiation simulated according to second scanning conditions. Operation 802 can be performed by the image data simulation unit 540. Similarly, the second distribution of the second radiation (e.g., X-ray photons) can refer to the incident intensity distribution of the second radiation before passing through the object being inspected.

[0117] In some embodiments, the second distribution of the second radiation may be related to a second dose level of the second radiation. The second dose level of the second radiation can be determined based on second scanning conditions. The second scanning conditions can be defined by values ​​of a plurality of second scanning parameters, including, for example, tube current, tube voltage, collimation width, exposure time (e.g., scan time), etc. The second dose level of the second radiation beam can be determined based on the values ​​of the plurality of second scanning parameters.

[0118] In some embodiments, a second distribution of radiation corresponding to a second dose level (or a second scanning condition) can be determined based on a first distribution of first radiation corresponding to a first dose level (or a first scanning condition). For example, a relationship between the radiation distribution (e.g., the number distribution of particles / photons in the radiation beam) and the scanning conditions (e.g., values ​​of scanning parameters as described above) can be determined based on the first scanning conditions and the first distribution of the first radiation, and then the second distribution of the second radiation beam can be determined based on that relationship. For example, based on the first distribution of the first radiation, the second distribution of the second radiation can be determined based on that relationship, according to the difference between the first and second scanning conditions.

[0119] In step 806, a third distribution of the second radiation under the second scanning condition can be determined based on the second radiation and high-dose projection data. Operation 802 can be performed by the image data simulation unit 540. As used herein, the third distribution of the second radiation beam (e.g., X-ray photons) can refer to the outgoing intensity distribution of the second radiation after it passes through the object being examined under the second scanning condition. In some embodiments, the third distribution of the second radiation beam can be determined based on the second distribution of the second radiation and the attenuation distribution of the object. In some embodiments, the attenuation distribution of the object can be related to the distribution of attenuation coefficients or absorption coefficients of different portions of the object. The distribution of attenuation coefficients or absorption coefficients can be determined by reconstructing an attenuation map of the object based on high-dose projection data. The third distribution of the second radiation beam can then be determined based on the second distribution of the second radiation and the attenuation distribution of the second radiation corresponding to the object.

[0120] In step 808, a noise estimate related to the scanner can be determined. Operation 802 can be performed by the image data simulation unit 540. In some embodiments, the noise estimate related to the scanner can be determined based on the scanner's detector unit. For example, when no radiation is emitted from the scanner, noise estimation can be performed by detecting data using the detector unit in the scanner. The noise may include electronic noise generated by circuitry connected to the detector unit.

[0121] In step 810, low-dose projection data can be determined based on a third distribution and noise estimation. Operation 802 can be performed by the image data simulation unit 540. In some embodiments, the low-dose projection data can refer to projection data corresponding to the second dose level described in steps 804 and 806. In some embodiments, a Poisson distribution associated with the second radiation can be determined based on a third distribution of the second radiation. The Poisson distribution can be determined as an approximation (e.g., by curve fitting) of the third distribution. Low-dose projection data can then be determined based on the Poisson distribution and noise estimation.

[0122] In some embodiments, the Poisson distribution and noise estimation can be mixed at a specific ratio to obtain low-dose projection data. For example, the noise estimation can be represented by a first matrix including a plurality of first elements. The Poisson distribution associated with the third distribution can be represented by a second matrix including a plurality of second elements. The plurality of first elements and the plurality of second elements can be multiplied by a first weight value and a second weight value, respectively. The low-dose projection data can be determined by a weighted sum of the weighted first elements and the weighted second elements. In some embodiments, the first weight value and the second weight value can be in the range of 0 to 1.

[0123] It should be noted that the above description is provided for illustrative purposes only and is not intended to limit the scope of this disclosure. Various modifications or variations can be made by those skilled in the art based on the teachings of this disclosure. However, such modifications and variations do not depart from the scope of this disclosure. For example, operations 802 and 808 can be performed simultaneously. As another example, operation 808 can be performed before operation 802.

[0124] Figure 9 This is a flowchart of an exemplary process 900 for determining a second neural network model, as described in some embodiments of this disclosure. Figure 6 The operation 608 shown can be performed according to process 900. In some embodiments, the second neural network model can be configured to improve the quality of the image (e.g., by improving contrast). The second neural network model can be determined by training a neural network model with multiple images of relatively high quality and multiple corresponding images of relatively low quality. Multiple images of relatively high quality and multiple corresponding images of relatively low quality can be reconstructed respectively based on the same image data but different reconstruction techniques. In some embodiments, the method used to determine the first neural network model... Figure 9 One or more operations of the process 900 shown can be performed in Figure 1 This is implemented in the CT imaging system 100 shown. For example, Figure 9 The process 900 shown can be stored in storage 130 as instructions and processed by processing engine 120 (e.g., as...). Figure 2 The processor 210 of the computing device 200 shown is, for example Figure 3 The CPU 340 of the mobile device 300 shown calls and / or executes.

[0125] In operation 902, projection data can be acquired. Operation 902 can be performed by acquisition module 410. In some embodiments, the projection data may include high-dose projection data, as described in conjunction with operation 702.

[0126] In step 904, a first image can be generated based on the projection data using a third reconstruction technique. Operation 904 can be performed by the image reconstruction unit 520. The third reconstruction technique includes iterative reconstruction techniques, analytical reconstruction techniques, or the like, or combinations thereof, as described in other parts of this disclosure. In some embodiments, one or more reconstruction parameters described in other parts of this disclosure can be determined prior to the reconstruction process of the first image. See, for example... Figure 7 Operations 706 and / or 708 and their related descriptions.

[0127] In step 906, a second image is generated based on the projection data using a fourth reconstruction technique. Operation 906 can be performed by the image reconstruction unit 520. The second image can have higher contrast than the first image.

[0128] In some embodiments, the fourth reconstruction technique may differ from the third reconstruction technique. For example, the third reconstruction technique may include analytical reconstruction techniques, and the fourth reconstruction technique may include iterative reconstruction techniques. In some embodiments, the reconstruction parameters used in the third reconstruction technique may differ from those used in the fourth reconstruction technique. For example, compared to the fourth reconstruction technique, the third reconstruction technique may use a larger slice thickness, a smaller reconstruction matrix, and / or a larger field of view. In some embodiments, the third and fourth reconstruction techniques may be of the same type but based on different reconstruction parameters. For example, the third and fourth reconstruction techniques may be iterative reconstruction techniques but based on different reconstruction parameters. In some embodiments, the third and fourth reconstruction techniques may be different types based on the same or different reconstruction parameters.

[0129] A denoising process or filtering kernel used to perform image smoothing can reduce image contrast. A filtering kernel used to perform image sharpening can increase image contrast. In some embodiments, the denoising process or filtering kernel used to perform image smoothing as described in other parts of this disclosure can be used in a third reconstruction technique. See, for example... Figure 7 Operations 706 and / or 708 and their related descriptions. Additionally or alternatively, a filter kernel for performing image sharpening functions as described in other parts of this disclosure may be used in the fourth reconstruction technique. See, for example... Figure 7 Operations 706 and / or 708 and their related descriptions. Therefore, the second image can exhibit higher contrast than the first image.

[0130] In step 908, a second neural network model can be determined based on the first image and the second image. In some embodiments, operation 908 can be performed by a neural network model training unit 560. In some embodiments, the second neural network model can be configured to improve image quality by, for example, increasing the contrast of the image. The effectiveness of the second neural network for improving image contrast can be related to the difference between the first image and the second image. As used herein, the difference between the first image and the second image can refer to the difference between the first contrast of the first image and the second contrast of the second image. The greater the difference between the first contrast of the first image and the second contrast of the second image, the more effective the second neural network may be in improving image quality by increasing the contrast of the image generated based on the second neural network model. As another example, the lower the first contrast of the first image, the more effective the second neural network may be in improving image quality by increasing the contrast of the image generated based on the second neural network model. The higher the second contrast of the second image, the more effective the second neural network may be in improving image quality by increasing the contrast of the image generated based on the second neural network model. In some embodiments, the second neural network model can be determined by training a neural network model based on a neural network training algorithm, multiple first images, and corresponding second images. Exemplary neural network training algorithms may include gradient descent, Newton's algorithm, quasi-Newton's algorithm, Levenberg-Marquardt algorithm, conjugate gradient algorithm, or the like, or combinations thereof.

[0131] In some embodiments, process 900 can be repeated for multiple training data sets including different projection data to improve or optimize the second neural network model. In different rounds of process 900 performed based on different pairs of high-dose and low-dose (or first and second) images, the high-dose (or first) image can be obtained based on the same or different reconstruction techniques. In different rounds of processing 900 performed based on different pairs of high-dose and low-dose (or first and second) images, the low-dose (or second) image can be obtained based on the same or different reconstruction techniques.

[0132] It should be noted that the above description is provided for illustrative purposes only and is not intended to limit the scope of this disclosure. Various modifications or variations can be made by those skilled in the art based on the teachings of this disclosure. However, such modifications and variations do not depart from the scope of this disclosure. For example, process 900 may include preprocessing the projection data. As another example, operations 904 and 906 may be performed simultaneously, or in conjunction with... Figure 9 The reverse order is shown.

[0133] Figure 10 This is a flowchart illustrating an exemplary process 1000 for training a neural network model according to some embodiments of this disclosure. Figure 7The operation shown in 710 and / or as follows Figure 9 The operation 908 shown can be performed according to process 1000. In some embodiments, the training of the neural network model is as follows: Figure 10 One or more operations of the process 1000 shown can be performed in Figure 1 This is implemented in the CT imaging system 100 shown. For example, Figure 10 The process 1000 shown can be stored in storage 130 as instructions and processed by processing engine 120 (e.g., as...). Figure 2 The processor 210 of the computing device 200 shown is, for example Figure 3 The CPU 340 of the mobile device 300 shown calls and / or executes.

[0134] In operation 1002, a pair of images, including a third image and a fourth image, can be acquired. Operation 1002 can be performed by the acquisition module 410. As used herein, the third image and the fourth image can refer to two images representing the same object or the same region (e.g., patient, tissue, etc.) of an object to be examined. In some embodiments, the third image and the fourth image can respectively correspond to, for example, Figure 7 The low-dose images and high-dose images are described above. In some embodiments, the third and fourth images may correspond to, respectively, the low-dose images and high-dose images described above. Figure 9 The first and second images are shown.

[0135] In step 1004, a neural network model including one or more parameters can be initialized. Operation 1004 can be performed by initialization module 562. In some embodiments, the initialization of the neural network model may include constructing the neural network model based on: convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory (LSTM), generative adversarial network (GAN), or the like, or combinations thereof, such as... Figure 11 And its description. In some embodiments, the neural network model may include multiple layers, such as an input layer, multiple hidden layers, and an output layer. The multiple hidden layers may include one or more convolutional layers, one or more batch normalization layers, one or more activation layers, fully connected layers, cost function layers, etc. Each of the multiple layers may include multiple nodes.

[0136] In some embodiments, the parameters of a neural network model may include the size of the convolutional kernel, the number of layers, the number of nodes in each layer, the connection weights between two connected nodes, and the bias vectors associated with the nodes. The connection weights between two connected nodes can be configured to represent a portion of the node's output value as the input value of another connected node. In some embodiments, the connection weights of the neural network model can be initialized to random values ​​in the range of -1 to 1. In some embodiments, the weights of all connections in the neural network model can have the same value in the range of -1 to 1, such as 0. The bias vectors associated with the nodes can be configured to control the output values ​​of nodes deviating from the origin. In some embodiments, the bias vectors of nodes in the neural network model can be initialized to random values ​​in the range of 0 to 1. In some embodiments, the parameters of the neural network model can be initialized based on Gaussian randomization algorithms, Havier's algorithm, etc.

[0137] In step 1006, a first region can be extracted from the third image. Operation 1006 can be performed by extraction block 564. In some embodiments, the first region can be extracted based on, for example, the size of the first region, the position of the first region, etc. For example, a first position can be determined in the first image, and then a first region with a specific size can be extracted at the first position in the first image. In some embodiments, the first region can be extracted based on a random sampling algorithm. Exemplary random sampling algorithms may include acceptance / rejection sampling algorithms, importance sampling algorithms, the Metropolis-Hasting algorithm, the Gibbs sampling algorithm, etc. In some embodiments, the first region can be extracted based on instructions provided by the user via terminal 140. For example, the user can determine the coordinates of a first position in the first image and the specific size of the first region, and then extraction block 564 can extract the first region based on the first position and the specific size of the first region.

[0138] In step 1008, a second region corresponding to the first region can be extracted from the fourth image. Operation 1008 can be performed by extraction block 564. As used herein, a second region corresponding to the first region can refer to a first region and a second region having the same size and each located at the same position in the third and fourth images. In some embodiments, the second region can be extracted based on the first region. For example, the third image can be divided into multiple first image blocks according to a segmentation rule such as uniform segmentation. The multiple first image blocks can be numbered according to a numbering rule, such as the position of each of the multiple first image blocks. A first block with a specific number can be extracted from the multiple first image blocks and designated as the first region. The fourth image can be divided into multiple second image blocks using the same segmentation rule as the first image. Each of the multiple second image blocks can be numbered using the same numbering rule as the first image. A second block with the same number as the extracted first region can be extracted from the multiple second image blocks and designated as the second region. As another example, the position of the first / second region relative to the third / fourth image can be related to the location of the first / second region stored in storage, such as storage unit 580. The second region relative to the fourth image can be determined based on the position of the first region relative to the third image in the storage.

[0139] In step 1010, the value of the cost function (also known as the loss function) can be determined. Operation 1010 can be performed by computation block 566. The cost function can be configured to evaluate the difference between a test value (e.g., a first region of a third image) and a desired value (e.g., a second region of a fourth image) of the neural network. In some embodiments, the first region of the third image can be transmitted via an input layer (e.g., ...). Figure 11 The input layer 1120 shown is input into the neural network model. The first region of the third image can be obtained from the first hidden layer of the neural network model (e.g., as shown in Figure 1120). Figure 11 The conventional layer 1140-1 shown is passed to the last hidden layer of the neural network model. The first region of the third image can be processed in each of the multiple hidden layers. For example, the first region of the input third image can be processed by one or more conventional layers (e.g., such as...). Figure 11 The conventional layer 1140-1 shown is used for processing. One or more conventional layers can be configured to perform image transformation, image enhancement, image denoising, or any other operation on a first region of the first image based on parameters associated with nodes in one or more conventional layers. The first region of the third image, processed by multiple hidden layers prior to the cost function layer, can be input to the cost function layer. The value of the cost function layer can be determined based on the first region of the third image and the second region of the fourth image, where the first region of the third image has undergone processing by several layers prior to the cost function layer.

[0140] At 1012, it is determined whether a first condition is met. Operation 1012 can be executed by decision block 568. If the first condition is met, process 1012 can proceed to operation 1016. If the first condition is not met, process 1000 can proceed to 1014. The first condition can provide an indication of whether the neural network model has been sufficiently trained. In some embodiments, the first condition may relate to the value of the cost function. For example, the first condition can be met if the value of the cost function is minimum or less than a threshold (e.g., a constant). As another example, the first condition can be met if the value of the cost function converges. In some embodiments, convergence can be considered to have occurred if the change in the value of the cost function over two or more consecutive iterations is equal to or less than a threshold (e.g., a constant). In some embodiments, convergence can be considered to have occurred if the difference between the value of the cost function and the target value is equal to or less than a threshold (e.g., a constant). In some embodiments, the first condition can be met when a specified number of iterations related to a first region of a third image and a second region of a fourth image have been performed during training.

[0141] In step 1014, one or more parameters of the neural network model can be updated. Operation 1014 can be performed by initialization module 562. In some embodiments, the parameter values ​​of at least some nodes can be adjusted until the value of the cost function associated with the first region of the third image satisfies a first condition. In some embodiments, the parameters of the neural network model can be adjusted based on the backpropagation (BP) algorithm. Exemplary backpropagation (BP) algorithms may include stochastic gradient descent, Adam, Adagrad, Adadelta, RMSprop, or the like, or combinations thereof.

[0142] At 1016, it is determined whether the second condition is met. Operation 1016 can be executed by decision block 568. If the second condition is met, process 1000 can proceed to 1018. If the second condition is not met, process 1000 can return to 1004, whereby another first region can be extracted from the third image. In some embodiments, the second condition can be met if a specified number of first and second regions have been processed in association with a neural network model.

[0143] In step 1018, the updated neural network model is determined. Operation 1018 can be performed by initialization block 562. In some embodiments, the updated neural network model can be determined based on the updated parameters.

[0144] In some embodiments, process 1000 can be repeated for multiple training data sets including different pairs of third and fourth images to improve or optimize the neural network model. In different rounds of process 1000 performed based on different pairs of third and fourth images, the third image can be obtained using the same or different reconstruction techniques. In different rounds of process 1000 performed based on different pairs of third and fourth images, the fourth image can be obtained using the same or different reconstruction techniques. Except for the first round of process 1000, in subsequent rounds of process 1000, the initialization of the neural network model in 1004 can be performed based on the updated parameters of the neural network model obtained in the previous round.

[0145] It should be noted that the above description is provided for illustrative purposes only and is not intended to limit the scope of this disclosure. Various modifications or variations can be made to the teachings of this disclosure by those skilled in the art. However, such modifications and variations do not depart from the scope of this disclosure. For example, process 1000 can be repeatedly performed based on multiple third and fourth images to obtain a first neural network model and / or a second neural network model. The training process can be performed until a termination condition is met. An exemplary termination condition is that a specific number of pairs of third and fourth images have been analyzed.

[0146] Figure 11 This is a schematic diagram of an exemplary convolutional neural network (CNN) model described according to some embodiments of the present disclosure.

[0147] A CNN model may include an input layer 1120, multiple hidden layers 1140, and an output layer 1160. The multiple hidden layers 1140 may include one or more convolutional layers, one or more rectified linear unit (ReLU) layers, one or more pooling layers, one or more fully connected layers, or the like, or a combination thereof.

[0148] For illustrative purposes, several exemplary hidden layers 1140 of a CNN model are shown, including convolutional layers 1140-1, pooling layers 1140-2, and fully connected layers 1140-N. As described in assembly process 708, the neural network training unit 560 may acquire a low-dose image as input to the CNN model. The low-dose image may be represented as a two-dimensional (2D) or three-dimensional (3D) matrix comprising multiple elements (e.g., pixels or voxels). Each of the multiple elements in the matrix may have a value representing a feature of the element (also referred to as a pixel / voxel value).

[0149] Convolutional layer 1140-1 may include multiple kernels (e.g., A, B, C, and D). These multiple kernels can be used to extract features from a low-dose image. In some embodiments, each of the multiple kernels may filter a portion (e.g., a region) of the low-dose image to produce specific features corresponding to that portion of the low-dose image. These features may include low-level features (e.g., edge features, texture features), high-level features, or complex features computed based on the kernels.

[0150] Pooling layer 1140-2 can take the output of convolutional layer 1140-1 as input. Pooling layer 1140-2 may include multiple pooling nodes (e.g., E, F, G, and H). These multiple pooling nodes can be used to sample the output of convolutional layer 1140-1, thereby reducing the computational burden of data processing in the CT imaging system 100 and increasing data processing speed. In some embodiments, neural network training unit 560 may reduce the size of the matrix corresponding to the low-dose image in pooling layer 1140-2.

[0151] Fully connected layers 1140-N may include multiple neurons (e.g., O, P, M, and N). These multiple neurons may be connected to multiple nodes from previous layers such as pooling layers. In fully connected layers 1140-N, neural network training unit 560 may determine multiple vectors corresponding to the multiple neurons based on features of the low-dose image, and further weight these multiple vectors with multiple weighting coefficients.

[0152] In the output layer 1160, the neural network training unit 560 can determine the output, such as a high-dose image, based on multiple vectors and weight coefficients obtained from the fully connected layer 708.

[0153] It should be noted that CNN models can be modified when applied to different conditions. For example, during training, a loss function (also referred to herein as a cost function) layer can be added to specify the bias between the predicted output (e.g., the predicted high-dose image) and the true label (e.g., a reference high-dose image corresponding to the low-dose image).

[0154] In some embodiments, the neural network training unit 560 can access multiple processing units, such as GPUs, within the CT imaging system 100. These multiple processing units can perform parallel processing in certain layers of the CNN model. Parallel processing can be performed in such a way that computations at different nodes within a layer of the CNN model can be distributed among two or more processing units. For example, one GPU can run computations corresponding to kernels A and B, while another (or more) GPUs can run computations in convolutional layer 1140-1 corresponding to kernels C and D. Similarly, computations corresponding to different nodes in other types of layers within the CNN model can be performed in parallel by multiple GPUs.

[0155] Having described the basic concepts, it will be readily apparent to those skilled in the art upon reading this detailed disclosure that the preceding detailed disclosure is intended to be illustrative only and not restrictive. Various alternatives, improvements, and modifications are possible and are intended for those skilled in the art, although not expressly stated herein. These alternatives, improvements, and modifications are intended to be suggested in this disclosure and fall within the spirit and scope of the exemplary embodiments of this disclosure.

[0156] Furthermore, certain terms have been used to describe embodiments of this disclosure. For example, the terms "one embodiment," "an embodiment," and / or "some embodiments" mean that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of this disclosure. Therefore, it should be emphasized and understood that two or more references to "one embodiment" or "an alternative embodiment" in various parts of this specification do not necessarily refer to the same embodiment. Moreover, specific features, structures, or characteristics may be suitably combined in one or more embodiments of this disclosure.

[0157] Furthermore, those skilled in the art will understand that aspects of this disclosure can be described and illustrated in any of the many patentable classes or contexts: any new and useful process, machine, product, or composition of matter, or any new and useful improvement thereof. Therefore, aspects of this disclosure can be implemented entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or a combination of software and hardware, all of which are generally referred to herein as “units,” “modules,” or “systems.” Furthermore, aspects of this disclosure can take the form of a computer program product embodied on one or more computer-readable media having computer-readable program code thereon.

[0158] Computer-readable signal media may include propagated data signals containing computer-readable program code, such as in baseband or as part of a carrier wave. Such propagated signals may take any form, including electromagnetic, optical, or similar, or any suitable combination thereof. Computer-readable signal media may be any computer-readable medium other than computer-readable storage media, which can communicate, propagate, or transmit a program for use by an instruction execution system, apparatus, or device, or communicate, propagate, or transmit a program relating to an instruction execution system, apparatus, or device. Any suitable medium, including wireless, wired, fiber optic cable, radio frequency (RF), or similar, or any suitable combination thereof, may be used to transmit program code contained on the computer-readable signal medium.

[0159] Computer program code used to perform the operations of various aspects of this disclosure can be written in any combination of one or more programming languages, including: object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, or the like; conventional programming languages ​​such as the "C" programming language, Visual Basic, Fortran 2103, Perl, COBOL 2102, PHP, ABAP; dynamic programming languages ​​such as Python, Ruby, and Groovy, and other programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, a remote computer can be connected to the user's computer via any type of network (including a local area network (LAN) or a wide area network (WAN), or the connection can be to an external computer (e.g., via the Internet through an Internet service provider) or in a cloud computing environment or as a service such as Software as a Service (SaaS).

[0160] Furthermore, the order of the listed processing elements or sequences, or the numbers, letters, or other designations used for this purpose, is not intended to limit the claimed processes and methods to any particular order, unless specified in the claims. While the foregoing disclosure has discussed various useful embodiments currently considered to be part of this disclosure by way of various examples, it should be understood that such details are merely for that purpose, and the appended claims are not limited to the disclosed embodiments, but rather are intended to cover modifications and equivalent arrangements within the spirit and scope of the disclosed embodiments. For example, while the implementation of the various components described above may be embodied in a hardware device, it may also be implemented solely in a software manner, such as an installation on an existing server or mobile device.

[0161] Similarly, it should be understood that in the above description of embodiments of this disclosure, various features are sometimes grouped together in one embodiment, drawing, or description thereof for the purpose of making the disclosure more fluent and aiding in understanding one or more of the various inventive embodiments. However, this method of disclosure should not be construed as reflecting an intention that the claimed subject matter requires more features than expressly recited in each claim. On the contrary, inventive embodiments contain fewer features than all the features of a single foregoing disclosed embodiment.

[0162] In some embodiments, the numbers used to describe and claim certain embodiments of the present application expressing quantities or properties will be understood to be modified in certain circumstances by the terms “about,” “approximately,” or “substantially.” For example, “about,” “approximately,” or “substantially” may indicate a variation of ±20% of the value they describe, unless otherwise stated. Thus, in some embodiments, the numerical parameters set forth in the written description and appended claims are approximate values ​​that may vary depending on the desired properties sought to be obtained by a particular embodiment. In some embodiments, numerical parameters should be interpreted based on the number of significant figures reported and by applying common rounding techniques. Although the wide range of numerical ranges and parameters set forth in some embodiments of the present application are approximate values, the numerical values ​​set forth in specific embodiments are reported as precisely as possible.

[0163] Every patent, patent application, patent application publication, and other material (e.g., articles, books, specifications, publications, documents, articles, or the like) cited herein is incorporated in its entirety for all purposes, except for any historical records of prosecution documents, anything inconsistent with or conflicting with this document, and anything that limits the maximum scope of protection of the present or later claims of this document. For example, if any cited material contains descriptions, definitions, and / or uses of terminology that are inconsistent with or conflict with this document, then the descriptions, definitions, and / or terminology in this document shall prevail.

[0164] Finally, it should be understood that the embodiments disclosed in this application are illustrative of the principles of the embodiments of this application. Other modifications that can be used are within the scope of this application. Therefore, as examples and not limitations, alternative configurations of the embodiments of this application can be made based on the teachings herein. Therefore, the embodiments of this application are not limited to exactly as shown and described.

Claims

1. A method implemented on a computing device having at least one processor, at least one computer-readable storage medium, and a communication port connected to an imaging device, the method comprising: Acquire low-dose image data; Obtain the first neural network model; The low-dose image data is processed based on the first neural network model to generate virtual high-dose image data corresponding to the low-dose image data; The first neural network model was obtained in the following way: Acquire high-dose projection data, which is generated by scanning the object being inspected using a scanner; Acquire low-dose projection data corresponding to the high-dose projection data, wherein the low-dose projection data is obtained from a scanner or determined by simulation based on the high-dose projection data; Based on the high-dose projection data, a high-dose image is generated using a first reconstruction technique; Based on the low-dose projection data, a low-dose image is generated using a second reconstruction technique; The first neural network model is determined by training a neural network model based on the neural network training algorithm, the high-dose image, and the corresponding low-dose image.

2. The method of claim 1, wherein the low-dose projection data is determined based on the high-dose projection data through simulation: A first distribution of the first radiation is determined, the first distribution being obtained when the scanner performs an air scan with no object placed between the X-ray generator and the detector unit, and a first dose level of the first radiation is determined according to the first scan conditions; A second distribution of second radiation from the scanner is determined based on a first distribution of the first radiation, the second radiation being a virtual radiation simulated according to a second scanning condition, the second distribution of the second radiation referring to the incident intensity distribution of the second radiation before passing through the object being inspected, and the second dose level of the second radiation being determined according to the second scanning condition. A third distribution of the second radiation under the second scanning condition is determined based on the second radiation and the high-dose projection data. The third distribution refers to the distribution of the emitted intensity of the second radiation after it passes through the object being inspected under the second scanning condition. Determine the noise estimate related to the scanner; The low-dose projection data are determined based on the third distribution and the noise estimation.

3. The method of claim 2, wherein the value of the first conditional scanning parameter includes at least one of tube current value, tube voltage value, collimation width, exposure time, antiscattering grid filtering, detector response, tube response, tube focal spot size, flyfocus mode, and detector integration time.

4. The method of claim 3, wherein there is a difference between the first scanning condition and the second scanning condition.

5. The method of claim 2, wherein the first dose level is higher than the second dose level.

6. The method of claim 2, wherein determining the third distribution of the second radiation under the second scanning condition based on the second radiation and the high-dose projection data comprises: The distribution of attenuation coefficients is determined by reconstructing the attenuation map of the object based on the high-dose projection data; The third distribution of the second radiation beam is determined based on the second distribution of the second radiation and the distribution of the attenuation coefficient of the second radiation corresponding to the object.

7. The method of claim 2, wherein determining the low-dose projection data based on the third distribution and the noise estimation comprises: Based on the third distribution, a Poisson distribution related to the second radiation is determined; The low-dose projection data are determined based on the Poisson distribution and the noise estimation.

8. The method according to any one of claims 1-7, further comprising: Obtain a second neural network model, which is configured to perform image data augmentation operations on the image data; The virtual high-dose image data is post-processed based on the second neural network model to obtain and output post-processed high-dose image data, which exhibits higher quality than the virtual high-dose image data.

9. A CT imaging system, comprising: A scanner includes a tube for emitting one or more radiation beams toward an object, and a detector for detecting one or more radiation beams emitted from the tube. A processing engine, connected to the scanner, is used for: Acquire low-dose image data; Obtain the first neural network model; Based on the first neural network model, low-dose image data is processed to generate virtual high-dose image data corresponding to the low-dose image data; The first neural network model was obtained in the following way: High-dose projection data is acquired, which is generated by scanning the object being inspected using a scanner. Acquire low-dose projection data corresponding to the high-dose projection data, wherein the low-dose projection data is obtained from a scanner or determined by simulation based on the high-dose projection data; Based on the high-dose projection data, a high-dose image is generated using a first reconstruction technique; Based on the low-dose projection data, a low-dose image is generated using a second reconstruction technique; The first neural network model is determined by training a neural network model based on the neural network training algorithm, the high-dose image, and the corresponding low-dose image.

10. The system of claim 9, wherein the processing engine is further configured to: Obtain a second neural network model, which is configured to perform image data augmentation operations on the image data; The virtual high-dose image data is post-processed based on the second neural network model to obtain post-processed high-dose image data, which exhibits higher quality than the virtual high-dose image data.

Images