WO2024051018A1 - 一种pet参数图像的增强方法、装置、设备及存储介质 - Google Patents

一种pet参数图像的增强方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2024051018A1
WO2024051018A1 PCT/CN2022/138173 CN2022138173W WO2024051018A1 WO 2024051018 A1 WO2024051018 A1 WO 2024051018A1 CN 2022138173 W CN2022138173 W CN 2022138173W WO 2024051018 A1 WO2024051018 A1 WO 2024051018A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
pet
parameter
dynamic
pet parameter
Prior art date
Application number
PCT/CN2022/138173
Other languages
English (en)
French (fr)
Inventor
陈泓兆
孙涛
吴亚平
王振国
王梅云
Original Assignee
深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2024051018A1 publication Critical patent/WO2024051018A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/90—Dynamic range modification of images or parts thereof
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10104—Positron emission tomography [PET]

Definitions

  • the present invention relates to the technical field of medical image processing, and in particular to a method, device, equipment and storage medium for enhancing PET parameter images.
  • PET Positron Emission Computed Tomography
  • tracers to detect metabolic characteristics of human or animal organs. It has the characteristics of high sensitivity, good accuracy, and accurate positioning.
  • dynamic PET imaging technology can provide tracer distribution images at continuous time points, revealing the changes in tracer activity over time.
  • PET parameter images that can reflect the functional parameters of tissues and organs can be further obtained, such as K 1 parameter images, k 2 parameter images, k 3 parameter images, K i parameter images, etc.
  • the first method can reduce the noise in the PET parameter image, it will also reduce the spatial resolution of the PET parameter image and destroy the image details of the PET parameter image.
  • the second method mostly requires PET parameter images with high image quality as training labels to train the image enhancement model.
  • PET parameter images with high image quality require longer scanning time or higher tracer injection dose, which does not meet clinical needs.
  • the image collection requirements bring great difficulty to the preparation of training labels.
  • Embodiments of the present invention provide a PET parameter image enhancement method, device, equipment and storage medium to solve the problem that the existing neural network model method needs to prepare high-quality PET parameter images while retaining the image details of the PET parameter image. At the same time, the image quality of PET parameter images is improved.
  • a method for enhancing PET parameter images includes:
  • the model parameters of the image enhancement model are adjusted until the preset number of iterations is met, and the predicted PET parameter image is used as the target corresponding to the original PET parameter image.
  • the input image is a noise image, a dynamic PET image corresponding to a preset acquisition time range in the dynamic PET image set, or a dynamic SUV image corresponding to the dynamic PET image.
  • a device for enhancing PET parameter images which device includes:
  • An input image acquisition module configured to determine the original PET parameter image based on the acquired dynamic PET image set, and acquire the input image corresponding to the original PET parameter image based on the preset mapping list;
  • a predicted PET parameter image determination module configured to input the input image into the image enhancement model to obtain an output predicted PET parameter image
  • a target PET parameter image determination module configured to adjust the model parameters of the image enhancement model based on the original PET parameter image and the predicted PET parameter image, until the preset number of iterations is met, using the predicted PET parameter image as The target PET parameter image corresponding to the original PET parameter image;
  • the input image is a noise image, a dynamic PET image corresponding to a preset acquisition time range in the dynamic PET image set, or a dynamic SUV image corresponding to the dynamic PET image.
  • an electronic device includes:
  • the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor, so that the at least one processor can execute the method described in any embodiment of the present invention. Enhancement method for PET parametric images.
  • a computer-readable storage medium stores computer instructions.
  • the computer instructions are used to implement any embodiment of the present invention when executed by a processor. Enhancement method for PET parametric images.
  • the technical solution of the embodiment of the present invention is to obtain the input image corresponding to the original PET parameter image determined based on the dynamic PET image set based on the preset mapping list, where the input image is a noise image, a dynamic PET image and a preset acquisition time
  • the dynamic PET image corresponding to the range or the dynamic SUV image corresponding to the dynamic PET image is input into the image enhancement model to obtain the output predicted PET parameter image.
  • the image enhancement model is The model parameters are adjusted until the preset number of iterations is met, and the predicted PET parameter image is used as the target PET parameter image corresponding to the original PET parameter image, which solves the problem that the existing neural network model method needs to prepare high-quality PET parameter images. , while retaining the image details of the PET parameter image, while improving the image quality of the PET parameter image.
  • Figure 1 is a flow chart of a PET parameter image enhancement method provided by Embodiment 1 of the present invention.
  • Figure 2 is a flow chart of a PET parameter image enhancement method provided in Embodiment 2 of the present invention.
  • Figure 3 is a schematic diagram of the model architecture of an image enhancement model provided in Embodiment 2 of the present invention.
  • Figure 4 is a flow chart of a specific example of a PET parameter image enhancement method provided in Embodiment 2 of the present invention.
  • FIG. 5 is a schematic structural diagram of a PET parameter image enhancement device provided in Embodiment 3 of the present invention.
  • FIG. 6 is a schematic structural diagram of an electronic device provided in Embodiment 4 of the present invention.
  • FIG 1 is a flow chart of a PET parameter image enhancement method provided in Embodiment 1 of the present invention. This embodiment can be applied to the situation of image enhancement of PET parameter images.
  • This method can be performed by an enhancement device for PET parameter images.
  • the PET parameter image enhancement device can be implemented in the form of hardware and/or software, and the PET parameter image enhancement device can be configured in the terminal device. As shown in Figure 1, the method includes:
  • the dynamic PET image set contains at least two dynamic PET images.
  • 18 F-FDG PET/CT dynamic imaging scanning technology can be used to perform imaging scanning on the tested object to obtain a dynamic PET image set.
  • the specific imaging technology used to obtain the dynamic PET image set is not limited here.
  • the original PET parameter image may be a kinetic parameter image or a functional parameter image.
  • the kinetic parameter image can be K 1 parameter image, k 2 parameter image, k 3 parameter image and k 4 parameter image
  • the functional parameter image can be K i parameter image.
  • the K i parameter image can be used to reflect the glucose uptake rate of tissues and organs.
  • the original PET parameter image is a kinetic parameter image
  • kinetic modeling is performed on the dynamic PET image set through kinetic parameter modeling to obtain the original PET parameter image.
  • the image quality of the original PET parameter image obtained based on kinetic modeling is poor, which is not conducive to subsequent image analysis.
  • the preset mapping list can be used to characterize the mapping relationship between at least one original PET parameter image and at least one input image.
  • the preset mapping list includes at least one of K 1 parameter image, k 2 parameter image, k 3 parameter image, k 4 parameter image and K i parameter image, as well as corresponding to each original PET parameter image respectively. Enter an image.
  • the input images corresponding to each original PET parameter image may be the same or different.
  • the input image is a noise image, a dynamic PET image corresponding to a preset collection time range in a dynamic PET image set, or a dynamic SUV image corresponding to a dynamic PET image.
  • the noise image may be a salt-and-pepper noise image, a Gaussian noise image or a mixed noise image, and the type of noise contained in the noise image is not limited here.
  • the preset collection time range is used to characterize the preset time period within the total collection duration corresponding to the dynamically collected image set. Taking the total collection time as 60 minutes as an example, the preset collection time range can be 0-5 minutes, 10-15 minutes or 50-60 minutes, etc.
  • the minimum acquisition time corresponding to the preset acquisition time range is 0, or the maximum acquisition time corresponding to the preset acquisition time range is the dynamic PET image set The corresponding total collection time.
  • the minimum acquisition time corresponding to the preset acquisition time range is 0, the maximum acquisition time is less than the first time threshold, and the first time threshold is less than half of the total acquisition time corresponding to the dynamic PET image set. Taking the total collection time as 60 minutes as an example, the first time threshold is less than 30 minutes. In an optional embodiment, the preset collection time range is 0-5 minutes. In this embodiment, the dynamic PET images corresponding to the preset acquisition time range are early dynamic PET images in the dynamic PET image set.
  • the maximum acquisition time corresponding to the preset acquisition time range is the total acquisition time corresponding to the dynamic PET image set, the minimum acquisition time is greater than the second time threshold, and the second time threshold is greater than the total acquisition time corresponding to the dynamic PET image set. Half the duration. Taking the total collection time as 60 minutes as an example, the second time threshold is greater than 30 minutes. In an optional embodiment, the preset collection time range is 50-60 minutes. In this embodiment, the dynamic PET images corresponding to the preset acquisition time range are the final dynamic PET images in the dynamic PET image set.
  • the early dynamic PET image or the late dynamic PET image is used as the input image corresponding to the K 1 parameter image, which can effectively improve the image quality of the K 1 parameter image.
  • the SUV (standard uptake value) image can represent the ratio between the activity concentration of the tracer taken up by tissues and organs and the average activity concentration of the whole body, and is used to reflect the metabolic activity of glucose.
  • the dynamic PET image is multiplied by the weight of the subject divided by the injection dose of the tracer to obtain the dynamic SUV image.
  • the advantage of this setting is that it can weaken the individual differences between different tested subjects, eliminate the influence of variables caused by weight variables and injection dose variables by unifying variables, thereby improving the image quality of the subsequent target PET parameter images.
  • the image enhancement model can perform image enhancement processing on the input image and output a predicted PET parameter image.
  • exemplary model architectures of image enhancement models include but are not limited to generative adversarial network architecture, U-NET architecture and super-resolution convolutional architecture (Super Resolution Convolutional Neural Networks, SRCNN), etc.
  • the image enhancement model The model architecture is not limited.
  • adjusting the model parameters of the image enhancement model based on the original PET parameter image and the predicted PET parameter image includes: based on the L2 loss function, determining the Euclidean difference between the original PET parameter image and the predicted PET parameter image. Distance difference; the L-BFGS iterative algorithm is used to adjust the model parameters of the image enhancement model by minimizing the Euclidean distance difference.
  • model parameters satisfy the formula:
  • * represents the L2 norm operator
  • f represents the image enhancement model
  • x 0 represents the original PET parameter image
  • x * represents the next predicted PET parameter image output by the image enhancement model based on the adjusted model parameter ⁇ * .
  • the advantage of this setting is that it can reduce the number of iterations of the image enhancement model and reduce the memory space occupied by the image enhancement model.
  • the predicted PET parameter image is continued to be output based on the image enhancement model corresponding to the adjusted model parameters.
  • the preset number of iterations may be 1000 times or 500 times, and the preset number of iterations is not limited here.
  • the technical solution of this embodiment is to obtain the input image corresponding to the original PET parameter image determined based on the dynamic PET image set based on the preset mapping list, where the input image is a noise image, a dynamic PET image and a preset acquisition time range
  • the corresponding dynamic PET image or the dynamic SUV image corresponding to the dynamic PET image is input into the image enhancement model to obtain the output predicted PET parameter image.
  • the image enhancement model is The model parameters are adjusted until the preset number of iterations is met, and the predicted PET parameter image is used as the target PET parameter image corresponding to the original PET parameter image, which solves the problem that the existing neural network model method needs to prepare high-quality PET parameter images. While retaining the image details of the PET parameter image, the image quality of the PET parameter image is improved.
  • FIG 2 is a flow chart of a PET parameter image enhancement method provided in Embodiment 2 of the present invention. This embodiment further optimizes the image enhancement model in the above embodiment. As shown in Figure 2, the method includes:
  • the model architecture of the image enhancement model is a U-NET architecture, where the U-NET architecture includes an encoder and a decoder.
  • the encoder includes at least two encoding convolutional networks
  • the decoder includes at least two decoding convolutional networks, and each encoding convolutional network and each decoding convolutional network are arranged symmetrically.
  • the encoding convolutional network model and the decoding convolutional network each contain multiple convolutional layers in series.
  • a convolutional layer is provided between every two adjacent coding convolutional networks in the encoder.
  • the stride of at least one convolutional layer is 2.
  • the convolution parameters corresponding to each convolution layer are not limited here.
  • the advantage of this setting is that it can reduce the artifacts present in the predicted PET parameter images output by the image enhancement model.
  • the first parameter feature map is determined based on the input input image, and the first parameter feature map is output to the first parameter feature map respectively.
  • the decoding convolutional network (j n-i+1) corresponding to the current encoding convolutional network in the encoder
  • a bilinear interpolation layer is provided between every two adjacent decoding convolutional networks in the decoder.
  • the advantage of this setting is that it can reduce the artifacts present in the predicted PET parameter images output by the image enhancement model.
  • the first upsampling feature map is determined based on the last parameter feature map output by the last encoding convolutional network in the encoder. , and output the first upsampling feature map to the first bilinear interpolation layer; through the first bilinear interpolation layer in the decoder, determine the first interpolation feature based on the first upsampling feature map map, and output the first interpolated feature map to the second decoding convolutional network; through the current decoding convolutional network (1 ⁇ j ⁇ n) in the decoder, output based on the i-1th bilinear interpolation layer
  • FIG. 3 is a schematic diagram of the model architecture of an image enhancement model provided in Embodiment 2 of the present invention.
  • the image enhancement model includes an encoder and a decoder, where the encoder contains n encoding convolutional networks, and a convolutional layer is set between each two adjacent encoding convolutional networks.
  • the decoder contains n decoding convolutional networks, and a bilinear interpolation layer is set between each two adjacent decoding convolutional networks.
  • the method further includes: when the input image is a dynamic PET image or a dynamic SUV In the case of images, the input image is normalized to obtain the normalized input image; the normalized input image is registered with the original PET parameter image to obtain the registered input image.
  • the input image is used as a floating image
  • the original PET parameter image is used as a standard image
  • a registration operation is performed on the input image and the original PET parameter image.
  • Exemplary registration algorithms used include, but are not limited to, affine registration, rigid registration, and so on.
  • the advantage of this setting is that it can improve the computational efficiency of the image enhancement model and improve the image quality of the target PET parameter image.
  • the method before inputting the input image into the image enhancement model to obtain the output predicted PET parameter image, the method further includes: performing cropping on the original PET parameter image and the input image based on the preset cropping size. Operation to obtain the cropped original PET parameter image and input image.
  • the region of interest is the rectangular frame area surrounding the brain, and the image size is 96*96*80.
  • the cropping area and cropping size are not limited here.
  • the advantage of this setting is that it can reduce the computational load of the subsequent image enhancement model and improve the computational efficiency of the image enhancement model.
  • FIG. 4 is a flow chart of a specific example of a PET parameter image enhancement method provided in Embodiment 2 of the present invention.
  • the input image is input into the improved U-NET model to determine whether the current number of iterations meets the preset number of iterations. If so, it means that the iteration process is over, and the final output predicted PET parameter image is used as the corresponding original PET parameter image.
  • the target PET parameter image if not, it means that the iterative process is not over.
  • the original PET parameter image is used as the training label of the modified U-NET model, and the L2 loss function is used to predict the output based on the training label and the improved U-NET model.
  • PET parameter image adjust the model weight of the image enhancement model, obtain an updated and improved U-NET model corresponding to the current iterative process, and continue the iterative process.
  • Table 1 shows the contrast-to-noise ratio (CNR) and contrast-to-noise ratio improvement rate (CNRIR) corresponding to a different image enhancement method provided in Embodiment 2 of the present invention.
  • IM5-G means that the dynamic PET image of 0-5 minutes in the dynamic PET image set is used as the input image
  • SUV-G means that the dynamic SUV image corresponding to the dynamic PET image of 50-60 minutes in the dynamic PET image set is used as the input image
  • BM4D stands for three-dimensional block matching filtering method
  • DIP stands for depth image prior method
  • GF stands for Gaussian filter
  • NLM non-local mean method.
  • dynamic PET images are brain PET images including blood vessel walls, gray matter, white matter and other areas.
  • the contrast-to-noise ratio of the IM5-G method and SUV-G provided in this embodiment are improved by 18.23% and 3.78% respectively.
  • the IM5-G method has greatly improved both the contrast-to-noise ratio and the contrast-to-noise ratio improvement rate.
  • the technical solution of this embodiment is to input the input image into the encoder in the image enhancement model, and output at least two parameter feature maps based on the input input image through at least two encoding convolutional networks in the encoder.
  • At least two decoding convolutional networks in the decoder based on at least two parameter feature maps output by the encoder, output a predicted PET parameter image, solving the problem of poor image quality of the target PET parameter image, making use of the image enhancement model and one-time
  • the dynamic PET image set obtained during the scanning process can obtain target PET parameter images with high contrast-to-noise ratio and rich image details, which improves the convergence speed of the image enhancement model.
  • FIG. 5 is a schematic structural diagram of a PET parameter image enhancement device provided in Embodiment 3 of the present invention. As shown in Figure 5, the device includes: an input image acquisition module 310, a predicted PET parameter image determination module 320 and a target PET parameter image determination module 330.
  • the input image acquisition module 310 is used to determine the original PET parameter image based on the acquired dynamic PET image set, and acquire the input image corresponding to the original PET parameter image based on the preset mapping list;
  • the predicted PET parameter image determination module 320 is used to input the input image into the image enhancement model to obtain the output predicted PET parameter image;
  • the target PET parameter image determination module 330 is used to adjust the model parameters of the image enhancement model based on the original PET parameter image and the predicted PET parameter image, until the preset number of iterations is met, and the predicted PET parameter image is used as the corresponding original PET parameter image.
  • Target PET parameter image is used to adjust the model parameters of the image enhancement model based on the original PET parameter image and the predicted PET parameter image, until the preset number of iterations is met, and the predicted PET parameter image is used as the corresponding original PET parameter image.
  • the input image is a noise image, a dynamic PET image corresponding to a preset collection time range in a dynamic PET image set, or a dynamic SUV image corresponding to a dynamic PET image.
  • the technical solution of this embodiment is to obtain the input image corresponding to the original PET parameter image determined based on the dynamic PET image set based on the preset mapping list, where the input image is a noise image, a dynamic PET image and a preset acquisition time range
  • the corresponding dynamic PET image or the dynamic SUV image corresponding to the dynamic PET image is input into the image enhancement model to obtain the output predicted PET parameter image.
  • the image enhancement model is The model parameters are adjusted until the preset number of iterations is met, and the predicted PET parameter image is used as the target PET parameter image corresponding to the original PET parameter image, which solves the problem that the existing neural network model method needs to prepare high-quality PET parameter images. While retaining the image details of the PET parameter image, the image quality of the PET parameter image is improved.
  • the minimum acquisition time corresponding to the preset acquisition time range is 0, or the maximum acquisition time corresponding to the preset acquisition time range is dynamic The total acquisition time corresponding to the PET image set.
  • the device further includes:
  • the input image registration module is used to adjust the model parameters of the image enhancement model based on the original PET parameter image and the predicted PET parameter image, in the case where the input image is a dynamic PET image or a dynamic SUV image. Normalization processing to obtain the normalized input image;
  • the model architecture of the image enhancement model is a U-NET architecture, where the U-NET architecture includes an encoder and a decoder.
  • the predicted PET parameter image determination module 320 is specifically used.
  • At least two parameter feature maps are output based on the input input image
  • a predicted PET parameter image is output based on at least two parameter feature maps output by the encoder.
  • a convolution layer is provided between every two adjacent coding convolutional networks in the encoder.
  • a bilinear interpolation layer is provided between every two adjacent decoding convolutional networks in the decoder.
  • the target PET parameter image determination module 330 is specifically used for:
  • the L-BFGS iterative algorithm is used to adjust the model parameters of the image enhancement model by minimizing the Euclidean distance difference.
  • the PET parameter image enhancement device provided by the embodiment of the present invention can execute the PET parameter image enhancement method provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.
  • FIG. 6 is a schematic structural diagram of an electronic device provided in Embodiment 4 of the present invention.
  • Electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.
  • Electronic devices may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (eg, helmets, glasses, watches, etc.), and other similar computing devices.
  • the components shown in the embodiments of the invention, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and/or claimed herein.
  • the electronic device 10 includes at least one processor 11, and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores There is a computer program executable by at least one processor 11, which may be based on a computer program stored in a read-only memory (ROM) 12 or loaded from a storage unit 18 into a random access memory (RAM) 13, to perform various appropriate actions and processing. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored.
  • the processor 11, the ROM 12 and the RAM 13 are connected to each other via the bus 14.
  • An input/output (I/O) interface 15 is also connected to bus 14 .
  • the I/O interface 15 Multiple components in the electronic device 10 are connected to the I/O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc. etc.; and communication unit 19, such as network card, modem, wireless communication transceiver, etc.
  • the communication unit 19 allows the electronic device 10 to exchange information/data with other devices through computer networks such as the Internet and/or various telecommunications networks.
  • Processor 11 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processing processor (DSP), and any appropriate processor, controller, microcontroller, etc.
  • the processor 11 performs various methods and processes described above, such as the enhancement method of PET parametric images.
  • the PET parametric image enhancement method may be implemented as a computer program, which is tangibly included in a computer-readable storage medium, such as the storage unit 18 .
  • part or all of the computer program may be loaded and/or installed onto the electronic device 10 via the ROM 12 and/or the communication unit 19.
  • the processor 11 may be configured to perform the enhancement method of the PET parametric image in any other suitable manner (eg, by means of firmware).
  • Various implementations of the systems and techniques described above may be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip implemented in a system (SOC), load programmable logic device (CPLD), computer hardware, firmware, software, and/or a combination thereof.
  • FPGAs field programmable gate arrays
  • ASICs application specific integrated circuits
  • ASSPs application specific standard products
  • SOC system
  • CPLD load programmable logic device
  • computer hardware firmware, software, and/or a combination thereof.
  • These various embodiments may include implementation in one or more computer programs executable and/or interpreted on a programmable system including at least one programmable processor, the programmable processor
  • the processor which may be a special purpose or general purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
  • An output device may be a special purpose or general purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
  • An output device may be a special purpose or general purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
  • the computer program for implementing the PET parametric image enhancement method of the present invention may be written using any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that the computer program, when executed by the processor, causes the functions/operations specified in the flowcharts and/or block diagrams to be implemented. A computer program may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
  • Embodiment 5 of the present invention also provides a computer-readable storage medium.
  • the computer-readable storage medium stores computer instructions.
  • the computer instructions are used to cause the processor to execute a PET parameter image enhancement method.
  • the method includes:
  • the model parameters of the image enhancement model are adjusted until the preset number of iterations is met, and the predicted PET parameter image is used as the target PET parameter image corresponding to the original PET parameter image;
  • the input image is a noise image, a dynamic PET image corresponding to a preset collection time range in a dynamic PET image set, or a dynamic SUV image corresponding to a dynamic PET image.
  • a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device.
  • Computer-readable storage media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any suitable combination of the foregoing.
  • the computer-readable storage medium may be a machine-readable signal medium.
  • machine-readable storage media would include electrical connections based on one or more wires, laptop disks, hard drives, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
  • RAM random access memory
  • ROM read only memory
  • EPROM or flash memory erasable programmable read only memory
  • CD-ROM portable compact disk read-only memory
  • magnetic storage device or any suitable combination of the above.
  • the systems and techniques described herein may be implemented on an electronic device having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display)) for displaying information to the user monitor); and a keyboard and pointing device (e.g., a mouse or a trackball) through which a user can provide input to the electronic device.
  • a display device eg, a CRT (cathode ray tube) or LCD (liquid crystal display)
  • a keyboard and pointing device e.g., a mouse or a trackball
  • Other kinds of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and may be provided in any form, including Acoustic input, voice input or tactile input) to receive input from the user.
  • the systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., A user's computer having a graphical user interface or web browser through which the user can interact with implementations of the systems and technologies described herein), or including such backend components, middleware components, or any combination of front-end components in a computing system.
  • the components of the system may be interconnected by any form or medium of digital data communication (eg, a communications network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.
  • Computing systems may include clients and servers.
  • Clients and servers are generally remote from each other and typically interact over a communications network.
  • the relationship of client and server is created by computer programs running on corresponding computers and having a client-server relationship with each other.
  • the server can be a cloud server, also known as cloud computing server or cloud host. It is a host product in the cloud computing service system to solve the problems of difficult management and weak business scalability in traditional physical hosts and VPS services. defect.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Processing (AREA)

Abstract

一种PET参数图像的增强方法、装置、设备及存储介质，该方法包括：基于预设映射列表，获取与基于动态PET图像集确定的原始PET参数图像对应的输入图像（S110）；将输入图像输入到图像增强模型中，得到输出的预测PET参数图像（S120），基于原始PET参数图像和输出的预测PET参数图像，对图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为原始PET参数图像对应的目标PET参数图像（S130）；其中，输入图像为噪声图像、动态PET图像集中与预设采集时间范围对应的动态PET图像或者与动态PET图像对应的动态SUV图像。该方法解决了现有的神经网络模型方法需要制备高质量的PET参数图像的问题。

Description

一种PET参数图像的增强方法、装置、设备及存储介质 技术领域
本发明涉及医学图像处理技术领域，尤其涉及一种PET参数图像的增强方法、装置、设备及存储介质。
背景技术
PET(Positron Emission Computed Tomography，正电子发射型计算机断层显像)成像是一种采用示踪剂检测人体或动物体器官代谢特征的医学成像技术，具备灵敏度高、准确性好、定位准确的特点。其中，动态PET成像技术可以提供连续时间点上的示踪剂的分布图像，揭示了示踪剂的活度随时间的变化规律。通过对动态PET图像序列应用动力学模型，可进一步得到能够反映组织器官的功能参数的PET参数图像，如K 1参数图像、k 2参数图像、k 3参数图像和K i参数图像等等。
目前为提高PET参数图像的图像质量，主要采用滤波算法或神经网络模型两种方式。其中，第一种方式虽然能够降低PET参数图像中的噪声，但同时也会降低PET参数图像的空间分辨率，破坏PET参数图像的图像细节。第二种方式大多需要图像质量高的PET参数图像作为训练标签对图像增强模型进行训练，而图像质量高的PET参数图像需要较长的扫描时间或较高的示踪剂注射剂量，不满足临床的图像采集要求，为训练标签的制备带来很大难度。
发明内容
本发明实施例提供了一种PET参数图像的增强方法、装置、设备及存储介质，以解决现有的神经网络模型方法需要制备高质量的PET参数图像的问题，在保留PET参数图像的图像细节的同时，提高PET参数图像的图像 质量。
根据本发明一个实施例提供了一种PET参数图像的增强方法，该方法包括：
基于获取到的动态PET图像集，确定原始PET参数图像，并基于预设映射列表，获取与所述原始PET参数图像对应的输入图像；
将所述输入图像输入到图像增强模型中，得到输出的预测PET参数图像；
基于所述原始PET参数图像和所述预测PET参数图像，对所述图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为所述原始PET参数图像对应的目标PET参数图像；
其中，所述输入图像为噪声图像、所述动态PET图像集中与预设采集时间范围对应的动态PET图像或者与所述动态PET图像对应的动态SUV图像。
根据本发明另一个实施例提供了一种PET参数图像的增强装置，该装置包括：
输入图像获取模块，用于基于获取到的动态PET图像集，确定原始PET参数图像，并基于预设映射列表，获取与所述原始PET参数图像对应的输入图像；
预测PET参数图像确定模块，用于将所述输入图像输入到图像增强模型中，得到输出的预测PET参数图像；
目标PET参数图像确定模块，用于基于所述原始PET参数图像和所述预测PET参数图像，对所述图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为所述原始PET参数图像对应的目标PET参数图像；
其中，所述输入图像为噪声图像、所述动态PET图像集中与预设采集时间范围对应的动态PET图像或者与所述动态PET图像对应的动态SUV图像。
根据本发明另一个实施例提供了一种电子设备，所述电子设备包括：
至少一个处理器；以及
与所述至少一个处理器通信连接的存储器；其中，
所述存储器存储有可被所述至少一个处理器执行的计算机程序，所述计算机程序被所述至少一个处理器执行，以使所述至少一个处理器能够执行本发明任一实施例所述的PET参数图像的增强方法。
根据本发明另一个实施例，提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机指令，所述计算机指令用于使处理器执行时实现本发明任一实施例所述的PET参数图像的增强方法。
本发明实施例的技术方案，通过基于预设映射列表，获取与基于动态PET图像集确定的原始PET参数图像对应的输入图像，其中，输入图像为噪声图像、动态PET图像中与预设采集时间范围对应的动态PET图像或者与动态PET图像对应的动态SUV图像，将输入图像输入到图像增强模型中，得到输出的预测PET参数图像，基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为原始PET参数图像对应的目标PET参数图像，解决了现有的神经网络模型方法需要制备高质量的PET参数图像的问题，在保留了PET参数图像的图像细节的同时，提高了PET参数图像的图像质量。
应当理解，本部分所描述的内容并非旨在标识本发明的实施例的关键或重要特征，也不用于限制本发明的范围。本发明的其它特征将通过以下的说明书而变得容易理解。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例一所提供的一种PET参数图像的增强方法的流程图；
图2为本发明实施例二所提供的一种PET参数图像的增强方法的流程图；
图3为本发明实施例二所提供的一种图像增强模型的模型架构的示意图；
图4为本发明实施例二所提供的一种PET参数图像的增强方法的具体实例的流程图；
图5为本发明实施例三所提供的一种PET参数图像的增强装置的结构示意图；
图6为本发明实施例四所提供的一种电子设备的结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本发明方案，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分的实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本发明保护的范围。
需要说明的是，本发明的说明书和权利要求书及上述附图中的术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
实施例一
图1为本发明实施例一所提供的一种PET参数图像的增强方法的流程图，本实施例可适用于对PET参数图像进行图像增强的情况，该方法可以由PET参数图像的增强装置来执行，该PET参数图像的增强装置可以采用硬件和/或软件的形式实现，该PET参数图像的增强装置可配置于终端设备中。如图1所示，该方法包括：
S110、基于获取到的动态PET图像集，确定原始PET参数图像，并基 于预设映射列表，获取与原始PET参数图像对应的输入图像。
其中，具体的，动态PET图像集中包含至少两个动态PET图像。示例性的，可采用 18F-FDG PET/CT动态成像扫描技术，对被测对象进行成像扫描，得到动态PET图像集。此处对获取动态PET图像集采用的具体成像技术不作限定。
其中，示例性的，原始PET参数图像可以是动力学参数图像或功能参数图像。如动力学参数图像可以是K 1参数图像、k 2参数图像、k 3参数图像和k 4参数图像，功能参数图像可以是K i参数图像。其中，K i参数图像可用于反映组织器官的葡萄糖摄取率。
在一个可选实施例中，具体的，当原始PET参数图像为动力学参数图像时，通过动态参数模型(kinetic modelling)，对动态PET图像集进行动力学建模，得到原始PET参数图像。此时，基于动力学建模得到的原始PET参数图像的图像质量较差，不利于后续的图像分析。
其中，具体的，预设映射列表可用于表征至少一个原始PET参数图像与至少一个输入图像之间的映射关系。其中，示例性的，预设映射列表中包含K 1参数图像、k 2参数图像、k 3参数图像、k 4参数图像和K i参数图像中至少一个，以及与各原始PET参数图像分别对应的输入图像。其中，各原始PET参数图像分别对应的输入图像可以相同也可以不同。
在本实施例中，输入图像为噪声图像、动态PET图像集中与预设采集时间范围对应的动态PET图像或者与动态PET图像对应的动态SUV图像。
其中，示例性的，噪声图像可以是椒盐噪声图像、高斯噪声图像或混合噪声图像，此处对噪声图像中包含的噪声类别不作限定。
其中，具体的，采集一次动态PET图像集需要一定的采集时长，通常为60分钟。在本实施例中，预设采集时间范围用于表征动态采集图像集对应的总采集时长内的预设时间段。以总采集时长为60分钟为例，预设采集时间范围可以为0-5分钟、10-15分钟或50-60分钟等等。
在一个可选实施例中，当原始PET参数图像为K 1参数图像时，预设采集时间范围对应的最小采集时间为0，或者，预设采集时间范围对应的最大采集时间为动态PET图像集对应的总采集时长。
在一个实施例中，预设采集时间范围对应的最小采集时间为0，最大采集时间小于第一时间阈值，第一时间阈值小于动态PET图像集对应的总采集时长的一半。以总采集时长为60分钟为例，第一时间阈值小于30分钟。在一个可选实施例中，预设采集时间范围为0-5分钟。在本实施例中，与该预设采集时间范围对应的动态PET图像为动态PET图像集中的早期动态PET图像。
在另一个实施例中，预设采集时间范围对应的最大采集时间为动态PET图像集对应的总采集时长，最小采集时间大于第二时间阈值，第二时间阈值大于动态PET图像集对应的总采集时长的一半。以总采集时长为60分钟为例，第二时间阈值大于30分钟。在一个可选实施例中，预设采集时间范围为50-60分钟。在本实施例中，与该预设采集时间范围对应的动态PET图像为动态PET图像集中的末期动态PET图像。
有研究证明，动态PET图像集中的早期动态PET图像或末期动态PET图像，与K 1参数图像之间具备一定的相关性。因此，本实施例将早期动态PET图像或末期动态PET图像作为K 1参数图像对应的输入图像，可以有效提高K 1参数图像的图像质量。
其中，具体的，SUV(standard uptake value，标准摄取值)图像可表征组织器官摄取的示踪剂的活度浓度与全身平均活度浓度之间的比值，用于反映葡萄糖的代谢活跃度。具体的，将动态PET图像乘以被测对象的体重除以示踪剂的注射剂量，得到动态SUV图像。
这样设置的好处在于，可以弱化不同被测对象之间的个体差异，通过统一变量来消除体重变量和注射剂量变量带来的变量影响，进而提高后续得到的目标PET参数图像的图像质量。
S120、将输入图像输入到图像增强模型中，得到输出的预测PET参数图像。
其中，具体的，图像增强模型可以对输入的输入图像进行图像增强处理，输出预测PET参数图像。其中，示例性的，图像增强模型的模型架构包括但不限于生成对抗网络架构、U-NET架构和超分辨率卷积架构(Super Resolution Convolutional Neural Networks，SRCNN)等等，此处对图像增强模型的模型架构不作限定。
S130、基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为原始PET参数图像对应的目标PET参数图像。
在一个可选实施例中，基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整，包括：基于L2损失函数，确定原始PET参数图像与预测PET参数图像之间的欧式距离差；采用L-BFGS迭代算法，通过最小化欧式距离差对图像增强模型的模型参数进行调整。
其中，示例性的，模型参数满足公式：
Figure PCTCN2022138173-appb-000001
其中，*表示L2范数算子，f表示图像增强模型，
Figure PCTCN2022138173-appb-000002
表示输入图像，x 0表示原始PET参数图像，x *表示图像增强模型基于调整后的模型参数θ *输出的下一预测PET参数图像。
这样设置的好处在于，可以降低图像增强模型的迭代次数，以及降低图像增强模型对内存空间的占用。
其中，具体的，在当前迭代次数不满足预设迭代次数的情况下，基于调整后的模型参数对应的图像增强模型，继续输出预测PET参数图像。其中，示例性的，预设迭代次数可以为1000次或500次，此处对预设迭代次数不作限定。
本实施例的技术方案，通过基于预设映射列表，获取与基于动态PET图像集确定的原始PET参数图像对应的输入图像，其中，输入图像为噪声图像、动态PET图像中与预设采集时间范围对应的动态PET图像或者与动态PET图像对应的动态SUV图像，将输入图像输入到图像增强模型中，得到输出的预测PET参数图像，基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为原始PET参数图像对应的目标PET参数图像，解决了现有的神经网络模型方法需要制备高质量的PET参数图像的问题，在保留了PET参数图像的图像细节的同时，提高了PET参数图像的图像质量。
实施例二
图2为本发明实施例二所提供的一种PET参数图像的增强方法的流程图，本实施例对上述实施例中的图像增强模型进行进一步优化。如图2所示，该方法包括：
S210、基于获取到的动态PET图像集，确定原始PET参数图像，并基于预设映射列表，获取与原始PET参数图像对应的输入图像。
S220、将输入图像输入到图像增强模型中的编码器中。
在本实施例中，图像增强模型的模型架构为U-NET架构，其中，U-NET架构包括编码器和解码器。
其中，具体的，编码器包含至少两个编码卷积网络，解码器中包含至少两个解码卷积网络，各编码卷积网络和各解码卷积网络对称设置。编码卷积网络模型和解码卷积网络中分别包含多个串联的卷积层。
S230、通过编码器中的至少两个编码卷积网络，基于输入的输入图像，输出至少两个参数特征图。
在一个可选实施例中，编码器中每两个相邻的编码卷积网络之间设置有一个卷积层。其中，示例性的，至少一个卷积层的步幅均为2。此处对各卷积层分别对应的卷积参数不作限定。
这样设置的好处在于，可以减少图像增强模型输出的预测PET参数图像中存在的伪影。
其中，具体的，通过编码器中的第一个编码卷积网络(i＝1)，基于输入的输入图像，确定第一个参数特征图，并将第一个参数特征图分别输出给第一个卷积层以及解码器中的最后一个解码卷积网络(j＝n)；通过编码器中的第一个卷积层，基于输入的第一个参数特征图，确定第一个卷积特征向量，并将第一个卷积特征向量输出给第二个编码卷积网络；通过编码器中的当前编码卷积网络(1<i<n，n表示编码器中编码卷积网络的总个数)，基于第i-1个卷积层输出的第i-1个卷积特征向量，确定第i个参数特征图，并将第i个参数特征图输出给第i个卷积层以及解码器中与当前编码卷积网络对应的解码卷积网络(j＝n-i+1)；以此类推，通过编码器中的最后一个编码卷积网络(i＝n)，基于第n-1个卷积层输出的第n-1个卷积特征向量，确定最后一个参数特征图，并将最后一个参数特征图输出给解码器中的第一个解码卷积网络(j＝1)。
S240、通过解码器中的至少两个解码卷积网络，基于编码器输出的至少两个参数特征图，输出预测PET参数图像。
在一个可选实施例中，解码器中每两个相邻的解码卷积网络之间设置有一个双线性插值层。
这样设置的好处在于，可以减少图像增强模型输出的预测PET参数图像中存在的伪影。
其中，具体的，通过解码器中的第一个解码卷积网络(j＝1)，基于编码器中的最后一个编码卷积网络输出的最后一个参数特征图，确定第一个上采样特征图，并将第一个上采样特征图输出给第一个双线性插值层；通过解码器中的第一个双线性插值层，基于第一个上采样特征图，确定第一个插值特征图，并将第一个插值特征图输出给第二个解码卷积网络；通过解码器中的当前解码卷积网络(1<j<n)，基于第i-1个双线性插值层输出的第i-1个插 值特征图以及编码器中与当前解码卷积网络对应的编码卷积网络(i＝n-j+1)输入的参数特征图，确定第i个上采样特征图，并将第i个上采样特征图输出给第i个双线性插值层；以此类推，通过解码器中的最后一个解码卷积网络(j＝n)，基于第n-1个双线性插值层输出的第n-1个插值特征图以及编码器中第一个编码卷积网络输入的第一个参数特征图，确定预测PET参数图像，并将预测PET参数图像进行输出。
图3为本发明实施例二所提供的一种图像增强模型的模型架构的示意图。具体的，图像增强模型包括编码器和解码器，其中，编码器中包含n个编码卷积网络，每两个相邻的编码卷积网络之间设置有一个卷积层。解码器中包含n个解码卷积网络，每两个相邻的解码卷积网络之间设置有一个双线性插值层。
S250、基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为原始PET参数图像对应的目标PET参数图像。
在上述实施例的基础上，可选的，在基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整之前，该方法还包括：在输入图像为动态PET图像或动态SUV图像的情况下，对输入图像进行归一化处理，得到归一化后的输入图像；将归一化后的输入图像配准原始PET参数图像，得到配准后的输入图像。
其中，具体的，将输入图像作为浮动图像，将原始PET参数图像作为标准图像，对输入图像和原始PET参数图像执行配准操作。其中，示例性的，采用的配准算法包括但不限于仿射配准和刚性配准等等。
这样设置的好处在于，可以提高图像增强模型的运算效率，以及提高目标PET参数图像的图像质量。
在一个可选实施例中，在将输入图像输入到图像增强模型中，得到输出的预测PET参数图像之前，该方法还包括：基于预设裁剪尺寸，对原始PET 参数图像和输入图像分别执行裁剪操作，得到裁剪后的原始PET参数图像和输入图像。
其中，示例性的，可仅保留感兴趣区域对应的区域图像，如感兴趣区域为大脑的外包矩形框区域，图像尺寸为96*96*80。此处对裁剪区域和裁剪尺寸不作限定。
这样设置的好处在于，可以降低后续图像增强模型的运算量，提高图像增强模型的运算效率。
图4为本发明实施例二所提供的一种PET参数图像的增强方法的具体实例的流程图。具体的，将输入图像输入到改进的U-NET模型中，判断当前迭代次数是否满足预设迭代次数，如果是，则说明迭代过程结束，将最终输出的预测PET参数图像作为原始PET参数图像对应的目标PET参数图像；如果否，则说明迭代过程未结束，将原始PET参数图像作为改建的U-NET模型的训练标签，采用L2损失函数，基于训练标签和改进的U-NET模型输出的预测PET参数图像，对图像增强模型的模型权重进行调整，得到与当前迭代过程对应的更新后的改进的U-NET模型，并继续迭代过程。
表1为本发明实施例二所提供的一种不同的图像增强方法分别对应的对比度噪声比(CNR)和对比度噪声比提升率(CNRIR)。
图像增强方法 CNR(Mean±SD) CNRIR(Mean±SD)
IM5-G 22.53±18.67 18.23％±9.12％
SUV-G 19.86±14.48 3.78％±9.88％
BM4D 19.89±17.03 3.83％±98.73％
DIP 19.46±18.53 2.01％±5.7％
GF 19.07±17.73 0.64％±15.68％
NLM 20.39±15.66 6.91％±14.34％
其中，IM5-G表示采用动态PET图像集中0-5分钟的动态PET图像作为输入图像，SUV-G表示采用动态PET图像集中50-60分钟的动态PET图像 对应的动态SUV图像作为输入图像，BM4D表示三维块匹配过滤法，DIP表示深度图像先验法，GF表示高斯滤波器，NLM表示非局部均值法。其中，动态PET图像为包含血管壁、灰质和白质等区域的大脑PET图像。
通过表1可以得到，本实施例提供的IM5-G方法和SUV-G相比于原始PET参数图像，其对比度噪声比分别提升了18.23％和3.78％。其中，IM5-G方法相比于现有的图像增强方法，无论是对比度噪声比，还是对比度噪声比提升率，均有大幅提升。
本实施例的技术方案，通过将输入图像输入到图像增强模型中的编码器中，通过编码器中的至少两个编码卷积网络，基于输入的输入图像，输出至少两个参数特征图，通过解码器中的至少两个解码卷积网络，基于编码器输出的至少两个参数特征图，输出预测PET参数图像，解决了目标PET参数图像的图像质量差的问题，使得利用图像增强模型和一次扫描过程得到的动态PET图像集即可得到对比度噪声比高、图像细节丰富的目标PET参数图像，提高了图像增强模型的收敛速度。
实施例三
图5为本发明实施例三所提供的一种PET参数图像的增强装置的结构示意图。如图5所示，该装置包括：输入图像获取模块310、预测PET参数图像确定模块320和目标PET参数图像确定模块330。
其中，输入图像获取模块310，用于基于获取到的动态PET图像集，确定原始PET参数图像，并基于预设映射列表，获取与原始PET参数图像对应的输入图像；
预测PET参数图像确定模块320，用于将输入图像输入到图像增强模型中，得到输出的预测PET参数图像；
目标PET参数图像确定模块330，用于基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为原始PET参数图像对应的目标PET参数图 像；
其中，输入图像为噪声图像、动态PET图像集中与预设采集时间范围对应的动态PET图像或者与动态PET图像对应的动态SUV图像。
本实施例的技术方案，通过基于预设映射列表，获取与基于动态PET图像集确定的原始PET参数图像对应的输入图像，其中，输入图像为噪声图像、动态PET图像中与预设采集时间范围对应的动态PET图像或者与动态PET图像对应的动态SUV图像，将输入图像输入到图像增强模型中，得到输出的预测PET参数图像，基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为原始PET参数图像对应的目标PET参数图像，解决了现有的神经网络模型方法需要制备高质量的PET参数图像的问题，在保留了PET参数图像的图像细节的同时，提高了PET参数图像的图像质量。
在上述实施例的基础上，可选的，当原始PET参数图像为K1参数图像时，预设采集时间范围对应的最小采集时间为0，或者，预设采集时间范围对应的最大采集时间为动态PET图像集对应的总采集时长。
在上述实施例的基础上，可选的，该装置还包括：
输入图像配准模块，用于在基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整之前，在输入图像为动态PET图像或动态SUV图像的情况下，对输入图像进行归一化处理，得到归一化后的输入图像；
将归一化后的输入图像配准原始PET参数图像，得到配准后的输入图像。
在上述实施例的基础上，可选的，图像增强模型的模型架构为U-NET架构，其中，U-NET架构包括编码器和解码器，相应的，预测PET参数图像确定模块320，具体用于：
将输入图像输入到图像增强模型中的编码器中；
通过编码器中的至少两个编码卷积网络，基于输入的输入图像，输出至 少两个参数特征图；
通过解码器中的至少两个解码卷积网络，基于编码器输出的至少两个参数特征图，输出预测PET参数图像。
在上述实施例的基础上，可选的，编码器中每两个相邻的编码卷积网络之间设置有一个卷积层。
在上述实施例的基础上，可选的，解码器中每两个相邻的解码卷积网络之间设置有一个双线性插值层。
在上述实施例的基础上，可选的，目标PET参数图像确定模块330，具体用于：
基于L2损失函数，确定原始PET参数图像与预测PET参数图像之间的欧式距离差；
采用L-BFGS迭代算法，通过最小化欧式距离差对图像增强模型的模型参数进行调整。
本发明实施例所提供的PET参数图像的增强装置可执行本发明任意实施例所提供的PET参数图像的增强方法，具备执行方法相应的功能模块和有益效果。
实施例四
图6为本发明实施例四所提供的一种电子设备的结构示意图。电子设备10旨在表示各种形式的数字计算机，诸如，膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置，诸如，个人数字处理、蜂窝电话、智能电话、可穿戴设备(如头盔、眼镜、手表等)和其它类似的计算装置。本发明实施例所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例，并且不意在限制本文中描述的和/或者要求的本发明的实现。
如图6所示，电子设备10包括至少一个处理器11，以及与至少一个处理器11通信连接的存储器，如只读存储器(ROM)12、随机访问存储器(RAM) 13等，其中，存储器存储有可被至少一个处理器11执行的计算机程序，处理器11可以根据存储在只读存储器(ROM)12中的计算机程序或者从存储单元18加载到随机访问存储器(RAM)13中的计算机程序，来执行各种适当的动作和处理。在RAM 13中，还可存储电子设备10操作所需的各种程序和数据。处理器11、ROM 12以及RAM 13通过总线14彼此相连。输入/输出(I/O)接口15也连接至总线14。
电子设备10中的多个部件连接至I/O接口15，包括：输入单元16，例如键盘、鼠标等；输出单元17，例如各种类型的显示器、扬声器等；存储单元18，例如磁盘、光盘等；以及通信单元19，例如网卡、调制解调器、无线通信收发机等。通信单元19允许电子设备10通过诸如因特网的计算机网络和/或各种电信网络与其他设备交换信息/数据。
处理器11可以是各种具有处理和计算能力的通用和/或专用处理组件。处理器11的一些示例包括但不限于中央处理单元(CPU)、图形处理单元(GPU)、各种专用的人工智能(AI)计算芯片、各种运行机器学习模型算法的处理器、数字信号处理器(DSP)、以及任何适当的处理器、控制器、微控制器等。处理器11执行上文所描述的各个方法和处理，例如PET参数图像的增强方法。
在一些实施例中，PET参数图像的增强方法可被实现为计算机程序，其被有形地包含于计算机可读存储介质，例如存储单元18。在一些实施例中，计算机程序的部分或者全部可以经由ROM 12和/或通信单元19而被载入和/或安装到电子设备10上。当计算机程序加载到RAM 13并由处理器11执行时，可以执行上文描述的PET参数图像的增强方法的一个或多个步骤。备选地，在其他实施例中，处理器11可以通过其他任何适当的方式(例如，借助于固件)而被配置为执行PET参数图像的增强方法。
本文中以上描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、场可编程门阵列(FPGA)、专用集成电路(ASIC)、 专用标准产品(ASSP)、芯片上系统的系统(SOC)、负载可编程逻辑设备(CPLD)、计算机硬件、固件、软件、和/或它们的组合中实现。这些各种实施方式可以包括：实施在一个或者多个计算机程序中，该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和/或解释，该可编程处理器可以是专用或者通用可编程处理器，可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令，并且将数据和指令传输至该存储系统、该至少一个输入装置、和该至少一个输出装置。
用于实施本发明的PET参数图像的增强方法的计算机程序可以采用一个或多个编程语言的任何组合来编写。这些计算机程序可以提供给通用计算机、专用计算机或其他可编程数据处理装置的处理器，使得计算机程序当由处理器执行时使流程图和/或框图中所规定的功能/操作被实施。计算机程序可以完全在机器上执行、部分地在机器上执行，作为独立软件包部分地在机器上执行且部分地在远程机器上执行或完全在远程机器或服务器上执行。
实施例五
本发明实施例五还提供了一种计算机可读存储介质，计算机可读存储介质存储有计算机指令，计算机指令用于使处理器执行一种PET参数图像的增强方法，该方法包括：
基于获取到的动态PET图像集，确定原始PET参数图像，并基于预设映射列表，获取与原始PET参数图像对应的输入图像；
将输入图像输入到图像增强模型中，得到输出的预测PET参数图像；
基于原始PET参数图像和预测PET参数图像，对图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为原始PET参数图像对应的目标PET参数图像；
其中，输入图像为噪声图像、动态PET图像集中与预设采集时间范围对应的动态PET图像或者与动态PET图像对应的动态SUV图像。
在本发明的上下文中，计算机可读存储介质可以是有形的介质，其可以 包含或存储以供指令执行系统、装置或设备使用或与指令执行系统、装置或设备结合地使用的计算机程序。计算机可读存储介质可以包括但不限于电子的、磁性的、光学的、电磁的、红外的、或半导体系统、装置或设备，或者上述内容的任何合适组合。备选地，计算机可读存储介质可以是机器可读信号介质。机器可读存储介质的更具体示例会包括基于一个或多个线的电气连接、便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦除可编程只读存储器(EPROM或快闪存储器)、光纤、便捷式紧凑盘只读存储器(CD-ROM)、光学储存设备、磁储存设备、或上述内容的任何合适组合。
为了提供与用户的交互，可以在电子设备上实施此处描述的系统和技术，该电子设备具有：用于向用户显示信息的显示装置(例如，CRT(阴极射线管)或者LCD(液晶显示器)监视器)；以及键盘和指向装置(例如，鼠标或者轨迹球)，用户可以通过该键盘和该指向装置来将输入提供给电子设备。其它种类的装置还可以用于提供与用户的交互；例如，提供给用户的反馈可以是任何形式的传感反馈(例如，视觉反馈、听觉反馈、或者触觉反馈)；并且可以用任何形式(包括声输入、语音输入或者、触觉输入)来接收来自用户的输入。
可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如，作为数据服务器)、或者包括中间件部件的计算系统(例如，应用服务器)、或者包括前端部件的计算系统(例如，具有图形用户界面或者网络浏览器的用户计算机，用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的实施方式交互)、或者包括这种后台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如，通信网络)来将系统的部件相互连接。通信网络的示例包括：局域网(LAN)、广域网(WAN)、区块链网络和互联网。
计算系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且 通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。服务器可以是云服务器，又称为云计算服务器或云主机，是云计算服务体系中的一项主机产品，以解决了传统物理主机与VPS服务中，存在的管理难度大，业务扩展性弱的缺陷。
应该理解，可以使用上面所示的各种形式的流程，重新排序、增加或删除步骤。例如，本发明中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行，只要能够实现本发明的技术方案所期望的结果，本文在此不进行限制。
上述具体实施方式，并不构成对本发明保护范围的限制。本领域技术人员应该明白的是，根据设计要求和其他因素，可以进行各种修改、组合、子组合和替代。任何在本发明的精神和原则之内所作的修改、等同替换和改进等，均应包含在本发明保护范围之内。

Claims (10)

  1. 一种PET参数图像的增强方法，其特征在于，包括：
    基于获取到的动态PET图像集，确定原始PET参数图像，并基于预设映射列表，获取与所述原始PET参数图像对应的输入图像；
    将所述输入图像输入到图像增强模型中，得到输出的预测PET参数图像；
    基于所述原始PET参数图像和所述预测PET参数图像，对所述图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为所述原始PET参数图像对应的目标PET参数图像；
    其中，所述输入图像为噪声图像、所述动态PET图像集中与预设采集时间范围对应的动态PET图像或者与所述动态PET图像对应的动态SUV图像。
  2. 根据权利要求1所述的方法，其特征在于，当所述原始PET参数图像为K 1参数图像时，所述预设采集时间范围对应的最小采集时间为0，或者，所述预设采集时间范围对应的最大采集时间为所述动态PET图像集对应的总采集时长。
  3. 根据权利要求2所述的方法，其特征在于，在基于所述原始PET参数图像和所述预测PET参数图像，对所述图像增强模型的模型参数进行调整之前，所述方法还包括：
    在所述输入图像为动态PET图像或动态SUV图像的情况下，对所述输入图像进行归一化处理，得到归一化后的输入图像；
    将归一化后的输入图像配准所述原始PET参数图像，得到配准后的输入图像。
  4. 根据权利要求1所述的方法，其特征在于，所述图像增强模型的模型架构为U-NET架构，其中，U-NET架构包括编码器和解码器，相应的，所述将所述输入图像输入到图像增强模型中，得到输出的预测PET参数图像，包括：
    将所述输入图像输入到图像增强模型中的编码器中；
    通过所述编码器中的至少两个编码卷积网络，基于输入的输入图像，输出至少两个参数特征图；
    通过所述解码器中的至少两个解码卷积网络，基于所述编码器输出的至少两个参数特征图，输出预测PET参数图像。
  5. 根据权利要求4所述的方法，其特征在于，所述编码器中每两个相邻的编码卷积网络之间设置有一个卷积层。
  6. 根据权利要求4所述的方法，其特征在于，所述解码器中每两个相邻的解码卷积网络之间设置有一个双线性插值层。
  7. 根据权利要求1-6任一项所述的方法，其特征在于，所述基于所述原始PET参数图像和所述预测PET参数图像，对所述图像增强模型的模型参数进行调整，包括：
    基于L2损失函数，确定原始PET参数图像与预测PET参数图像之间的欧式距离差；
    采用L-BFGS迭代算法，通过最小化所述欧式距离差对所述图像增强模型的模型参数进行调整。
  8. 一种PET参数图像的增强装置，其特征在于，包括：
    输入图像获取模块，用于基于获取到的动态PET图像集，确定原始PET参数图像，并基于预设映射列表，获取与所述原始PET参数图像对应的输入图像；
    预测PET参数图像确定模块，用于将所述输入图像输入到图像增强模型中，得到输出的预测PET参数图像；
    目标PET参数图像确定模块，用于基于所述原始PET参数图像和所述预测PET参数图像，对所述图像增强模型的模型参数进行调整，直到满足预设迭代次数时，将预测PET参数图像作为所述原始PET参数图像对应的目标PET参数图像；
    其中，所述输入图像为噪声图像、所述动态PET图像集中与预设采集时间范围对应的动态PET图像或者与所述动态PET图像对应的动态SUV图像。
  9. 一种电子设备，其特征在于，所述电子设备包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述至少一个处理器执行的计算机程序，所述计算机程序被所述至少一个处理器执行，以使所述至少一个处理器能够执行权利要求1-7中任一项所述的PET参数图像的增强方法。
  10. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质存储有计算机指令，所述计算机指令用于使处理器执行时实现权利要求1-7中任一项所述的PET参数图像的增强方法。
PCT/CN2022/138173 2022-09-09 2022-12-09 一种pet参数图像的增强方法、装置、设备及存储介质 WO2024051018A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211103152.5A CN115375583A (zh) 2022-09-09 2022-09-09 一种pet参数图像的增强方法、装置、设备及存储介质
CN202211103152.5 2022-09-09

Publications (1)

Publication Number Publication Date
WO2024051018A1 true WO2024051018A1 (zh) 2024-03-14

Family

ID=84071595

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/138173 WO2024051018A1 (zh) 2022-09-09 2022-12-09 一种pet参数图像的增强方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN115375583A (zh)
WO (1) WO2024051018A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115375583A (zh) * 2022-09-09 2022-11-22 深圳先进技术研究院 一种pet参数图像的增强方法、装置、设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112489158A (zh) * 2021-01-13 2021-03-12 河北大学 基于cGAN的自适应网络用于低剂量PET图像的增强方法
WO2021062413A1 (en) * 2019-09-26 2021-04-01 The General Hospital Corporation System and method for deep learning for inverse problems without training data
US20210239863A1 (en) * 2020-01-31 2021-08-05 INSERM (Institut National de la Santé et de la Recherche Médicale) Super resolution in positron emission tomography imaging using ultrafast ultrasound imaging
CN114332271A (zh) * 2021-12-07 2022-04-12 深圳先进技术研究院 基于静态pet图像的动态参数图像合成方法、系统
US20220287671A1 (en) * 2019-08-30 2022-09-15 The Research Foundation For The State University Of New York Dilated convolutional neural network system and method for positron emission tomography (pet) image denoising
CN115375583A (zh) * 2022-09-09 2022-11-22 深圳先进技术研究院 一种pet参数图像的增强方法、装置、设备及存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220287671A1 (en) * 2019-08-30 2022-09-15 The Research Foundation For The State University Of New York Dilated convolutional neural network system and method for positron emission tomography (pet) image denoising
WO2021062413A1 (en) * 2019-09-26 2021-04-01 The General Hospital Corporation System and method for deep learning for inverse problems without training data
US20210239863A1 (en) * 2020-01-31 2021-08-05 INSERM (Institut National de la Santé et de la Recherche Médicale) Super resolution in positron emission tomography imaging using ultrafast ultrasound imaging
CN112489158A (zh) * 2021-01-13 2021-03-12 河北大学 基于cGAN的自适应网络用于低剂量PET图像的增强方法
CN114332271A (zh) * 2021-12-07 2022-04-12 深圳先进技术研究院 基于静态pet图像的动态参数图像合成方法、系统
CN115375583A (zh) * 2022-09-09 2022-11-22 深圳先进技术研究院 一种pet参数图像的增强方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN115375583A (zh) 2022-11-22

Similar Documents

Publication Publication Date Title
WO2020224406A1 (zh) 图像分类方法、计算机可读存储介质和计算机设备
WO2020119679A1 (zh) 三维左心房分割方法、装置、终端设备及存储介质
Guo et al. Dual attention enhancement feature fusion network for segmentation and quantitative analysis of paediatric echocardiography
CN111369562B (zh) 图像处理方法、装置、电子设备及存储介质
WO2022213654A1 (zh) 一种超声图像的分割方法、装置、终端设备和存储介质
CN113298831B (zh) 图像分割方法、装置、电子设备及存储介质
Tan et al. Automated vessel segmentation in lung CT and CTA images via deep neural networks
CN112396605B (zh) 网络训练方法及装置、图像识别方法和电子设备
CN112767505B (zh) 图像处理方法、训练方法、装置、电子终端及存储介质
WO2024051018A1 (zh) 一种pet参数图像的增强方法、装置、设备及存储介质
CN110570394A (zh) 医学图像分割方法、装置、设备及存储介质
Xie et al. Contextual loss based artifact removal method on CBCT image
CN113888566B (zh) 目标轮廓曲线确定方法、装置、电子设备以及存储介质
WO2020253138A1 (zh) 一种分类方法、装置、设备和存储介质
CN114419375B (zh) 图像分类方法、训练方法、装置、电子设备以及存储介质
WO2022127318A1 (zh) 一种扫描定位方法、装置、存储介质及电子设备
TW202346826A (zh) 影像處理方法
WO2022227193A1 (zh) 肝脏区域分割方法、装置、电子设备及存储介质
CN115239655A (zh) 一种甲状腺超声影像肿瘤分割分类方法及装置
CN111539926B (zh) 一种图像检测方法及装置
CN114494070A (zh) 血管图像的增强方法、装置、设备和存储介质
Liu et al. MM-Net: A mixformer-based multi-scale network for anatomical and functional image fusion
Luo et al. AMCNet: attention-based multiscale convolutional network for DCM MRI segmentation
CN113112507B (zh) 灌注影像分析方法、系统、电子设备及存储介质
CN116630634B (zh) 一种图像处理方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22957977

Country of ref document: EP

Kind code of ref document: A1