WO2020162296A1 - 画像処理装置および画像処理方法 - Google Patents

画像処理装置および画像処理方法 Download PDF

Info

Publication number
WO2020162296A1
WO2020162296A1 PCT/JP2020/003268 JP2020003268W WO2020162296A1 WO 2020162296 A1 WO2020162296 A1 WO 2020162296A1 JP 2020003268 W JP2020003268 W JP 2020003268W WO 2020162296 A1 WO2020162296 A1 WO 2020162296A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
tomographic image
tomographic
creating
output
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/JP2020/003268
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
二三生 橋本
希望 大手
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hamamatsu Photonics KK
Original Assignee
Hamamatsu Photonics KK
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 Hamamatsu Photonics KK filed Critical Hamamatsu Photonics KK
Priority to US17/428,693 priority Critical patent/US11893660B2/en
Priority to EP20752373.9A priority patent/EP3923033B1/en
Priority to CN202080012799.XA priority patent/CN113454489B/zh
Publication of WO2020162296A1 publication Critical patent/WO2020162296A1/ja
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01T—MEASUREMENT OF NUCLEAR OR X-RADIATION
    • G01T1/00—Measuring X-radiation, gamma radiation, corpuscular radiation, or cosmic radiation
    • G01T1/29—Measurement performed on radiation beams, e.g. position or section of the beam; Measurement of spatial distribution of radiation
    • G01T1/2914—Measurement of spatial distribution of radiation
    • G01T1/2985—In depth localisation, e.g. using positron emitters; Tomographic imaging (longitudinal and transverse section imaging; apparatus for radiation diagnosis sequentially in different planes, steroscopic radiation diagnosis)
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T12/00—Tomographic reconstruction from projections
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T12/00—Tomographic reconstruction from projections
    • G06T12/20—Inverse problem, i.e. transformations from projection space into object space
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2210/00—Indexing scheme for image generation or computer graphics
    • G06T2210/41—Medical
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00—Image generation
    • G06T2211/40—Computed tomography
    • G06T2211/412—Dynamic
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00—Image generation
    • G06T2211/40—Computed tomography
    • G06T2211/441—AI-based methods, deep learning or artificial neural networks

Definitions

  • the present disclosure relates to an apparatus and a method for creating a tomographic image after noise removal processing based on list data collected by a radiation tomography apparatus.
  • a PET (Positron Emission Tomography) device and a SPECT (Single Photon Emission Computed Tomography) device are examples of radiation tomography devices that can acquire tomographic images of a subject (living body).
  • the PET device is equipped with a detector having a large number of small radiation detectors arranged around the measurement space in which the subject is placed.
  • the PET device detects a photon pair of energy 511 keV generated by the annihilation of electron-positron pairs in a subject into which a positron-emitting isotope (RI radiation source) has been injected by the detection unit by the coincidence counting method, and the coincidence counting is performed. Collect information. Then, a tomographic image representing the spatial distribution of the photon pair occurrence frequency in the measurement space (that is, the spatial distribution of the RI radiation source) can be reconstructed based on the collected large number of coincidence information.
  • the list data in which the coincidence counting information collected by the PET device is arranged in time series is divided into a plurality of frames in the order of collection, and the image reconstruction process is performed using the data group included in each frame of the list data.
  • This PET device plays an important role in the field of nuclear medicine and the like, and can be used to study, for example, biological functions and higher brain functions.
  • Image filters used for noise removal include a Gaussian filter and a guided filter.
  • a Gaussian filter has been used conventionally.
  • the guided filter has been developed in recent years, and is characterized in that it can well preserve the grayscale boundary in an image as compared with the Gaussian filter.
  • Patent Document 1 and Non-Patent Documents 1 and 2 describe a technique for removing noise in a dynamic PET image by a guided filter.
  • the techniques described in Patent Document 1 and Non-Patent Document 1 use, as a guidance image, an image obtained by integrating dynamic PET images composed of tomographic images of a plurality of frames during noise removal processing by a guided filter.
  • the technique described in Non-Patent Document 2 enables more effective noise removal by using a more appropriate guidance image.
  • Non-Patent Document 3 uses the Deep Image Prior technology (Non-Patent Document 4) that uses a Convolutional Neural Network (CNN), which is a type of deep neural network (DNN: Deep Neural Network), to reduce noise in PET images. Techniques for removing are described.
  • CNN Convolutional Neural Network
  • DNN Deep Neural Network
  • the noise removal processing for a PET image by the techniques described in Patent Document 1 and Non-Patent Documents 1 to 3 is superior in noise removal performance to the case where a Gaussian filter is used. However, it is desired to further improve the noise removal performance for PET images and SPECT images.
  • An object of the present invention is to provide an apparatus and method capable of creating a high-performance noise-removed tomographic image based on list data collected by a radiation tomography apparatus.
  • the embodiment of the present invention is an image processing apparatus.
  • the image processing apparatus is an apparatus for creating a noise-removed tomographic image based on list data collected by a radiation tomography apparatus, and (1) divides the list data into a plurality of frames in the order of collection, A first image creation unit that creates a first tomographic image of the frame by performing a reconstruction process using a data group included in the frame of the list data for each of the plurality of frames; A second image creating unit that creates a second tomographic image by performing reconstruction processing using a data group having a larger number of data than the data group used when creating the first tomographic image of each frame in the one image creating unit; (3) With respect to each of the plurality of frames, the convolutional neural network inputs the second tomographic image, the convolutional neural network outputs the output tomographic image, and the convolutional neural network is based on the comparison between the output tomographic image and the first tomographic image. And a CNN processing unit that generates a plurality of
  • the embodiment of the present invention is a radiation tomography system.
  • the radiation tomography system is a radiation tomography apparatus that collects list data for reconstructing a tomographic image of a subject, and creates a tomographic image after noise removal processing based on the list data collected by the radiation tomography apparatus.
  • an image processing apparatus having the above configuration.
  • the embodiment of the present invention is an image processing method.
  • the image processing method is a method for creating a noise-removed tomographic image based on list data collected by a radiation tomography apparatus, and (1) divides the list data into a plurality of frames in the order of collection, A first image creating step of creating a first tomographic image of the frame by performing reconstruction processing using a data group included in the frame of the list data for each of the plurality of frames; A second image creating step of creating a second tomographic image by performing reconstruction processing using a data group having a larger number of data than the data group used when creating the first tomographic image of each frame in the one image creating step; (3) With respect to each of the plurality of frames, the convolutional neural network inputs the second tomographic image, the convolutional neural network outputs the output tomographic image, and the convolutional neural network is based on the comparison between the output tomographic image and the first tomographic image.
  • FIG. 1 is a diagram showing the configuration of a radiation tomography system.
  • FIG. 2 is a diagram illustrating the CNN processing unit.
  • FIG. 3 is a flowchart illustrating the image processing method.
  • FIG. 4 is a flowchart illustrating the operation of the CNN processing unit and the CNN processing step S13.
  • FIG. 5 is a diagram illustrating the list data, the frame, the first tomographic image D m, and the second tomographic image S m .
  • FIG. 6 is a graph showing a temporal change in the acquisition frequency of the coincidence counting information in the subject to which the RI radiation source is applied.
  • FIG. 1 is a diagram showing the configuration of a radiation tomography system.
  • FIG. 2 is a diagram illustrating the CNN processing unit.
  • FIG. 3 is a flowchart illustrating the image processing method.
  • FIG. 4 is a flowchart illustrating the operation of the CNN processing unit and the CNN processing step S13.
  • FIG. 5 is a diagram illustrating the list
  • FIG. 7 is a diagram for explaining the simulation method, and shows (a) a numerical phantom, (b) a sinogram, (c) a noise-added sinogram, and (d) a reconstructed image.
  • FIG. 8 is a graph showing a model of changes over time in the activities of the white matter part (WM) and the gray matter part (GM) used in the simulation.
  • FIG. 9 is a diagram showing a numerical phantom image.
  • FIG. 10 is a diagram showing a first tomographic image of the sixth frame.
  • FIG. 11 is a diagram showing an image obtained by performing the noise removal processing of Comparative Example 1 on the first tomographic image of FIG. FIG.
  • FIG. 12 is a diagram showing an image obtained by performing the noise removal processing of Comparative Example 2 on the first tomographic image of FIG.
  • FIG. 13 is a diagram showing an output tomographic image selected by performing the CNN processing of this embodiment on the first tomographic image of FIG.
  • FIG. 14 is a graph showing changes over time in the PSNR of the reconstructed image in each of Comparative Example 1, Comparative Example 2, and this embodiment obtained in the simulation.
  • FIG. 15 is a graph showing a temporal change of SSIM of the reconstructed image in each of Comparative Example 1, Comparative Example 2, and this embodiment obtained in the simulation.
  • FIG. 16 is a diagram showing a second tomographic image created using the entire list data.
  • FIG. 17 is a diagram showing a first tomographic image of the sixth frame.
  • FIG. 18 is a diagram showing an image obtained by performing the noise removal processing of Comparative Example 1 on the first tomographic image of FIG.
  • FIG. 19 is a diagram showing an image obtained by performing the noise removal processing of Comparative Example 2 on the first tomographic image of FIG.
  • FIG. 24 is a diagram showing TAC in white matter (WM).
  • FIG. 25 is a diagram showing TAC in gray matter (GM).
  • FIG. 1 is a diagram showing a configuration of a radiation tomography system 1.
  • the radiation tomography system 1 includes a radiation tomography apparatus 2 and an image processing apparatus 10.
  • the image processing apparatus 10 includes a first image creating unit 11, a second image creating unit 12, a CNN processing unit 13, an image selecting unit 14, and a storage unit 15.
  • a computer having a CPU, a RAM, a ROM, a hard disk drive and the like is used.
  • the image processing apparatus 10 also includes an input unit (for example, a keyboard and a mouse) that receives an input from an operator, and a display unit (for example, a liquid crystal display) that displays images and the like.
  • an input unit for example, a keyboard and a mouse
  • a display unit for example, a liquid crystal display
  • the radiation tomography apparatus 2 is an apparatus that collects list data for reconstructing a tomographic image of the subject.
  • Examples of the radiation tomography apparatus 2 include a PET apparatus and a SPECT apparatus.
  • the radiation tomography apparatus 2 will be described as a PET apparatus.
  • the radiation tomography apparatus 2 includes a detection unit having a large number of small radiation detectors arranged around the measurement space in which the subject is placed.
  • the radiation tomography apparatus 2 detects a photon pair of energy 511 keV generated by the annihilation of electron-positron pairs in a subject into which a positron emission isotope (RI ray source) has been injected, by a coincidence counting method, This coincidence counting information is accumulated. Then, the radiation tomography apparatus 2 outputs, to the image processing apparatus 10, a list of a large number of accumulated coincidence counting information arranged in time series as list data.
  • RI ray source positron emission isotope
  • List data includes identification information and detection time information of a pair of radiation detectors that simultaneously count photon pairs.
  • the list data may further include energy information of photons detected by each radiation detector, and detection time difference information of the pair of radiation detectors.
  • the image processing device 10 reconstructs a tomographic image based on the list data.
  • an ML-EM (maximum likelihood expectation maximization) method and an iterative image reconstruction technique by a block iterative method which is an improvement thereof are known.
  • an iterative image reconstruction technique by the block iterative method an OSEM (ordered subset ML-EM) method, a RAMLA (row-action maximum likelihood algorithm) method, a DRAMA (dynamic RAMLA) method, and the like are known.
  • the image processing apparatus 10 creates a tomographic image after noise removal processing using a convolutional neural network (CNN).
  • CNN convolutional neural network
  • the first image creation unit 11 divides the list data into a plurality of frames (first to Mth frames) in the order of collection, and uses a data group included in the mth frame of the list data for each of the plurality of frames.
  • the reconstruction process is performed to create the first tomographic image D m of the m-th frame.
  • the first tomographic images D 1 to D M are dynamic PET images.
  • the second image creating unit 12 reconstructs the list data by using a data group having a larger data number than the data group used when creating the first tomographic image D m of the m-th frame in the first image creating unit 11.
  • the second tomographic image S m is created by performing the processing.
  • the second tomographic image S m may be a static PET image.
  • the second tomographic image may be created for each frame, but may be created commonly for all frames, or may be created commonly for some frames.
  • the first image creating unit 11 and the second image creating unit 12 may be common or may be provided separately.
  • the CNN processing unit 13 causes the CNN to input the second tomographic image S m for each of the first to M-th frames, and causes the CNN to output the output tomographic image O m .
  • the CNN processing unit 13 compares the output tomographic image O m with the first tomographic image D m, and learns CNN based on the comparison result.
  • the CNN processing unit 13 repeats this learning operation and generates an output tomographic image O m,n each time learning is performed.
  • the output tomographic image O m,n is an output tomographic image output from the CNN after learning (n ⁇ 1) times of the CNN using the first tomographic image D m and the second tomographic image S m for the m-th frame. is there.
  • the output tomographic image O m,1 is an output tomographic image output from the CNN in a state in which learning has never been performed.
  • the image selection unit 14 selects one of the output tomographic images O m,1 to O m,N for each of the first to Mth frames as a tomographic image after noise removal processing. To do. Preferably, the image selection unit 14 selects one of the plurality of output tomographic images O m,1 to O m,N based on the comparison between the output tomographic image O m,n and the second tomographic image S m . Select the output tomographic image of.
  • the output tomographic image O m, to n and the error between the second tomographic image S m may select the output tomographic image which minimizes the output tomographic image O m, n and a second tomographic image S m
  • Any output tomographic image may be selected from one or two or more output tomographic images in which the error between and is less than or equal to the threshold value.
  • a doctor or a technician may select any output tomographic image from the plurality of output tomographic images O m,1 to O m,N .
  • the storage unit 15 stores the list data and stores the first tomographic image D m and the second tomographic image S m of each frame.
  • the storage unit 15 also stores a plurality of output tomographic images O m,1 to O m,N of each frame , and stores an output tomographic image selected from these.
  • FIG. 3 is a flowchart explaining the image processing method.
  • This image processing method includes a list data acquisition step S10 for acquiring list data collected by the radiation tomography apparatus 2, a first image creation step S11 performed by the first image creation unit 11, and a second image creation unit 12.
  • the second image creating step S12 is performed, the CNN processing step S13 is performed by the CNN processing unit 13, and the image selecting step S14 is performed by the image selecting unit 14.
  • the list data collected by the radiation tomography apparatus 2 is acquired.
  • the list data is divided into a plurality of frames (first to Mth frames) in the order of collection, and a data group included in the mth frame of the list data is used for each of the plurality of frames.
  • the reconstruction process is performed to create the first tomographic image D m of the m-th frame.
  • the second image creating step S12 reconstruction is performed using a data group having a larger number of data than the data group used when creating the first tomographic image D m of the m-th frame in the first image creating unit 11 in the list data.
  • the second tomographic image S m is created by performing the processing.
  • the first image creating step S11 and the second image creating step S12 may be performed in any order, or may be performed in parallel.
  • the second tomographic image S m is input to the CNN and the output tomographic image O m is output from the CNN for each of the first to Mth frames.
  • CNN is learned based on the comparison between the output tomographic image O m and the first tomographic image D m . Then, this learning operation is repeated and an output tomographic image O m,n is generated each time learning is performed.
  • the image selecting step S14 for each of the first to Mth frames, one of the output tomographic images O m,1 to O m,N is selected as a tomographic image after noise removal processing. To do.
  • FIG. 4 is a flowchart illustrating the operation of the CNN processing unit 13 and the CNN processing step S13.
  • CNN is repeatedly learned N times to store N output tomographic images O m,1 to O m,N .
  • Output tomographic image O m output whenever obtaining n tomographic images O m, by comparing the n second tomographic image S m, CNN processing when the error is equal to or less than a predetermined threshold value between the two step S13 May be ended, and the finally obtained output tomographic image may be the tomographic image after the noise removal processing.
  • FIG. 5 is a diagram illustrating the list data, the frame, the first tomographic image D m, and the second tomographic image S m .
  • the horizontal axis represents the time when the list data is collected, and the time axis is divided into a plurality of frames.
  • the first tomographic image D m of the m-th frame is a tomographic image reconstructed using the data group included in the m-th frame of the list data.
  • the second tomographic image S m corresponding to the first tomographic image D m of the m-th frame is a tomographic image reconstructed using a data group having a larger data number than the data group included in the m-th frame in the list data. is there.
  • the second tomographic image S m corresponding to the first tomographic image D m of the m-th frame is created using a data group having a data number that is at least twice as large as the data group of the m-th frame.
  • the second tomographic image S m corresponding to the first tomographic image D m of the m-th frame may be created using a data group including the data group of the m-th frame, or may be created before and after the data group of the m-th frame. It may be created using a data group.
  • FIG. 6 is a graph showing the time change of the acquisition frequency of the coincidence counting information in the subject to which the RI radiation source was input.
  • the frequency of electron-positron pair annihilation (acquisition frequency of coincidence counting information) in the subject in which the RI source is turned on gradually increases as time passes immediately after the RI source is turned on, as shown in FIG. It rises, reaches a peak at a certain time Tp, and then gradually lowers after that peak.
  • the period of each frame is set to be relatively short in the period in which the coincidence counting information is frequently acquired immediately after the RI source is turned on, and the frequency of the subsequent coincidence counting information is gradually decreased.
  • the period of each frame is set relatively long. Further, the tomographic images are greatly different between the period immediately after the RI source is turned on until the frequency of acquisition of coincidence counting reaches the peak (before time Tp) and the period after the peak (after time Tp).
  • Tp the period after the peak
  • the second tomographic image S m corresponding to the first tomographic image D m of the m-th frame is the time of the list data. It is suitable to be created using a data group before Tp.
  • the second tomographic image S m corresponding to the first tomographic image D m of the m-th frame is the time Tp in the list data. It is suitable to be created using the following data group.
  • the second tomographic image S m corresponding to the first tomographic image D m of the m-th frame may be created using the entire list data.
  • the second tomographic image S m corresponding to the first tomographic image D m of the m-th frame has the maximum pixel value of the second tomographic image S m of the first It is preferable that the tomographic image D m is created by performing normalization so as to be equal to the maximum pixel value. That is, it is preferable to use the normalized second tomographic image S′ m represented by the following equation (1).
  • MaxD m is the maximum pixel value of the first tomographic image D m of the m-th frame.
  • MaxS m is the maximum pixel value of the second tomographic image S m corresponding to the first tomographic image D m of the m-th frame.
  • FIG. 7 is a diagram illustrating a simulation method.
  • a numerical phantom (Fig. 7(a)) was prepared. This numerical phantom simulates a tomographic image of the brain of a monkey into which 18 F-FDG (fluorodeoxyglucose) has been injected. This numerical phantom contains the white matter (WM) and gray matter (GM) of the brain.
  • WM white matter
  • GM gray matter
  • FIG. 8 is a graph showing a model of time change of each activity of the white matter part (WM) and the gray matter part (GM) used in the simulation.
  • TAC Time-Activity Curve
  • the total number of counts was set to 15 million.
  • the sinogram is a histogram of coincidence counting information for each pair of radiation detectors of the radiation tomography apparatus 2.
  • Each period of the first to fourth frames is 20 seconds (subtotal 80 seconds)
  • each period of the fifth to eighth frames is 40 seconds (subtotal 160 seconds)
  • each period of the ninth to twelfth frames Is set to 60 seconds (subtotal 240 seconds)
  • each period of the 13th to 16th frames is set to 180 seconds (subtotal 720 seconds)
  • each period of the 17th to 30th frames is set to 300 seconds (subtotal 4200 seconds)
  • FIG. 9 is a diagram showing a numerical phantom image.
  • FIG. 10 is a diagram showing a first tomographic image of the sixth frame.
  • FIG. 11 is a diagram showing an image obtained by performing the noise removal processing of Comparative Example 1 on the first tomographic image of FIG.
  • the noise removal processing of Comparative Example 1 is processing by a Gaussian filter.
  • FIG. 12 is a diagram showing an image obtained by performing the noise removal processing of Comparative Example 2 on the first tomographic image of FIG.
  • the noise removal processing of Comparative Example 2 is the processing described in Non-Patent Document 2.
  • FIG. 13 is a diagram showing an output tomographic image selected by performing the CNN processing of this embodiment on the first tomographic image of FIG.
  • the noise removal processing of the present embodiment preserves the boundary of light and shade in the image well and removes noise well. In addition, the unnaturalness of the processed image is reduced.
  • FIG. 14 is a graph showing a temporal change in PSNR of a reconstructed image in each of Comparative Example 1, Comparative Example 2, and this embodiment obtained in the simulation.
  • PSNR Peak Signal to Noise Ratio
  • dB decibels
  • FIG. 15 is a graph showing the temporal change of the SSIM of the reconstructed image in each of Comparative Example 1, Comparative Example 2, and this embodiment obtained in the simulation.
  • the SSIM Structuretural Similarity Index
  • the SSIM is an index that quantifies changes in image brightness, contrast, and structure, and the higher the value, the better the image quality.
  • Both of the indicators of PSNR and SSIM indicate that the noise removal processing of the present embodiment is superior in performance to the noise removal processing of Comparative Examples 1 and 2. For other frames, the performance of the noise removal processing of this embodiment is superior.
  • a rhesus monkey brain into which 18 F-FDG had been injected was used as a subject.
  • an animal PET device (SHR-38000) manufactured by Hamamatsu Photonics KK was used. Transmission measurement was performed for 30 minutes, emission measurement was performed for 90 minutes, and absorption correction was performed on the emission measurement result based on the transmission measurement result. The list data was divided into the 1st to 30th frames. A reconstructed image of each frame was created by the DRAMA method.
  • FIG. 16 is a diagram showing a second tomographic image created using the entire list data.
  • FIG. 17 is a diagram showing a first tomographic image of the sixth frame.
  • FIG. 18 is a diagram showing an image obtained by performing the noise removal processing of Comparative Example 1 on the first tomographic image of FIG.
  • FIG. 19 is a diagram showing an image obtained by performing the noise removal processing of Comparative Example 2 on the first tomographic image of FIG.
  • FIG. 20 to 23 are diagrams showing output tomographic images selected by performing the CNN process of the embodiment on the first tomographic image of FIG.
  • the noise removal process of this example is superior in performance to the noise removal process of Comparative Examples 1 and 2.
  • FIG. 24 is a diagram showing TAC in white matter (WM).
  • FIG. 25 is a diagram showing TAC in gray matter (GM).
  • the TAC changes more slowly with time in the case of the noise removal processing of the present embodiment than in the noise removal processing of Comparative Examples 1 and 2. From this, it can be seen that the performance of the noise removal processing of the present embodiment is superior.
  • Statistical analysis and model analysis can be performed by using this TAC, and a large time variation of TAC may cause a large error in the quantitative value.
  • the radiation tomography apparatus 2 is the PET apparatus in the above embodiment, it may be a SPECT apparatus.
  • the image processing apparatus is an apparatus for creating a tomographic image after noise removal processing based on list data collected by a radiation tomography apparatus.
  • a first image creating unit that creates a first tomographic image of the frame by performing reconstruction processing using a data group included in the frame of the list data for each of the plurality of frames, and (2) list
  • a second tomographic image is created by performing reconstruction processing using a data group having a larger number of data than the data group used when creating the first tomographic image of each frame in the first image creating unit.
  • the image creation unit inputs the second tomographic image to the convolutional neural network, outputs the output tomographic image from the convolutional neural network, and compares the output tomographic image with the first tomographic image.
  • a CNN processing unit that generates a plurality of output tomographic images by repeating learning of a convolutional neural network based on the configuration.
  • the above image processing apparatus may further include an image selection unit that selects one of the output tomographic images for each of the plurality of frames as the tomographic image after the noise removal processing.
  • the image selection unit may be configured to select one of the output tomographic images from the plurality of output tomographic images based on the comparison between the output tomographic image and the second tomographic image for each of the plurality of frames.
  • the second image creating unit creates the first tomographic image of the frame in the first image creating unit when creating the second tomographic image corresponding to the first tomographic image of each frame.
  • a data group including the data group used at that time may be used.
  • the second image creation unit when creating the second tomographic image corresponding to the first tomographic image of each frame, the data used when the first image creating unit creates the first tomographic image of the frame.
  • the data groups before and after the group may be used.
  • the radiation tomography system is a radiation tomography apparatus that collects list data for reconstructing a tomographic image of a subject, and after noise removal processing based on the list data collected by the radiation tomography apparatus.
  • the image processing device having the above-described configuration for creating a tomographic image is provided.
  • the image processing method is a method for creating a tomographic image after noise removal processing based on list data collected by a radiation tomography apparatus.
  • a second tomographic image is created by performing reconstruction processing using a data group having a larger number of data than the data group used to create the first tomographic image of each frame in the first image creating step.
  • the above image processing method may further include an image selection step of selecting one of the output tomographic images for each of the plurality of frames as the tomographic image after the noise removal processing.
  • one of the output tomographic images may be selected from the plurality of output tomographic images based on the comparison between the output tomographic image and the second tomographic image for each of the plurality of frames.
  • the first tomographic image of the frame is created in the first image creating step.
  • a data group including the data group used at that time may be used.
  • the second image creating step when the second tomographic image corresponding to the first tomographic image of each frame is created, the data used in creating the first tomographic image of the frame in the first image creating step The data groups before and after the group may be used.
  • the present invention can be used as an apparatus and method capable of creating a high-performance noise-removed tomographic image based on list data collected by a radiation tomography apparatus.
  • SYMBOLS 1... Radiation tomography system, 2... Radiation tomography apparatus, 10... Image processing apparatus, 11... 1st image creation part, 12... 2nd image creation part, 13... CNN processing part, 14... Image selection part, 15... Memory section.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Molecular Biology (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Image Processing (AREA)
  • Nuclear Medicine (AREA)
  • Image Analysis (AREA)
PCT/JP2020/003268 2019-02-07 2020-01-29 画像処理装置および画像処理方法 Ceased WO2020162296A1 (ja)

Priority Applications (3)

Application Number Priority Date Filing Date Title
US17/428,693 US11893660B2 (en) 2019-02-07 2020-01-29 Image processing device and image processing method
EP20752373.9A EP3923033B1 (en) 2019-02-07 2020-01-29 Image processing device and image processing method
CN202080012799.XA CN113454489B (zh) 2019-02-07 2020-01-29 图像处理装置和图像处理方法

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2019-020467 2019-02-07
JP2019020467A JP7237624B2 (ja) 2019-02-07 2019-02-07 画像処理装置および画像処理方法

Publications (1)

Publication Number Publication Date
WO2020162296A1 true WO2020162296A1 (ja) 2020-08-13

Family

ID=71947437

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2020/003268 Ceased WO2020162296A1 (ja) 2019-02-07 2020-01-29 画像処理装置および画像処理方法

Country Status (5)

Country Link
US (1) US11893660B2 (https=)
EP (1) EP3923033B1 (https=)
JP (1) JP7237624B2 (https=)
CN (1) CN113454489B (https=)
WO (1) WO2020162296A1 (https=)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7562223B2 (ja) * 2020-11-17 2024-10-07 キヤノンメディカルシステムズ株式会社 核医学診断装置および核医学データ処理装置
KR102561613B1 (ko) * 2021-12-10 2023-08-01 광주과학기술원 Ste 및 최적 중단시점 자동 결정 알고리즘을 적용한 dip를 사용하여 노이즈 이미지를 디노이징하는 방법 및 장치
KR102476433B1 (ko) * 2021-12-22 2022-12-12 이화여자대학교 산학협력단 신경망을 이용하여 영상의 노이즈를 저감하기 위한 학습 및 복원 방법과 이를 수행하는 컴퓨팅 장치
JP2023112818A (ja) * 2022-02-02 2023-08-15 浜松ホトニクス株式会社 画像処理装置および画像処理方法
JP2023173737A (ja) * 2022-05-26 2023-12-07 浜松ホトニクス株式会社 画像処理装置および画像処理方法
JP7815033B2 (ja) * 2022-05-31 2026-02-17 浜松ホトニクス株式会社 画像処理装置および画像処理方法
JP2024054668A (ja) * 2022-10-05 2024-04-17 浜松ホトニクス株式会社 画像処理装置および画像処理方法
WO2025142535A1 (ja) * 2023-12-27 2025-07-03 浜松ホトニクス株式会社 画像処理装置および画像処理方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103955899A (zh) 2014-05-02 2014-07-30 南方医科大学 基于组合图像引导的动态pet图像去噪方法
US20170039738A1 (en) * 2004-01-13 2017-02-09 Spectrum Dynamics Llc Gating with anatomically varying durations
US20180018757A1 (en) * 2016-07-13 2018-01-18 Kenji Suzuki Transforming projection data in tomography by means of machine learning
JP2018110866A (ja) * 2017-01-13 2018-07-19 キヤノンメディカルシステムズ株式会社 医用画像生成装置及び医用画像生成方法
JP2019202087A (ja) * 2018-05-25 2019-11-28 キヤノンメディカルシステムズ株式会社 医用情報処理システム、医用情報処理装置、放射線診断装置、超音波診断装置、学習用データの生産方法及びプログラム

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018048507A1 (en) 2016-09-06 2018-03-15 Han Xiao Neural network for generating synthetic medical images
RU2667879C1 (ru) * 2017-05-30 2018-09-24 Общество с ограниченной ответственностью "Аби Продакшн" Обработка и анализ данных на изображениях компьютерной томографии
EP3698269B1 (en) * 2017-11-22 2025-07-30 Zhejiang Dahua Technology Co., Ltd. An image processing method and system
US11880962B2 (en) * 2018-02-15 2024-01-23 General Electric Company System and method for synthesizing magnetic resonance images
US10685445B2 (en) * 2018-06-09 2020-06-16 Uih-Rt Us Llc Systems and methods for generating augmented segmented image set
CN109166161B (zh) 2018-07-04 2023-06-30 东南大学 一种基于噪声伪影抑制卷积神经网络的低剂量ct图像处理系统
CN109009179B (zh) * 2018-08-02 2020-09-18 浙江大学 基于深度置信网络的相同同位素标记双示踪剂pet分离方法
US10991092B2 (en) * 2018-08-13 2021-04-27 Siemens Healthcare Gmbh Magnetic resonance imaging quality classification based on deep machine-learning to account for less training data
US11446008B2 (en) * 2018-08-17 2022-09-20 Tokitae Llc Automated ultrasound video interpretation of a body part with one or more convolutional neural networks
CN109272472B (zh) 2018-10-15 2022-07-15 天津大学 面向医用能谱ct图像的噪声及伪影消除方法
US11222447B2 (en) * 2020-05-06 2022-01-11 Siemens Medical Solutions Usa, Inc. Inter-frame motion correction in whole-body direct parametric image reconstruction

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170039738A1 (en) * 2004-01-13 2017-02-09 Spectrum Dynamics Llc Gating with anatomically varying durations
CN103955899A (zh) 2014-05-02 2014-07-30 南方医科大学 基于组合图像引导的动态pet图像去噪方法
US20180018757A1 (en) * 2016-07-13 2018-01-18 Kenji Suzuki Transforming projection data in tomography by means of machine learning
JP2018110866A (ja) * 2017-01-13 2018-07-19 キヤノンメディカルシステムズ株式会社 医用画像生成装置及び医用画像生成方法
JP2019202087A (ja) * 2018-05-25 2019-11-28 キヤノンメディカルシステムズ株式会社 医用情報処理システム、医用情報処理装置、放射線診断装置、超音波診断装置、学習用データの生産方法及びプログラム

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
F. HASHIMOTO ET AL.: "Denoising of Dynamic Sinogram by Image Guided Filtering for Positron Emission Tomography", IEEE TRANSACTIONS ON RADIATION AND PLASMA MEDICAL SCIENCES, vol. 2, no. 6, 2018, pages 541 - 548, XP011698873, DOI: 10.1109/TRPMS.2018.2869936
KIBO OTE, AOI TOKUI, FUMIO HASHIMOTO, TAKASHI ISOBE, AKINORI SAITO, TOMOHIDE OMURA: "Using a Convolutional Neural Network to Eliminate PET Image Noise", MEDICAL IMAGING TECHNOLOGY, vol. 37, no. 1, 25 January 2019 (2019-01-25), pages 35 - 45, XP055837428, ISSN: 0288-450X, DOI: 10.11409/mit.37.35 *
KUANG GONG ET AL.: "PET Image Reconstruction Using Deep Image Prior", IEEE TRANSACTIONS ON MEDICAL IMAGING, 2018
LIJUN LU ET AL.: "Dynamic PET Denoising Incorporating a Composite Image Guided Filter", IEEE NUCLEAR SCIENCE SYMPOSIUM AND MEDICAL IMAGING CONFERENCE (NSS/MIC, 2014

Also Published As

Publication number Publication date
CN113454489B (zh) 2024-07-26
CN113454489A (zh) 2021-09-28
US20220114772A1 (en) 2022-04-14
EP3923033A4 (en) 2022-11-16
JP2020128882A (ja) 2020-08-27
JP7237624B2 (ja) 2023-03-13
US11893660B2 (en) 2024-02-06
EP3923033B1 (en) 2025-06-25
EP3923033A1 (en) 2021-12-15

Similar Documents

Publication Publication Date Title
JP7237624B2 (ja) 画像処理装置および画像処理方法
CN110097611B (zh) 图像重建方法、装置、设备及存储介质
EP3685350B1 (en) Image reconstruction using machine learning regularizers
US10475214B2 (en) Tomographic reconstruction based on deep learning
US12444098B2 (en) Image processing device and image processing method
US7949172B2 (en) Iterative image processing
CN110415311B (zh) Pet图像重建方法、系统、可读存储介质和设备
Whiteley et al. FastPET: Near real-time PET reconstruction from histo-images using a neural network
JP6974159B2 (ja) 画像処理装置および画像処理方法
CN105488824B (zh) 一种重建pet图像的方法和装置
Rashid et al. Optimization of PET Image Reconstruction for Enhanced Image Quality in Various Tasks Using a Conventional PET Scanner
JP6495615B2 (ja) 画像処理装置および画像処理方法
JP6986961B2 (ja) 画像処理装置および画像処理方法
Chen et al. A novel edge-preserving filter for medical image enhancement
JP2023173737A (ja) 画像処理装置および画像処理方法
JP7018306B2 (ja) 画像処理装置および画像処理方法
US20250157099A1 (en) Image processing device and image processing method
Suzuki et al. Image Correction in Emission Tomography Using Deep Convolution Neural Network
WO2024224878A1 (ja) 画像処理装置および画像処理方法
WO2024035674A1 (en) Methods and related aspects for producing processed images with controlled images quality levels
JP2025037333A (ja) 減弱分布画像作成装置、画像処理装置、放射線断層撮影システム、減弱分布画像作成方法、画像処理方法
Bian et al. Dynamic PET image reconstruction using a spatial-temporal edge-preserving prior

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: 20752373

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2020752373

Country of ref document: EP

Effective date: 20210907

WWG Wipo information: grant in national office

Ref document number: 2020752373

Country of ref document: EP