Radiation diagnostic apparatus, method and program

Abstract

To improve a dead-time correction of measured radiation.SOLUTION: A radiation diagnosis device according to an embodiment is provided with a processing unit. The processing unit is configured to: acquire radiation data showing intensity of radiation detected by a plurality of detection elements, and including a count filtered using an energy band-pass filter of an energy range including energy of the radiation when not scattering; estimate a singles spectrum of radiation including a count other than the count filtered using the energy band-pass filter using a scatter model on the basis of the radiation data; and correct the radiation data for detection omission occurring from dead time of each of the plurality of detection elements on the basis of the singles spectrum, and thereby implement a dead-time correction generating the corrected radiation data.SELECTED DRAWING: Figure 4A

Description

[Technical field]

[0001] The embodiments disclosed in the present specification and drawings relate to a radiological diagnostic apparatus, a method, and a program. [Background technology]

[0002] The embodiments described herein relate to estimating a singles spectrum from radiation measurements showing energy window filtered counts, and in particular to using the estimated singles spectrum to perform dead time correction of the measured radiation.

[0003] Positron Emission Tomography (PET) is a nuclear medicine imaging method based on the use of weakly radiolabeled agents (tracers) to image specific features of the body. PET images display the spatial distribution of radiopharmaceuticals, allowing a physician or clinician to draw conclusions about, for example, metabolic activity or blood flow.

[0004] In PET imaging, a tracer agent is administered to the subject (e.g., a patient) to be imaged, for example, by injection, inhalation, or ingestion. After administration, the tracer agent is concentrated in a specific location in the subject's body due to its physical or biomolecular properties. The actual spatial distribution of the tracer agent, the intensity of the area of ​​accumulation of the tracer agent, and the kinetics of the process from administration to final excretion are all important factors that may have clinical significance.

[0005] During this process, the tracer applied to the tracer agent emits a positron, which is the antimatter equivalent of an electron. When the emitted positron collides with an electron, the electron and positron are annihilated, resulting in the emission of a pair of gamma rays, each with an energy of 511 keV, traveling in directions approximately 180 degrees apart from each other.

[0006] The spatiotemporal distribution of the tracer is reconstructed using the principles of tomographic reconstruction, for example, by characterizing each detection event for its energy (i.e., amount of light emitted), location, and timing. If two gamma rays are detected within a coincidence time window, they are identified as a coincidence pair, since they are presumed to have arisen from the same positron annihilation event. By drawing a line between their positions, i.e., the Line-Of-Response (LOR), the approximate location of the positron annihilation event can be identified. Such timing information can also be used to measure the statistical distribution along the LOR of annihilation based on the Time-Of-Flight (TOF) information of the two gamma rays. Also, by accumulating a large number of LORs, tomographic reconstruction can be performed to obtain a volumetric image of the spatial distribution of radiation (e.g., tracer density) inside the subject.

[0007] Here, some counts are not recorded due to the dead time of the detector. For example, a detector detects one quantum of radiation (e.g., an X-ray or a gamma ray) and then has a dead time during which no detection occurs. The image quality can be improved by correcting the acquired data to account for the counts that arrived during the dead time and were missed. This dead time correction depends on how accurately the flux rate can be determined. However, in PET devices that filter out the counts outside the energy window and thus filter out the scattered events, the accuracy of the flux rate is limited because the flux rate is not measured taking into account the flux over the entire singles spectrum.

[0008] Therefore, an improved method of dead-time correction is desired for estimating the singles spectrum from a windowed portion of the radiation spectrum. [Prior art documents] [Patent documents]

[0009] [Patent Document 1] International Publication No. 2019 / 145398 [Patent Document 2] US Patent Application Publication No. 2019 / 0066342 [Non-patent literature]

[0010] [Non-Patent Document 1] John M Ollinger,"Model-based scatter correction for fully 3D PET",Physics in Medicine & Biology,Volume 41,Number 1,153-176,1996 Summary of the Invention [Problem to be solved by the invention]

[0011] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to improve dead time correction of measured radiation. However, the problems to be solved by the embodiments disclosed in this specification and the drawings are not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described later can also be positioned as other problems. [Means for solving the problem]

[0012] A radiological diagnostic apparatus according to an embodiment includes a processor. The processor acquires radiation data indicating the intensity of radiation detected by a plurality of detector elements and including counts filtered using an energy bandpass filter of an energy range including the energy of the radiation when not scattered. The processor estimates a singles spectrum of the radiation using a scattering model based on the radiation data, including counts other than the counts filtered using the energy bandpass filter. The processor also performs deadtime correction based on the singles spectrum to generate corrected radiation data by correcting the radiation data for missed detections resulting from deadtimes of each of the plurality of detector elements. [Brief description of the drawings]

[0013] A more complete understanding of the present invention and many of the attendant advantages thereof will be readily obtained as the same becomes better understood by reference to the following detailed description when considered in conjunction with the accompanying drawings, in which:

[0014] [Figure 1A] FIG. 1A illustrates a PET image of a small subject in accordance with one embodiment. [Figure 1B] FIG. 1B illustrates an energy spectrum of gamma rays acquired for the field of view (FOV) shown in FIG. 1A according to one embodiment. [Figure 2A] FIG. 2A illustrates a PET image of a large subject according to one embodiment. [Figure 2B] FIG. 2B illustrates an energy spectrum of gamma rays acquired for the field of view (FOV) shown in FIG. 2A according to one embodiment. [Figure 3A] FIG. 3A is a diagram showing the FOV of true coincidence counting in a PET device according to one embodiment. [Figure 3B] FIG. 3B is a diagram showing the FOV of random coincidence counting in a PET device according to one embodiment. [Figure 4A] FIG. 4A is a flow diagram illustrating a method for performing dead-time correction after estimating a singles spectrum from energy filtered counts and then performing image reconstruction according to one embodiment. [Figure 4B] FIG. 4B is a flow diagram illustrating a method for estimating a singles spectrum using attenuation and emission images and then performing dead-time correction according to one embodiment. [Diagram 5] FIG. 5 is a schematic diagram illustrating scattering in a PET device according to one embodiment. [Figure 6A] FIG. 6A is a schematic diagram illustrating Compton scattering according to one embodiment. [Figure 6B]FIG. 6B shows a plot of electron and photon energy as a function of angle for Compton scattering in accordance with one embodiment. [Figure 7] FIG. 7 is a flow diagram illustrating a method for estimating a singles spectrum using a deep learning (DL) network according to one embodiment. [Figure 8] FIG. 8 is a perspective view showing a PET device according to an embodiment. [Figure 9] FIG. 9 is a schematic diagram showing a PET device according to an embodiment. [Figure 10] FIG. 10 is a schematic diagram illustrating primary flux, first scattered flux, and multi-scattered flux according to one embodiment. [Figure 11] FIG. 11 is a flow diagram illustrating a process for implementing the Radiative Transfer Equation (RTE) method according to one embodiment. [Figure 12] FIG. 12 is a flow diagram illustrating training a DL network to estimate a singles spectrum according to one embodiment. [Figure 13] FIG. 13 is a diagram illustrating a DL network according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] According to an aspect of an embodiment, there is provided a radiological diagnostic apparatus including a processor. The processor acquires radiation data indicating an intensity of radiation detected by a plurality of detector elements and including counts filtered using an energy bandpass filter of an energy range including the energy of the radiation when not scattered. The processor also estimates a singles spectrum of the radiation using a scattering model based on the radiation data, including counts other than the counts filtered using the energy bandpass filter. The processor also performs deadtime correction based on the singles spectrum to generate corrected radiation data by correcting the radiation data for missed detections resulting from deadtimes of each of the plurality of detector elements.

[0016] According to another aspect of the embodiment, there is provided a method for controlling a radiological diagnostic apparatus having a processor, the method including the steps of: acquiring radiation data by the processor indicating an intensity of radiation detected by a plurality of detector elements and including counts filtered using an energy band-pass filter of an energy range including an energy of the radiation when not scattered; estimating a singles spectrum of the radiation using a scattering model based on the radiation data, the singles spectrum including counts other than the counts filtered using the energy band-pass filter; and performing dead-time correction by the processor to generate corrected radiation data by correcting the radiation data for detection errors resulting from dead-times of each of the plurality of detector elements based on the singles spectrum.

[0017] In PET, gamma rays are detected by a combination of scintillator crystals and photodetectors, such as Photomultiplier Tubes (PMTs) or Silicon Photomultipliers (SiPMs). After detecting a gamma ray, these photodetectors have a period of dead time before the photodetector recovers and can detect another gamma ray. During this dead time, gamma rays incident on the photodetector are not detected, resulting in missed counts. The number of missed counts can be measured based on the incident flux rate, the dead time, and photon statistics (e.g., Poisson distribution). Thus, by estimating the total flux (e.g., full singles spectrum), the measured counts can be corrected for the dead time loss, thereby improving accuracy and image quality.

[0018] In PET, a narrow energy window for the detected counts is often used to remove scatter events. In conventional methods, the dead time estimation is based on the measured counts and excludes the counts outside the energy window. Such conventional methods underestimate the dead time because they do not consider gamma rays outside the energy window, which also causes dead time loss. Therefore, estimating the count rate outside the energy window would be useful for more accurate estimation and correction of dead time loss. Thus, a method that takes into account counts both outside and inside the energy window would lead to more accurate scaling of PET reconstructions, which is important for clinical diagnosis and treatment monitoring.

[0019] As mentioned above, dead time occurs whether the gamma rays detected have an energy of 511 keV (i.e., direct gamma rays from a positron emission event) or are scattered (e.g., gamma rays resulting from Compton scattering of direct gamma rays). In a PET device, the measured singles counts are selected to be within an energy window that spans an energy range that peaks at 511 keV for positron emission. As mentioned above, performing dead time correction using only counts within the energy window would result in an underestimation of the total flux (i.e., gamma ray flux due to both direct / primary gamma rays and scattered gamma rays), and therefore an underestimation of the dead time loss. Furthermore, the ratio of primary flux to scattered flux may vary from one object to another, since gamma rays passing through a larger object pass through more material that is more likely to scatter the gamma rays.

[0020] Furthermore, in PET, the measured coincidence counts include both true coincidence counts and background signals (e.g., random coincidence counts). The background signal includes counts due to random events and scatter events. In PET, the background signal is mainly composed of Accidental Coincidence (AC) and scatter. Incidental coincidence counts are also called random.

[0021] For many annihilation events, one photon of a photon pair is absorbed or scattered outside the plane of the PET detector ring, so only the other photon is detected. Additionally, some photons that reach the scintillating detectors of the PET detector ring are not detected because they are less than the unity quantum efficiency of the detector. A detection event in which one photon of a photon pair is detected is also called a "single". This includes cases where only one of the photon pairs is detected, and cases where both photon pairs are detected. If two singles from separate annihilations are detected within a coincidence timing window, they are erroneously recorded as arising from the same annihilation. This is called an accidental coincidence (AC) event or random event. In other words, an AC event occurs when two unrelated singles are detected within the coincidence timing window. If an AC event has one LOR that passes outside the object being imaged (OBJ), the detection is said to be in the "wing".

[0022] Photons scattered in the body leave the detector plane mostly undetected, but some are detected and recorded, resulting in an inaccurate LOR. In some embodiments, some of these scattering events that cause an inaccurate LOR can be eliminated by energy discrimination because the photons lose a portion of their energy during the Compton interaction that causes the scattering event. Even so, some scattered photons (scatter) and some random coincidences (random) are unavoidably recorded, and as a result, the background signal includes random and scattered.

[0023] Singles may include not only AC, but also random and scatter that do not occur within the time window of another gamma ray and therefore are not recorded as part of a gamma coincidence pair. All gamma rays detected, whether true coincidences, AC, or random and scatter that are not recorded as coincidences (i.e., all singles), cause a dead-time period in the detector during which the detector does not detect gamma rays and thus misses counts. The accuracy of the dead-time correction depends on how accurately the totality of missed counts can be estimated. This requires determining the total singles that cause the dead-time. In addition to the above time windows used to remove AC, an energy window is also used to remove scatter with energies less than 511 keV, which indicates positron emission in the absence of scatter. Since singles of all energies cause dead-time in the detector response, it is preferable for the dead-time correction to take into account the complete spectrum of singles counts.

[0024] In this specification, a count of multiple singles is referred to as a "singles count," a spectrum of singles counts is referred to as a "singles spectrum," and a complete spectrum of singles counts is referred to as a "full singles spectrum."

[0025] The methods described herein estimate a full singles spectrum of gamma rays based on emission and attenuation images of a subject, which can then be used to improve correction of dead-time losses.

[0026] Referring now to the drawings, in which like reference numbers indicate the same or similar parts throughout the several views, FIG. 1A shows a PET image of a small subject, and FIG. 1B shows a singles count spectrum of the small subject. The singles count spectrum represents singles counts as a function of energy. In the singles count spectrum, the positron emission peak is centered at 511 keV, but many gamma rays are lower in energy due to inelastic scattering (e.g., Compton scattering). In PET scans, an energy window is often used to selectively record counts around the 511 keV peak. However, knowledge of the full singles spectrum can improve dead-time correction of PET imaging. Thus, the method described herein estimates the full singles spectrum from counts measured using an energy window.

[0027] Similar to FIGS. 1A and 1B, FIG. 2A shows a PET image of a large object, and FIG. 2B shows a singles count spectrum of the large object. As can be seen by comparing FIG. 1B and FIG. 2B, a large object has a large percentage of scattered gamma rays. Furthermore, in the example shown in FIG. 2A, a large percentage of the emission density is outside the imaged FOV. As can be seen, various factors contribute to the relative number of gamma counts corresponding to energies inside and outside a given energy window. A method that models these various factors to estimate the full singles spectrum and provide the ratio of counts inside and outside the energy window improves the accuracy of the dead-time loss correction, resulting in improved accuracy of the PET image and improved image quality.

[0028] 1A and 2A, multiple images are stitched together from a series of single scans, each of which corresponds to the single scan width (i.e., the width of the FOV of a single bed scan) shown in Figures 1A and 2A. Note that a single scan is a single scan, and a single bed scan is a scan performed at one couch position.

[0029] 3A shows the PET device 800 in relation to an imaged object OBJ. Fig. 3A also shows an imaging FOV where both gamma ray pairs reach multiple detectors without scattering (i.e., true coincidence FOV). For example, the FOV shown in Fig. 3A may be the true coincidence FOV when the PET device 800 performs a single couch scan.

[0030] FIG. 3B shows a schematic diagram of the region where gamma rays that give rise to random coincidences originate. Thus, random coincidences can originate from regions outside the imaging FOV (i.e., the FOV of true coincidences shown in FIG. 3A). Singles include scattering in addition to random (i.e., random coincidences). Thus, singles can include gamma rays where scattering events occur within the random FOV, and gamma ray emissions occur both within and outside the random FOV. That is, as long as there are scattering events within the random coincidence FOV, singles spectra corresponding to scattered gamma rays can result from gamma rays that originate / emit both within and outside the random coincidence FOV.

[0031] In two-dimensional (2D) PET and / or three-dimensional (3D) PET, scattering effects are one of the most significant physical factors that degrade image quality. In a typical 3D PET system, scattering events can be between 30% and 50% of the total detected events. Model-based scattering estimation can provide an accurate estimate of scattering. When the subject is axially longer than the PET device, it is preferable to take into account out-of-FOV scattering (i.e., scattered gamma rays coming from outside the imaging FOV). In particular, when the PET system has a large aperture without end shielding, such out-of-FOV scattering can exceed 40% of the total scattering. Out-of-FOV scattering events are classified into two categories, similar to the general scattering described above: (i) events that occur inside the FOV when one or both of the detected photons are scattered from materials placed outside the FOV, and (ii) events that arise from annihilation events (i.e., activity) that occur outside the FOV. For both categories, the attenuation distribution and activity distribution outside the FOV can be used to model the detected scattering distribution.

[0032] To better understand the methods described herein, it is to be understood the difference between coincidence counting and single counting. In PET image reconstruction, coincidence counting is important because it determines the LOR of the positron emission events. However, in dead time correction, singles counting is important because all counts, whether coincidence or not, result in detector dead time. Thus, it is to be understood that when the context of the description herein is intended for PET image reconstruction, the relevant gamma ray detection events are coincidences, and when the context of the description herein is intended for dead time correction, for example, the relevant gamma ray detection events are singles.

[0033] In the context of PET image reconstruction, two conditions are often imposed for gamma rays scattered into the FOV from outside the FOV to be detected and recorded. First, the event must scatter at an angle that causes the gamma ray to hit the detector surface. Second, after scattering, the energy of the gamma ray must be within an energy window. In terms of these two detection conditions, the majority of events outside the FOV occur at directly adjacent couch positions, because the probability of scattering (i.e., the scattering cross section) increases as the scattering angle decreases, and because scattering events occurring at directly adjacent couch positions are scattered and detected at larger angles.

[0034] In view of the energy window used to filter out the scatter, the counts acquired in the PET scan do not include detection of scattered gamma rays with energies lower than the energy window. Nevertheless, the counts of these low energy gamma rays contribute to the dead time, and therefore, in order to perform the dead time correction as accurately as possible, the leakage of counts caused by the dead time caused by low energy gamma rays should be taken into account. Thus, Figures 4A and 4B show a flow diagram of a method 400 for estimating a full singles spectrum from the acquired counts (e.g., the counts filtered within the energy window). The estimated full singles spectrum 415 is then used to perform a dead time correction of the PET data (raw scan data) 407, and a PET image 435 is reconstructed from the corrected PET data 425.

[0035] First, in step 410, a full singles spectrum 415 is estimated from the emission image 403 and the attenuation image 405. Figure 4A shows that PET data 407 can optionally be an input to step 410. In some embodiments, the PET data 407 can be used to generate the emission image 403 and the attenuation image 405.

[0036] For example, a joint-estimation PET method can be used to simultaneously acquire both an emission image 403 (also referred to as an emission map or activity distribution) and an attenuation image 405 (also referred to as an attenuation map). When using a joint-estimation PET method, the emission image 403 and the attenuation image 405 can be generated from the PET data 407.

[0037] In another embodiment, the PET data 407 can be acquired using a PET / CT scanner that scans the object OBJ being imaged to generate both PET and X-ray Computed Tomography (CT) data. The PET data can be used to generate the emission image 403, and the CT data can be used to generate the attenuation image 405. Additional techniques can also be used to generate the emission image 403 and the attenuation image 405. For example, the emission image 403 can be generated from the PET data 407. The attenuation image 405 is used to estimate the scattering cross section of the gamma rays as a function of position within the object OBJ being imaged.

[0038] A variety of methods can be used to estimate the full singles spectrum 415 from the emission image 403 and the attenuation image 405. In each method, the attenuation image 405 can be used to generate a scattering cross section model (e.g., a model that represents the scattering cross section of gamma rays as a function of position). Different materials, such as bone and water, have different scattering cross sections due to their different angular dependences that depend on the atomic constituents. The magnitude of the scattering cross section can also depend on the density of the material components at a given point. Thus, in some embodiments, material decomposition can be used to decompose the attenuation image 405 into material components to determine the density of various material components as a function of position.

[0039] For example, voxels with Hounsfield values ​​above a certain threshold can be determined as bone, while voxels with Hounsfield values ​​below another threshold can be identified as soft tissue. Thresholding and region growing methods can then be used to segment the attenuation image 405 into material components. Furthermore, if spectral CT was used to generate the attenuation image 405, material decomposition can be used to determine the scattering cross section as a function of location. Furthermore, human body atlases and / or machine learning methods can be used to determine the scattering cross section as a function of location from the attenuation image 405, as well as other methods that will be appreciated by those skilled in the art.

[0040] Thus, in some embodiments, a location-specific scattering cross section can be obtained from the attenuation image 405 and from material components derived from the attenuation image 405. The location-specific x-ray scattering cross section can be obtained as a linear superposition of the products of the volume pixel (i.e., voxel) density fractions of the material components of the material decomposition and the pre-calculated scattering cross sections corresponding to each material component.

[0041] Returning to the various methods used to estimate the full singles spectrum 415 from the emission image 403 and the attenuation image 405, in one embodiment, a neural network can be trained to estimate the full singles spectrum 415 in response to the input of the emission image 403 and the attenuation image 405. In another embodiment, a Monte Carlo (MC) method can be used to estimate the full singles spectrum 415 based on the emission image 403 and the attenuation image 405. In another embodiment, a Radiative Transfer Equation (RTE) method can be used to estimate the full singles spectrum 415 based on the emission image 403 and the attenuation image 405. In another embodiment, an analytical calculation can be used to estimate the full singles spectrum 415. Each of these embodiments will be described in more detail below.

[0042] Also, in step 420, using the full singles spectrum 415, dead time correction is performed on the PET data. The dead time of the gamma ray detector can depend on the flux and energy spectrum of the incident gamma rays. [Number] In the above equation, E is the photon energy, τ(E) is the dead time induced by a single photon having the energy E, and p(E) is the incident flux distribution as a function of the energy E. If the dead time is assumed to be a constant rather than depending on the gamma ray energy, the above equation is simplified as follows. [Number] In the above equation, N Total = ∫0 ∞ p(E)dE is the count rate. In practice, the measured counts only include a narrow window [E1, E2] where E1 < 511 keV < E2. That is, the PET device acquires the incident flux integrated within the narrow window [E1, E2]. Nevertheless, using the measured count rate N meas = ∫ E1 E2 p(E)dE, the dead time per count τ can be scaled to match the effective dead time per count τ c to calculate the dead time. Here, the dead time is the dead time per measured count rate N meas and is calculated as follows. [Number] In the above equation, p c (E) is the flux variance during the calibration of the dead time. The total count N total and the measured count N measIf the ratio of p(E) to p(E) is subject independent, the dead time for a given flux rate p0(E) during a PET scan is calculated as follows:

number

number

[0043] Also, in step 430, a PET image is reconstructed from the corrected PET data 425. In PET, the emission data can be organized as sinograms (similar to projection data in X-ray CT), and thus activity images showing activity / tracer density as a function of positron / voxel index can be reconstructed using methods similar to those used to reconstruct attenuation images in X-ray CT. Such reconstruction methods include, for example, Filtered Back-Projection (FBP) and Iterative Reconstruction (IR) reconstruction methods.

[0044] In some embodiments using IR reconstruction techniques, the reconstructed image x(x) is generated by solving an optimization problem.

number

[0045] In some embodiments, the regularizer can be a patch-based edge-preserving penalty function, where the penalty is calculated using neighboring patches rather than individual pixels, as illustrated in the following equation:

number

number

[0046] Returning to step 410, four methods for estimating the full singles spectrum are described in detail: (i) Monte Carlo (MC) simulation, (ii) analytical calculation, (iii) radiative transport equation (RTE) method, and (iv) machine learning method.

[0047] First, the full singles spectrum 415 can be generated using MC simulation methods such as the Geant4 Tomographic Emission Application (GATE) toolkit, as described in S. Jan et al., "GATE Tomographic Emission Application of Geant4, a Simulation Toolkit for PET and SPECT," Phys Med Biol. Vol. 49, pp. 4543-4561, 2004, the entire contents of which are incorporated herein by reference. MC simulations can be computationally intensive and slow to run a sufficient number of simulations to overcome the inherent noise in the MC simulation, but they have their own advantages. For example, MC simulations can be accurate and the simulation results can provide a ground truth. This is important for evaluating the system and validating some of the other algorithms described below. Additionally, some hidden variables in the real acquisition, such as scattering and randomness, are measured directly in the Monte Carlo simulation. Using the reconstructed emission image 403 and attenuation image 405, the MC simulation directly generates the full singles spectrum corresponding to the PET acquisition.

[0048] Second, the full singles spectrum 415 can be generated using analytical calculations as described in CC Watson, "New Fast Image-Based Scatter Correction for PET in 3D," IEEE Trans. Nucl. Sci., vol. 47, no. 4, pp. 1587-1594, 2000, the entire contents of which are incorporated herein by reference. Figure 5 illustrates first order scattering in a PET device 800. For example, first order scattering is calculated by the following equation:

number

number

[0049] FIG. 6A shows the relationship between the energy and momentum of the incident gamma ray, the outgoing gamma ray, and the scattered electron for Compton scattering. The angles φ and θ are determined by the fact that scattering follows the laws of conservation of energy and momentum. The energy of the scattered electron is retained in the first crystal. Furthermore, the remaining energy is carried away by the outgoing gamma ray. As a result, if there is only a single scattering event, the energy of the outgoing gamma ray is retained in the second crystal (e.g., at a wavelength in the optical, ultraviolet, or infrared spectrum) by converting the energy of the outgoing gamma ray into a second photon via photoelectric absorption. In general, these energies are related by the "Compton shift", which is the difference between the wavelengths of the outgoing and incident gamma ray, given by the following equation: λ'-λ=h / (mc)(1-cosθ) In the above equation, λ is the initial wavelength, λ' is the wavelength after scattering, h is Planck's constant, m is the rest mass of the electron, c is the speed of light, and θ is the scattering angle of the gamma ray. The energy is inversely proportional to the wavelength by the Planck's constant h, i.e., E=h / λ. Figure 6B shows the relationship between the scattered photon energy and the electron energy as a function of the scattering angle, with the initial gamma ray photon energy set to 511 keV.

[0050] From the above, the energy E' of the scattered photon is given by the following formula:

number

number

[0051] As a third method, the full singles spectrum 415 can be generated using the RTE method as described in Y. Lu et al., "Deterministic Integrated Spherical Harmonic Method for Scattering Simulation in Computed Tomography," Proc. SPIE 10132, Medical Imaging 2017: Physics of Medical Imaging, p. 101322H, 2017, the entire contents of which are incorporated herein by reference. Compared to the MC method, the RTE can calculate first-order and multi-scattering events, but does not suffer from statistical noise like the MC method. The RTE method solves the following problems:

number

number

[0052] When discretizing the entire volume, each discretized region is considered a scattering source. Successively higher order scattering can be included by iteratively computing the scattering map using the following formula:

number

number

number

[0053] Fourth, the full singles spectrum 415 can be generated using a machine learning model such as an Artificial Neural Network (ANN), also referred to herein as a Deep Learning (DL) network. A DL network is trained using a training data set to learn implicit patterns in the data, allowing it to rapidly predict the full singles spectrum from the emission and attenuation images 403 and 405. Although training a DL network can take a significant amount of time, training the DL network only occurs once, and the trained DL network is then used iteratively for each image acquisition. Using a DL network to estimate the full singles spectrum can reduce computational costs compared to MC simulations or analytical calculations.

[0054] FIG. 7 shows a flow diagram of a method 700 for PET imaging in which an emission image 403 and an attenuation image 405 are used to estimate a full singles spectrum 415, which is then used to correct the PET data 407 used to reconstruct a PET image 435.

[0055] Here, steps 410, 420, and 430 are performed similarly to the example described with reference to Figures 4A and 4B, except that in step 410, a DL network 470 is received as input and used to estimate the full singles spectrum 415. Furthermore, Figure 7 illustrates that PET data 407 (raw data) is used as input in step 420, rather than in step 410 as shown in Figure 4A. In general, step 410 can be performed without PET data 407. However, in some embodiments, PET data 407 can be used in step 410 (e.g., to determine a scale factor when analytical equations are used to calculate scatter). Thus, both embodiments in which PET data 407 is used as input in step 410 and those in which it is not used are considered embodiments of the disclosed method. Furthermore, the PET data 407 can be used to generate an emission image 403 and an attenuation image 405 (e.g., using a joint estimation PET method).

[0056] Also in process 460, input data 457 and target data 453 are used to train DL network 470. Input data 457 includes pairs of emission and attenuation images used as input to the DL network. Target data 453 is the full singles spectrum corresponding to each pair of emission and attenuation images included in input data 457. Target data 453 provides a gold standard against which the output of DL network 470 is compared such that weighting coefficients between nodes of DL network 470 are adjusted (e.g., using backpropagation) until DL network 470 is fully trained to generate an output that matches the full singles spectrum of target data 453. The method of training DL network 470 in process 460 is described in detail below.

[0057] 8 and 9 show a PET device 800 that includes several Gamma-Ray photon counting detectors (GRDs) (e.g., GRD1, GRD2, ... GRDN) that are configured as rectangular detector modules. In one embodiment, the detector ring includes 40 GRDs. In another embodiment, 48 GRDs are used, and a greater number of GRDs are used to create a larger aperture PET device 800.

[0058] Each GRD may include a two-dimensional array of individual detector crystals that absorb gamma radiation and emit scintillation photons. The scintillation photons may be detected by a two-dimensional array of photomultiplier tubes (PMTs) disposed within the GRD. A light guide may be disposed between the detector crystals and the array of PMTs. Additionally, each GRD may include several PMTs of various sizes, each positioned to receive scintillation photons from multiple detector crystals. Each PMT may generate an analog signal indicating when a scintillation event occurred and the energy of the gamma ray that produced the detection event. Additionally, photons emitted from one detector crystal may be detected by multiple PMTs, and based on the analog signal generated by each PMT, the detector crystal corresponding to the detection event may be determined, for example, using Anger logic and crystal decoding.

[0059] FIG. 9 shows a schematic diagram of a PET device system having a GRD arranged to detect gamma rays emitted from an object OBJ. The GRD can measure the timing, position and energy corresponding to each gamma ray detection. In one embodiment, multiple gamma ray detectors are arranged in a ring as shown in FIGS. 8 and 9. The detector crystals may be scintillator crystals each having individual scintillator elements arranged in a two-dimensional array, and the scintillator elements may be any known scintillating material. The PMTs are arranged such that light from each scintillator element is detected by the multiple PMTs to enable Anger calculation and crystal decoding of the scintillation events.

[0060] 9 shows a PET device 800 in which an object OBJ to be imaged is placed on a bed top 816, and GRD modules GRD1 to GRDN are arranged to surround the object OBJ and the bed top 816. The GRD is fixedly connected to a circular part 820 that is fixedly connected to a gantry 840. The gantry 840 houses many parts of the PET device. The gantry 840 of the PET device further includes an open opening through which the object and the bed top 816 can pass, and gamma rays emitted in opposite directions from the object OBJ due to an annihilation event are detected by the GRD, and the coincidence counting of the gamma ray pair can be determined using timing and energy information.

[0061] 9 also shows circuitry and hardware for acquiring, storing, processing and distributing gamma ray detection data. The circuitry and hardware includes a processor 870, a network controller 874, a memory 878 and a data acquisition system (DAS) 876. The PET device further includes a data channel for transmitting detection measurements from the GRD to the DAS 876, the processor 870, the memory 878 and the network controller 874. The data acquisition system 876 can control the acquisition, digitization and transmission of detection data from the detector. In one embodiment, the DAS 876 controls the operation of the couch top 816. The processor 870 performs various functions, including reconstructing an image from the detection data according to methods 400 and 700, performing pre-reconstruction processing of the detection data, and performing post-reconstruction processing of image data, as described herein. Here, the processor 870 is an example of a processing unit.

[0062] The processor 870 is configured to execute the methods 400 and 700 described herein. The processor 870 may include a central processing unit (CPU) that may be implemented as individual logic gates such as an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or another complex programmable logic device (CPLD). The FPGA or CPLD may be implemented by coding in a hardware description language such as VHDL, Verilog, etc., and the code may be stored directly in the electronic memory in the FPGA or CPLD or in a separate electronic memory. Here, the electronic memory may be a non-volatile memory such as a read only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. Alternatively, the electronic memory may be a volatile memory such as a static or dynamic RAM, and a processor such as a microcontroller or microprocessor may be provided to manage the electronic memory and the interaction of the FPGA or CPLD with the memory.

[0063] Alternatively, the CPU of the processor 870 may execute a computer program including a set of computer readable instructions to perform the methods 400 and 700 described herein. The program may be stored in a non-transitory electronic memory and / or a hard disk drive, a compact disc (CD), a digital versatile disc (DVD), a flash drive, or other known storage medium as described above. Furthermore, the computer readable instructions may be provided as a utility application, a background daemon, a component of an operating system, or a combination thereof, and may be implemented in a processor such as an Intel Xenon processor or an AMD Opteron processor, and an operating system such as Microsoft VISTA, UNIX, Solaris, LINUX, Apple, MAC-OS, and other operating systems known to those skilled in the art. Furthermore, the CPU may be implemented as multiple processors cooperating in parallel to execute instructions.

[0064] In one embodiment, the reconstructed image may be displayed on a display, which may be a Liquid Crystal Display (LCD) display, a Cathode Ray Tube (CRT) display, a Plasma display, an Organic Light Emitting Diode (OLED) display, a Light Emitting Diode (LED) display, or any other display known in the art.

[0065] Memory 878 may be a hard disk drive, a CD-ROM drive, a DVD drive, a flash drive, Random Access Memory (RAM), ROM, or any other electronic storage known in the art.

[0066] The network controller 303, for example an Intel Ethernet PRO network interface card from Intel Corporation, provides connections between various elements of the PET imaging apparatus. The network controller 303 also provides connections to, for example, an external network. The external network may be a public network such as the Internet, or a private network such as a Local Area Network (LAN) or a Wide Area Network (WAN), or any combination thereof, including a Public Switched Telephone Network (PSTN) or an Integrated Services Digital Network (ISDN) sub-network. For example, the external network may be a wired network such as an Ethernet network, or a wireless network such as a cellular network including Enhanced Data Rates for GSM Evolution (EDGE), 3G and 4G wireless cellular systems. For example, the wireless network may be WiFi, Bluetooth, or other known wireless communication formats.

[0067] The RTE method will be described in more detail below. FIG. 10 shows a scattering process in which gamma rays are emitted from a point shown as a gamma ray source, and shows three processes until the gamma rays of the gamma ray source reach the same detector among multiple detectors arranged in a ring shape. Here, the three processes are (i) a primary beam, (ii) first scattering, and (iii) multiple scattering. First, a primary flux is sent through the object OBJ while being attenuated, and is detected by a predetermined detector in the array of detectors. Next, when the gamma rays are scattered once before reaching the detector, first scattering (described as single scattering) occurs. Next, when the gamma rays are scattered multiple times before reaching the detector, multiple scattering (also called high order scattering) occurs. In some embodiments, the RTE method can be performed using three terms corresponding to the primary beam, first scattering, and multiple scattering.

[0068] 11 shows a flow diagram of process 410 for simulating scattering using the RTE three-step integral formulation. Here, a non-limiting embodiment is described in which process 410 is performed using a spherical harmonic expansion of the integral to accelerate the calculation of the integral, but process 410 is not limited to being performed using a spherical harmonic expansion. In some examples, for X-ray CT, the RTE scattering calculation using process 410 can calculate both Compton scattering and Rayleigh scattering, and for PET, the RTE scattering calculation using process 410 can calculate only Compton scattering.

[0069] The method described herein solves the RTE described above by splitting it into three simpler integration problems to obtain accurate, yet uncomplicated, scattering simulations for scattering estimation. This is achieved by first rewriting the RTE as the following integral equation:

number

[0070] As shown in FIG. 10, the flux is grouped into three components: (i) primary flux, (ii) first order scattered flux, and (iii) multi-scattered flux. Process 410 is configured to solve the integral equation of the RTE in steps corresponding to each of these three flux components. The RTE can be split into at least two components, i.e., scattered flux and primary flux. Thus, as mentioned above, given the primary flux ψ0(r'(with →),E',Ω'(with ^), the scattered flux ψ can be solved by the integral equation S (r,E,Ω(^)) can be calculated.

number

[0071] The integral equation for the scattered flux, ψ S (r(→),E,Ω(^)) is further divided into a first scattering component ψ1(r(→),E,l,m) and a multi-scattering component ψ S k It can be divided into (r(with →), E, ​​l, m).

[0072] First, the first order scattering flux is calculated in step 1120 of process 410. The first order scattering flux can be calculated by the discretized integral formula given by the following equation:

number

number

[0073] Different materials, such as bone and water, have different scattering cross sections due to different angular dependences depending on their atomic constituents. The magnitude of the scattering cross section may also depend on the density of the material components at point r'. Thus, by decomposing the attenuation image 405 into material components and determining the density of the various material components as a function of position, the scattering cross section as a function of position can be determined.

[0074] In one non-limiting example, material decomposition can be performed using dual-energy CT, and another option for material decomposition is to use threshold and region growing methods that infer regions corresponding to material components by voxel attenuation (e.g., voxels with attenuation values ​​above a certain threshold in Hounsfield units are determined to be bone).

[0075] Also, in step 1130 of process 410, higher order scattering terms (i.e., multiple scattering) are calculated. The first order scattering flux includes only X-ray photons scattered once, while the kth order scattering term includes X-ray photons scattered k times. After calculating the first order scattering flux, the flux of multiple scattering can be calculated using the discretized integral spherical harmonic formula as follows:

number

number

number

[0076] Also, in step 1140 of process 410, a stopping criterion is evaluated to determine whether the approximation up to order k is sufficiently accurate, i.e., whether the scattering calculated by the iterative multi-scattering calculation described above when calculated up to order k satisfies a predefined stopping criterion, e.g., related to the accuracy requirements of the scattering correction for a given medical application. In some embodiments, the stopping criterion may be satisfied if the following formula is true:

number

[0077] Also, the flux at the detector is determined in step 1150 of process 410. After calculating the multi-scattering flux, the above results can be combined to calculate the intensity of the scattered flux at the detector 540.

[0078] The training of the DL network is described in more detail below (e.g., process 460), where the target data 453 is a full singles spectrum, which can be calculated, for example, using the MC method or the RTE method, and the input data is a pair of emission and attenuation images corresponding to the full singles spectrum of the target data 453.

[0079] 12 shows a flow diagram of one embodiment of DL training process 460. In process 460, the input data 457 and the target data 453 are used as training data to train DL network 470, which outputs trained DL network 470 in step 1240. In offline DL training process 460, a training data set for training DL network 470 is used, in which each pair of emission and attenuation images in the input data 457 is combined with a corresponding full singles spectrum of the target data 453, to train DL network 470 to generate a full singles spectrum that approximates the full singles spectrum of the target data 453.

[0080] In process 460, a set of training data is obtained and the DL network 470 is iteratively updated to reduce the error (e.g., the value generated by the loss function). The DL network infers the mapping implied by the training data, and the cost function generates an error value related to the discrepancy between the target data 453 and the results generated by applying the DL network 470 to the input data 457. For example, in some implementations, the cost function may use a mean-squared error to minimize the average value of the squared error. In the case of a Multilayer Perceptrons (MLP) neural network, the network may be trained by minimizing a mean-squared error-based cost function using (stochastic) gradient descent using an error backpropagation algorithm.

[0081] Also, in step 1210 of the process 460, an initial guess is made for the coefficients of the DL network 470. The initial guess is made based on any one of LeCun initialization, Xavier initialization, and Kaiming initialization.

[0082] Additionally, steps 1210 through 1240 of process 460 are a non-limiting example of an optimization method for training DL network 470.

[0083] An error is calculated (e.g., using a loss or cost function) to indicate a measure of the difference (e.g., a distance measure) between the target data 453 and the output from the network generated by applying the input data 457 to the current version of the DL network 470. The error may be calculated using any known cost function, including the cost functions described above, or a distance measure between image data. Additionally, in some embodiments, the error / loss function may be calculated using one or more of a hinge loss and a cross-entropy loss. In some embodiments, the loss function may be a measure of the difference between the target data and the result of applying the input data to the DL network 470. p It may be a norm. p By setting different values ​​of “p” in the norm, the difference in noise can be emphasized.

[0084] In some embodiments, the DL network 470 is trained using backpropagation. Backpropagation can be used to train neural networks and is used in conjunction with gradient descent optimization. During the forward pass, the algorithm calculates predictions for the network based on the current parameters θ. These predictions are input to a loss function and compared to the corresponding ground truth labels (i.e., the target data 453). During the backward pass, the model calculates the gradient of the loss function with respect to the current parameters, which are then updated by making a predefined size step in the direction of the minimized loss (e.g., accelerated methods such as Nesterov's momentum method and various adaptive methods allow the size of the step to be selected to converge faster and optimize the loss function).

[0085] The optimization method for performing backpropagation can use one or more of gradient descent, batch gradient descent, stochastic gradient descent, and mini-batch stochastic gradient descent. Forward and backward passes can be performed incrementally through each layer of the network. In the forward pass, execution begins by feeding inputs through the first layer, thereby creating output activations for the subsequent layers. This process is repeated until the loss function of the final layer is reached. During the backward pass, the final layer computes gradients with respect to its learnable parameters (if any) and also with respect to its inputs, which act as upstream derivatives of the previous layer. This process is repeated until the input layer is reached.

[0086] Returning to Figure 12, in step 1220 of process 460, a function of the changes in the network can be calculated to determine a change in error (e.g., an error gradient) that can be used to select the direction and step size of subsequent changes to the weights / coefficients of DL network 470. Calculating the error gradient in this manner is consistent with some embodiments of gradient descent optimization methods. In some other embodiments, this step can be omitted and / or replaced with another step according to a different algorithm (e.g., a non-gradient descent optimization algorithm such as simulated annealing or a genetic algorithm), as will be appreciated by those skilled in the art.

[0087] Also in step 1220 of process 460, a new set of coefficients for DL ​​network 470 is determined. For example, the changes calculated in step 1220 are used to update the weights / coefficients, similar to a gradient descent optimization method or an over-allocation acceleration method.

[0088] Also, in step 1230 of process 460, the updated weights / coefficients of DL network 470 are used to calculate new error values.

[0089] Also, in step 1240, a predefined stopping criterion is used to determine whether the training of the network is complete. For example, the predefined stopping criterion may evaluate whether the new error and / or the total number of iterations performed exceeds a predefined value. For example, the stopping criterion may be met if the new error is below a predefined threshold or if a maximum number of iterations has been reached. If the criterion is not met, the training process performed in process 460 continues by going back to the start of the iterative loop by repeating step 1220 with new weights and coefficients (the iterative loop includes steps 1220, 1230, 1240). If the criterion is satisfied, the training process performed in process 460 is complete.

[0090] FIG. 13 shows examples of interconnections between layers of DL network 470. DL network 470 may include fully connected, convolutional, and pooling layers, all of which are described below. In some preferred embodiments of DL network 470, convolutional layers are placed near the input layer, while fully connected layers performing high level inference are placed further up the architecture towards the loss function. Pooling layers are inserted after the convolutions and may provide a reduction that lowers the spatial extent of the filters and therefore the amount of learnable parameters. Activation functions are also incorporated in the various layers to introduce non-linearity and allow the network to make complex predictive relationships. The activation functions may be saturating activation functions (e.g., sigmoid or hyperbolic tangent functions) or regularized activation functions (e.g., Rectified Linear Units (ReLUs) as applied in the first and second examples above). As also illustrated in the first and second examples above, layers of DL network 470 may also incorporate batch regularization.

[0091] FIG. 13 shows an example of a general artificial neural network (ANN) with N inputs, K hidden layers, and three outputs. Each layer is composed of nodes (also called neurons), which perform a weighted sum of the inputs and compare the result of the weighted sum with a threshold to generate an output. Each ANN forms a class of functions whose members are obtained by varying architectural details such as the respective thresholds, connection weights, or the number of nodes and / or their connectivity. The nodes of an ANN may also be called neurons (or neuron nodes), which may have interconnections between different layers of the ANN system. Synapses (i.e., connections between neurons) store values ​​called "weights" (interchangeably called "coefficients" or "weighting coefficients") that manipulate the data of a computation. The output of an ANN depends on three types of parameters: (i) the interconnection pattern between the different layers of neurons, (ii) a learning process that updates the interconnection weights, and (iii) an activation function that converts the weighted inputs of a neuron into an output activation.

[0092] Mathematically, a function m(x) of a network of neurons is a function of other functions n i (x), and other functions may be defined as components of still other functions. This can be conveniently represented as a network structure, as shown in Figure 13, with arrows representing the dependencies between the variables. For example, an ANN may use a nonlinear weighted sum, where m(x)=K(Σ i w i n i (x)).

[0093] In Figure 13, the circles around the threshold function represent neurons (i.e., nodes). In the non-limiting example shown in Figure 13, the inputs are depicted as circles around linear functions, and the arrows represent directed connections between neurons.

[0094] In the above-mentioned embodiment, an example of acquiring an emission image from PET data, acquiring an attenuation image from X-ray CT data, and an example of acquiring an emission image and an attenuation image from PET data have been described, but the embodiment is not limited thereto. For example, PET data and MRI (Magnetic Resonance Imaging) data may be acquired, an emission image may be acquired by reconstructing the PET image from the PET data, and an attenuation image may be acquired from the MRI data. For example, the MRI data may be converted into data equivalent to X-ray CT data, and the attenuation image may be acquired using the data in a manner similar to the method of acquiring an attenuation image from X-ray CT data.

[0095] In the above-described embodiment, an example of estimating a full singles spectrum has been described, but the estimated singles spectrum does not necessarily have to be a complete spectrum of singles counts. For example, a singles spectrum including counts filtered using an energy bandpass filter and a portion of other counts may be estimated. In other words, the estimated singles spectrum may include counts other than those filtered using an energy bandpass filter.

[0096] In the above-mentioned embodiment, the technology disclosed in the present application is applied to a PET device, but the embodiment is not limited thereto. For example, the technology disclosed in the present application can be similarly applied to various types of radiation diagnostic devices such as a SPECT (Single Photon Emission Computed Tomography) device, an X-ray CT device, and an X-ray diagnostic device.

[0097] At least one of the embodiments described above provides improved dead time correction of measured radiation.

[0098] The above disclosure also encompasses the following embodiments.

[0099] (1) An apparatus comprising a circuit configured to: (i) obtain radiation data indicative of an intensity of radiation detected at a plurality of detector elements, the radiation data including counts, filtered with an energy bandpass filter to generate filtered data, the energy range of the energy bandpass filter including an energy of the radiation if the radiation is not scattered, (ii) estimate a singles spectrum of the radiation detected at a plurality of detector elements based on the radiation data, the singles spectrum estimated using a scattering model, and (iii) perform a dead-time correction on the radiation data based on the estimated singles spectrum, correcting the radiation data for missed detections resulting from dead-time of each detector element of the plurality of detector elements to generate corrected radiation data.

[0100] (2) The apparatus described in (1), wherein the radiation detected by the plurality of detector elements includes gamma rays, and the circuitry is further configured to acquire radiation data by performing a Positron Emission Tomography (PET) scan that generates filtered data including energies in an energy range of 511 keV, and to reconstruct a PET image from the corrected radiation data.

[0101] (3) The apparatus described in (2), wherein the circuitry is further configured to acquire an emission image showing the density of gamma ray emitters as a function of position within a subject being imaged, acquire an attenuation image showing attenuation as a function of position within the subject, determine a scattering cross section of the gamma rays as a function of position within the subject based on the attenuation image, and estimate a singles spectrum from the emission image and the determined scattering cross section.

[0102] (4) The apparatus described in (3), wherein the radiation data includes X-ray Computed Tomography (CT) data and PET data, and the circuitry is further configured to (i) obtain the emission image by reconstructing a PET image from the PET data, and (ii) obtain the attenuation image by reconstructing a CT image from the X-ray CT data.

[0103] (5) The apparatus described in (3), wherein the radiation data includes PET data, and the circuitry is further configured to acquire the emission image and the attenuation image using a joint estimation PET method.

[0104] (6) The apparatus according to any one of (1) to (5), wherein the circuit is further configured to estimate a singles spectrum using a Monte Carlo method as a scattering model.

[0105] (7) An apparatus according to any of (1) to (6), wherein the circuit estimates the singles spectrum using an analytical equation as the scattering model, the analytical equation representing first order scattering that is scaled to match the tail of the radiation data.

[0106] (8) An apparatus according to any one of (1) to (7), wherein the circuit is further configured to estimate the singles spectrum using a Radiative Transfer Equation (RTE) method as the scattering model.

[0107] (9) An apparatus according to any one of (1) to (8), wherein, for performing the RTE method, the circuit is further configured to: integrate the attenuation of the primary gamma ray flux from an emission position of a gamma ray emitter to a scattering position where the gamma ray is scattered; and determine a scattering flux at the scattering position by integrating a scattering contribution from the emission position in the emission image, the scattering flux being a function of the radiation energy and solid angle of the gamma ray after scattering.

[0108] (10) The apparatus described in (9), wherein the calculation of the RTE is accelerated by the circuitry further configured to calculate the scattering contributions from emission positions across the emission image using an approximation in which the scattering contributions are approximated using a predetermined number of irreducible fractions of a spherical harmonic expansion.

[0109] (11) The apparatus according to any one of (1) to (8), wherein the circuit is further configured to estimate a singles spectrum using a neural network as the scattering model.

[0110] (12) The apparatus of (11), wherein the neural network is trained using a training data set including input data and target data, the input data comprising pairs of attenuation and emission images, the target data comprising singles spectra each corresponding to a respective pair of attenuation and emission data of the input data, and the neural network is trained to generate an output that substantially matches a singles spectrum from the target data when the corresponding pairs of attenuation and emission images are applied as inputs to the neural network.

[0111] (13) The apparatus of (11), wherein the circuitry is further configured to train the neural network by iteratively adjusting weighting coefficients of the neural network to optimize a value of a loss function, the loss function measuring the discrepancy between each singles spectrum of target data and output data from the neural network when a pair of attenuation and emission images corresponding to each singles spectrum is applied as an input to the neural network.

[0112] (14) A method comprising: (i) obtaining radiation data indicative of an intensity of radiation detected by a plurality of detector elements; (ii) estimating a singles spectrum of the radiation detected by a plurality of detector elements based on the radiation data; and (iii) performing a dead-time correction on the radiation data based on the estimated singles spectrum, wherein the radiation data includes counts filtered using an energy band-pass filter, an energy range of the energy band-pass filter including an energy of the radiation when the radiation is not scattered, the singles spectrum being estimated using a scattering model, and the dead-time correction corrects the radiation data for missed detections resulting from dead-time of each detector element of the plurality of detector elements to generate corrected radiation data.

[0113] (15) The method according to (14), wherein the radiation detected by the plurality of detector elements includes gamma rays, and the step of acquiring the radiation data further includes performing a Positron Emission Tomography (PET) scan that generates counts that are filtered using an energy range that includes an energy of 511 keV, and the method further includes reconstructing a PET image from the corrected radiation data.

[0114] (16) The method according to (15) further comprises: (i) acquiring an emission image showing the density of a gamma-ray emitter as a function of position within a subject being imaged, and acquiring an attenuation image showing attenuation as a function of position within the subject; (ii) determining a scattering cross section of the gamma rays as a function of position within the subject based on the attenuation image; and (iii) estimating a singles spectrum from the emission image and the determined scattering cross section.

[0115] (17) The method according to (16), wherein the radiation data includes X-ray Computed Tomography (CT) data and PET data, and the method further includes (i) obtaining the emission image by reconstructing a PET image from the PET data, and (ii) obtaining the attenuation image by reconstructing a CT image from the X-ray CT data.

[0116] (18) The method according to (16), wherein the radiation data includes PET data, and the method further includes acquiring the emission image and the attenuation image using a joint estimation PET method.

[0117] (19) A method according to any one of (14) to (18), wherein the step of estimating the singles spectrum is performed using one of a Monte Carlo method, an analytical equation, a radiation transport equation, or a neural network as the scattering model, and the analytical equation represents first order singles scattering that is scaled to match the tail of the radiation data.

[0118] (20) A method according to any one of (14) to (18), wherein the step of estimating the singles spectrum uses a Radiative Transfer Equation (RTE) as the scattering model, and the method further includes: (i) integrating the attenuation of the primary gamma ray flux from an emission position of a gamma ray emitter to a scattering position where the gamma ray is scattered; and (ii) determining a scattering flux at the scattering position by integrating a scattering contribution from the emission position in the emission image, wherein the scattering flux is a function of the radiation energy and solid angle of the gamma ray after scattering.

[0119] (21) The method according to (20), wherein calculating the RTE is accelerated by a circuit further configured to calculate the scattering contributions from emission positions across the emission image using an approximation in which the scattering contributions are approximated using a predetermined number of irreducible fractions of a spherical harmonic expansion.

[0120] (22) A method according to any of (14) to (18), wherein the step of estimating the singles spectrum uses a neural network as the scattering model, the neural network being trained using a training data set including input data and target data, the input data comprising pairs of attenuation and emission images, the target data comprising singles spectra each corresponding to a respective pair of attenuation and emission data of the input data, and the neural network being trained to generate an output that substantially matches the singles spectrum from the target data when the corresponding pairs of attenuation and emission images are applied as inputs to the neural network.

[0121] (23) A method according to any of (14) to (18), wherein the step of estimating the singles spectrum uses a neural network as the scattering model, and the method further includes training the neural network by iteratively adjusting weighting coefficients of the neural network to optimize a value of a loss function, the loss function measuring the discrepancy between each singles spectrum of the target data and output data from the neural network when a pair of the attenuation image and emission image corresponding to each singles spectrum is applied as input to the neural network.

[0122] (24) A non-transitory computer-readable storage medium comprising executable instructions that, when executed by a circuit, cause the circuit to perform any one of the methods recited in (14) to (23).

[0123] Although several embodiments have been described, these embodiments are presented by way of example only and are not intended to limit the teachings of the present disclosure. Indeed, the novel methods, apparatus, and systems described herein may be embodied in a variety of other forms. Furthermore, various omissions, substitutions, and changes in the form of the methods, apparatus, and systems described herein may be made without departing from the spirit of the present disclosure.

[0124] Obviously, numerous modifications and variations of the present invention are possible in light of the above teachings. It is therefore to be understood that, within the scope of the appended claims, the invention may be practiced otherwise than as specifically described herein. [Explanation of symbols]

[0125] 800 PET equipment 870 Processor

Claims

1. obtaining an emission image showing the density of the gamma ray emitter as a function of position within the imaged object and an attenuation image showing attenuation as a function of position within the object based on radiation data including counts showing the intensity of radiation detected by the plurality of detector elements and filtered using an energy band pass filter with an energy range that includes the energy of the radiation when unscattered; estimating a singles spectrum of the radiation using a scattering model based on the emission image and the attenuation image, the singles spectrum including counts other than those filtered using the energy bandpass filter; performing dead-time correction to generate corrected radiation data by correcting the radiation data for detection misses resulting from dead-times of each of the plurality of detection elements based on the singles spectrum; A radiological diagnostic apparatus comprising a processing unit.

2. The radiation is gamma radiation, The processing unit includes: acquiring said radiation data by performing a Positron Emission Tomography (PET) scan that generates filtered counts using an energy range that includes an energy of 511 keV; reconstructing a PET image from the corrected radiation data; The radiological diagnostic apparatus according to claim 1 .

3. The processing unit includes: determining a scattering cross section of the gamma rays as a function of position within the subject based on the attenuation image, and estimating the singles spectrum from the emission image and the scattering cross section. The radiological diagnostic apparatus according to claim 2 .

4. Obtaining radiation data indicating the intensity of gamma rays detected by a plurality of detector elements by performing a Positron Emission Tomography (PET) scan that produces counts filtered using an energy range that includes an energy of 511 keV, the counts being filtered using an energy bandpass filter in an energy range that includes the energy of the gamma rays when unscattered; estimating a singles spectrum of the gamma rays using a scattering model based on the radiation data, the singles spectrum including counts other than those filtered using the energy bandpass filter; performing a dead time correction based on the singles spectrum to generate corrected radiation data by correcting the radiation data for detection errors resulting from dead times of each of the plurality of detection elements; Reconstructing a PET image from the corrected radiation data. A processing unit is provided, The processing unit includes: obtaining emission images showing the density of gamma ray emitters as a function of position within an imaged object, and attenuation images showing attenuation as a function of position within said object; determining a scattering cross section of the gamma rays as a function of position within the subject based on the attenuation image, and estimating the singles spectrum from the emission image and the scattering cross section. Radiological diagnostic equipment.

5. The processing unit acquires PET data and X-ray computed tomography (CT) data, acquires the emission image by reconstructing a PET image from the PET data, and acquires the attenuation image by reconstructing a CT image from the X-ray CT data. The radiological diagnostic apparatus according to claim 3 or 4.

6. The processor acquires PET data and acquires the emission image and the attenuation image from the PET data using a joint-estimation PET method. The radiological diagnostic apparatus according to claim 3 or 4.

7. The processing unit acquires PET data and MRI (Magnetic Resonance Imaging) data, acquires the emission image by reconstructing a PET image from the PET data, and acquires the attenuation image from the MRI data. The radiological diagnostic apparatus according to claim 3 or 4.

8. The processing unit estimates the singles spectrum using a Monte Carlo method as the scattering model. The radiation diagnostic apparatus according to any one of claims 1 to 6.

9. The processor estimates the singles spectrum using an analytical equation representing first order scattering scaled to match the tails of the radiation data as the scattering model. The radiation diagnostic apparatus according to any one of claims 1 to 6.

10. Acquiring radiation data including counts indicating an intensity of radiation detected by a plurality of detector elements and filtered using an energy bandpass filter having an energy range including the energy of the radiation when not scattered; estimating a singles spectrum of the radiation based on the radiation data using a scattering model, the singles spectrum including counts other than those filtered using the energy bandpass filter; performing dead-time correction to generate corrected radiation data by correcting the radiation data for detection misses resulting from dead-times of each of the plurality of detection elements based on the singles spectrum; A processing unit is provided, The processor estimates the singles spectrum using, as the scattering model, an analytical equation describing first order scattering scaled to match the tails of the radiation data. Radiological diagnostic equipment.

11. The processing unit estimates the singles spectrum using a Radiative Transfer Equation (RTE) method as the scattering model. The radiation diagnostic apparatus according to any one of claims 1 to 6.

12. Acquiring radiation data including counts indicating an intensity of radiation detected by a plurality of detector elements and filtered using an energy bandpass filter having an energy range including the energy of the radiation when not scattered; estimating a singles spectrum of the radiation based on the radiation data using a scattering model, the singles spectrum including counts other than those filtered using the energy bandpass filter; performing dead-time correction to generate corrected radiation data by correcting the radiation data for detection misses resulting from dead-times of each of the plurality of detection elements based on the singles spectrum; A processing unit is provided, The processing unit estimates the singles spectrum using a Radiative Transfer Equation (RTE) method as the scattering model. Radiological diagnostic equipment.

13. The processing unit performs the RTE method by integrating the attenuation of the primary gamma ray flux from an emission location of the gamma ray emitter to a scattering location where the gamma ray is scattered, and determining a scattering flux, which is a function of the radiation energy and solid angle of the gamma ray after scattering at the scattering location, by integrating the scattering contribution from the emission location. The radiological diagnostic apparatus according to claim 11.

14. The processing unit accelerates the calculation of the RTE method by calculating the scattering contribution from the emission position using an approximation calculation that approximates the scattering contribution using a predetermined number of irreducible fractions of a spherical harmonic expansion. The radiological diagnostic apparatus according to claim 13.

15. The processing unit estimates the singles spectrum using a neural network as the scattering model. The radiation diagnostic apparatus according to any one of claims 1 to 6.

16. acquiring radiation data indicating an intensity of radiation detected by a plurality of detector elements, the radiation data including counts filtered using an energy bandpass filter having an energy range including the energy of the radiation when not scattered; estimating a singles spectrum of the radiation based on the radiation data using a scattering model, the singles spectrum including counts other than those filtered using the energy bandpass filter; performing dead-time correction to generate corrected radiation data by correcting the radiation data for detection misses resulting from dead-times of each of the plurality of detection elements based on the singles spectrum; A processing unit is provided, The processing unit estimates the singles spectrum using a neural network as the scattering model. Radiological diagnostic equipment.

17. the processing unit uses a training data set including input data including pairs of attenuation and emission images and target data including singles spectra corresponding to the pairs of attenuation and emission images included in the input data to train the neural network to generate an output that substantially matches the singles spectra of the target data when the corresponding pairs of attenuation and emission images are applied as inputs to the neural network. The radiological diagnostic apparatus according to claim 15 or 16.

18. the processing unit trains the neural network by iteratively adjusting weighting coefficients of the neural network to optimize a value of a loss function that measures a discrepancy between an output from the neural network and the attenuation and emission image pair corresponding to a singles spectrum of target data when the pair is applied as an input to the neural network. The radiological diagnostic apparatus according to claim 17.

19. A method for controlling a radiological diagnostic apparatus having a processing unit, comprising: The processor acquires an emission image showing the density of gamma ray emitters as a function of position within the object being imaged and an attenuation image showing attenuation as a function of position within the object based on radiation data including counts showing the intensity of radiation detected by the plurality of detector elements and filtered using an energy band pass filter with an energy range including the energy of the radiation when not scattered; the processor estimating, based on the emission image and the attenuation image, a singles spectrum of the radiation using a scattering model, the singlets spectrum including counts other than those filtered using the energy bandpass filter; a step of performing dead time correction by the processing unit to generate corrected radiation data by correcting the radiation data for detection errors caused by dead times of the respective detection elements based on the singles spectrum; A method comprising:

20. A method for controlling a radiological diagnostic apparatus having a processing unit, comprising: acquiring radiation data indicating the intensity of gamma rays detected by the plurality of detector elements, the radiation data including counts filtered using an energy bandpass filter having an energy range including the energy of the gamma rays when not scattered, by performing a Positron Emission Tomography (PET) scan, the PET scan generating counts filtered using an energy range including an energy of 511 keV; The processing unit estimates a singles spectrum of the gamma ray including counts other than those filtered using the energy bandpass filter based on the radiation data using a scattering model; performing dead time correction by the processing unit to generate corrected radiation data by correcting the radiation data for detection errors resulting from dead times of each of the plurality of detection elements based on the singles spectrum; the processor reconstructing a PET image from the corrected radiation data; Including, The processing unit includes: obtaining emission images showing the density of gamma ray emitters as a function of position within an imaged object, and attenuation images showing attenuation as a function of position within said object; determining a scattering cross section of the gamma rays as a function of position within the subject based on the attenuation image, and estimating the singles spectrum from the emission image and the scattering cross section. method.

21. A method for controlling a radiological diagnostic apparatus having a processing unit, comprising: acquiring radiation data including counts indicative of an intensity of radiation detected by a plurality of detector elements and filtered using an energy bandpass filter having an energy range that includes the energy of the radiation when not scattered; the processor estimating, based on the radiation data, a singles spectrum of the radiation using a scattering model, the singles spectrum including counts other than those filtered using the energy bandpass filter; performing dead time correction by the processing unit to generate corrected radiation data by correcting the radiation data for detection errors resulting from dead times of each of the plurality of detection elements based on the singles spectrum; Including, The processor estimates the singles spectrum using, as the scattering model, an analytical equation describing first order scattering scaled to match the tails of the radiation data. method.

22. A method for controlling a radiological diagnostic apparatus having a processing unit, comprising: acquiring radiation data including counts indicative of an intensity of radiation detected by a plurality of detector elements and filtered using an energy bandpass filter having an energy range that includes the energy of the radiation when not scattered; the processor estimating, based on the radiation data, a singles spectrum of the radiation using a scattering model, the singles spectrum including counts other than those filtered using the energy bandpass filter; performing dead time correction by the processing unit to generate corrected radiation data by correcting the radiation data for detection errors resulting from dead times of each of the plurality of detection elements based on the singles spectrum; Including, The processing unit estimates the singles spectrum using a Radiative Transfer Equation (RTE) method as the scattering model. method.

23. A method for controlling a radiological diagnostic apparatus having a processing unit, comprising: acquiring radiation data including counts indicative of an intensity of radiation detected by a plurality of detector elements and filtered using an energy bandpass filter having an energy range that includes the energy of the radiation when not scattered; the processor estimating, based on the radiation data, a singles spectrum of the radiation using a scattering model, the singles spectrum including counts other than those filtered using the energy bandpass filter; performing dead time correction by the processing unit to generate corrected radiation data by correcting the radiation data for detection errors resulting from dead times of each of the plurality of detection elements based on the singles spectrum; Including, The processing unit estimates the singles spectrum using a neural network as the scattering model. method.

24. A program causing a computer to execute the method according to any one of claims 19 to 23.

Images