Tomographic imaging apparatus, tomographic imaging method, and TOF-PET apparatus

Abstract

To provide a tomographic image creation device that creates a tomographic image of a subject using a DNN on the basis of information on a plurality of γ-ray pair simultaneous count events collected by a PET detector, with which device it is possible to easily learn the DNN.SOLUTION: A tomographic image creation device 10 comprises a time difference calculation unit 12, a signal waveform processing unit 13, an error inference unit 14, a γ-ray pair occurrence position calculation unit 15, and an image creation unit 16. The time difference calculation unit 12 determines, for each of a number of γ-ray pair simultaneous count events, a time difference tled in timing at which the respective values of a first signal and a second signal outputted from a signal waveform acquisition unit 11 reach a threshold. The signal waveform processing unit 13 relatively shifts the waveform of the first signal or the waveform of the second signal by the time difference tled in a direction to approach each other in a time-axis direction. The error inference unit 14 infers, by a DNN, an error terr that is included in the time difference tled on the basis of the respective waveforms of the first signal and second signal after the shift.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a tomographic image creation apparatus, a tomographic image creation method, and a TOF-PET apparatus.

Background Art

[0002] A PET (Positron Emission Tomography) apparatus includes a PET detector including a large number of radiation detectors provided surrounding a measurement space, and a tomographic image creation apparatus that creates a tomographic image of a subject based on information on a large number of γ-ray pair coincidence events collected for the subject by this PET detector. A subject administered with a drug labeled with a positron-emitting radionuclide is placed in the measurement space of the PET detector. When a positron is emitted from the positron-emitting radionuclide in the body of the subject, two γ-ray photons with an energy of 511 KeV are generated by the annihilation of the positron and the electron. These two γ-ray photons (γ-ray pair) fly in opposite directions to each other and are simultaneously counted by any two radiation detectors of the PET detector. Then, by performing a required image reconstruction process based on the information on the large number of γ-ray pair coincidence events collected by the tomographic image creation apparatus, an image representing the distribution of γ-ray pair generation positions (that is, a tomographic image of the subject) can be created.

[0003] Among PET apparatuses, in a TOF-PET (Time-of-Flight PET) apparatus, for each γ-ray pair coincidence event, based on the time difference between the detection timings of the two radiation detectors that simultaneously counted the γ-ray pair, the γ-ray pair generation position on the coincidence line connecting these two radiation detectors can be detected. Then, by detecting the γ-ray pair generation positions for a large number of γ-ray pair coincidence events, an image representing the distribution of γ-ray pair generation positions (that is, a tomographic image of the subject) can be created. Hereinafter, such a technique is referred to as "Comparative Example 1". In a TOF-PET apparatus, in order to create a tomographic image with high spatial resolution, it is desirable to obtain the time difference between the detection timings of the two radiation detectors that simultaneously counted the γ-ray pair with high time resolution.

[0004] In the technique described in Non-Patent Document 1 (hereinafter referred to as "Comparative Example 2"), in a TOF-PET apparatus, waveforms of a first signal and a second signal respectively output from two radiation detectors that simultaneously count a pair of γ-rays are input into a convolutional neural network (CNN), which is a type of deep neural network (DNN). The CNN is used to estimate the time difference between the detection timings of the two radiation detectors that simultaneously count the pair of γ-rays. Compared with Comparative Example 1, in Comparative Example 2, it is said that the time difference between the detection timings of the two radiation detectors that simultaneously count the pair of γ-rays can be obtained with high time resolution.

Prior Art Documents

Non-Patent Documents

[0005]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] Compared with Comparative Example 1, in Comparative Example 2, since the time difference between the detection timings of the two radiation detectors that simultaneously count the pair of γ-rays can be obtained with high time resolution, it is expected that a tomographic image with high spatial resolution can be created. However, the inventors have found that in Comparative Example 2, there are problems that the amount of data required to train the CNN is huge, so the training of the CNN is not easy, and the created tomographic image is distorted.

[0007] The present invention has been made to solve the above problems, and aims to provide a device and method for creating a tomographic image of a subject using a DNN based on information on multiple gamma-ray pair coincidence events collected by a PET detector, which can easily train the DNN and create a tomographic image with little distortion. It also aims to provide a TOF-PET device equipped with such a tomographic image creation device and a PET detector. [Means for solving the problem]

[0008] The tomographic image creation device of the present invention is a device for creating a tomographic image of a subject based on information on a plurality of gamma-ray pair coincidence events collected from the subject placed in a measurement space of a PET detector including a plurality of radiation detectors, and the device (1) calculates a time difference t between the timing at which the values of a first signal and a second signal output from two radiation detectors that have coincidentally counted a gamma-ray pair among the plurality of radiation detectors reach a threshold value for each of the plurality of gamma-ray pair coincidence events. led (2) a time difference calculation unit for calculating a time difference t led (3) a signal waveform processor that relatively shifts the first signal and the second signal by a time difference t based on the waveforms of the first signal and the second signal after the shift by the signal waveform processor, and led Error t err (4) the time difference t led and error t err and (5) an image creation unit that creates a tomographic image of the subject based on the gamma ray pair generation positions determined by the gamma ray pair generation position calculation unit for each of a plurality of gamma ray pair coincidence events.

[0009] The tomography image creation apparatus of the present invention preferably further includes a learning unit that learns a deep neural network. Based on information on a plurality of γ-ray pair coincidence events collected for a positron-emitting radionuclide placed in a measurement space, for each of the plurality of γ-ray pair coincidence events, the waveforms of the first signal and the second signal after shifting by the signal waveform processing unit are used as input data to the deep neural network, and the time difference t led calculated by the time difference calculation unit is used as teacher data to train the deep neural network, where the teacher data is the difference between the time difference t and the true time difference based on the position of the positron-emitting radionuclide.

[0010] The TOF-PET apparatus of the present invention includes a PET detector including a plurality of radiation detectors, and the above-described tomography image creation apparatus of the present invention that creates a tomographic image of a subject based on information on a plurality of γ-ray pair coincidence events collected for the subject placed in the measurement space of the PET detector.

[0011] The tomography image creation method of the present invention is a method for creating a tomographic image of a subject based on information on a plurality of γ-ray pair coincidence events collected for the subject placed in the measurement space of a PET detector including a plurality of radiation detectors, and includes: (1) for each of the plurality of γ-ray pair coincidence events, the time difference t between the timings at which the values of the first signal and the second signal output from the two radiation detectors that simultaneously counted the γ-ray pair among the plurality of radiation detectors reach a threshold value led is calculated in a time difference calculation step; (2) in a signal waveform processing step, the waveform of the first signal or the waveform of the second signal is relatively shifted by the time difference t led in a direction approaching each other in the time axis direction; (3) based on the waveforms of the first signal and the second signal after shifting in the signal waveform processing step, an error t led of the time difference t err is estimated in an error estimation step by a deep neural network; and (4) the time difference t led and the error t errBased on this, a γ-ray pair generation position calculation step for obtaining the γ-ray pair generation position on the coincidence counting line connecting two radiation detectors to each other, and (5) an image creation step for creating a tomographic image of the subject based on the γ-ray pair generation position obtained by the γ-ray pair generation position calculation step for each of a plurality of γ-ray pair coincidence counting events.

[0012] The tomographic image creation method of the present invention preferably further includes a learning step of training a deep neural network. This learning step is based on information of a plurality of γ-ray pair coincidence counting events collected for a positron-emitting radionuclide placed in the measurement space. For each of the plurality of γ-ray pair coincidence counting events, the waveforms of the first signal and the second signal after shifting by the signal waveform processing step are used as input data to the deep neural network, and the time difference t obtained by the time difference calculation step led and the difference from the true time difference based on the position of the positron-emitting radionuclide are used as teacher data to train the deep neural network.

Advantages of the Invention

[0013] According to the present invention, a tomographic image of a subject can be created using a DNN based on information of a plurality of γ-ray pair coincidence counting events collected by a PET detector, and the DNN can be easily trained to create a tomographic image with little distortion.

Brief Description of the Drawings

[0014]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

DETAILED DESCRIPTION OF THE INVENTION

[0015] Hereinafter, embodiments for implementing the present invention will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same elements are denoted by the same reference numerals, and duplicate descriptions are omitted. The present invention is not limited to these examples, but is defined by the claims, and is intended to include all modifications within the meaning and scope equivalent to the claims.

[0016] FIG. 1 is a diagram showing the configuration of a TOF-PET apparatus 1. The TOF-PET apparatus 1 includes a tomographic image creation apparatus 10 and a PET detector 20.

[0017] The PET detector 20 includes a large number of radiation detectors provided in a ring shape surrounding the measurement space where the subject 2 is placed. A subject 2 administered with a drug labeled with a positron-emitting radionuclide is placed in the measurement space of the PET detector 20. When a positron is emitted from the positron-emitting radionuclide in the body of the subject 2, two γ-ray photons with an energy of 511 KeV are generated by the annihilation of the positron and the electron. These two γ-ray photons (γ-ray pair) fly in opposite directions to each other and are simultaneously counted by any two of the plurality of radiation detectors 21, 22 of the PET detector 20. Each of the plurality of radiation detectors of the PET detector 20 outputs a pulse signal in response to a γ-ray detection event. In FIG. 1, the flight path of a certain γ-ray pair generated at a certain position (γ-ray pair generation position) in the body of the subject 2 is indicated by an arrow, and the two of the plurality of radiation detectors that detected the γ-ray pair are shown as the radiation detectors 21, 22.

[0018] The tomographic image creation apparatus 10 creates a tomographic image of the subject 2 based on information on a large number of γ-ray pair simultaneous counting events collected for the subject 2 placed in the measurement space of the PET detector 20. The tomographic image creation apparatus 10 includes a signal waveform acquisition unit 11, a time difference calculation unit 12, a signal waveform processing unit 13, an error estimation unit 14, a γ-ray pair generation position calculation unit 15, an image creation unit 16, and a learning unit 17.

[0019] The signal waveform acquisition unit 11 is connected to each of the plurality of radiation detectors of the PET detector 20 by signal lines, and inputs the pulse signals output from each of the plurality of radiation detectors in response to γ-ray detection events. In FIG. 1, the signal lines between the signal waveform acquisition unit 11 and two radiation detectors 21 and 22 that have detected a certain pair of γ-rays among the plurality of radiation detectors of the PET detector 20 are shown, and the signal lines between the other radiation detectors and the signal waveform acquisition unit 11 are not shown for simplicity of illustration.

[0020] Based on the pulse signals output from each of the plurality of radiation detectors of the PET detector 20 in response to γ-ray detection events, the signal waveform acquisition unit 11 detects γ-ray pair coincidence events by any two of the plurality of radiation detectors and identifies those two radiation detectors. Then, for each of a large number of γ-ray pair coincidence events, the signal waveform acquisition unit 11 outputs the waveforms of the pulse signals (first signal, second signal) output from each of the two radiation detectors that have simultaneously counted the γ-ray pair to the time difference calculation unit 12.

[0021] The time difference calculation unit 12 inputs the waveforms of the first signal and the second signal output from the signal waveform acquisition unit 11 for each of a large number of γ-ray pair coincidence events. Then, the time difference calculation unit 12 calculates the time difference t led of the timings at which the values of the first signal and the second signal, which are pulse signals, reach the threshold value.

[0022] FIG. 2 is a diagram for explaining the processing content of the time difference calculation unit 12. The time difference calculation unit 12 obtains the timing t1 at which the value of the first signal reaches the threshold value and the timing t2 at which the value of the second signal reaches the threshold value, and calculates the time difference t led between these timings t1 and t2. This processing is called LED (Lead Edge Discriminator) and is also performed in Comparative Example 1.

[0023] The signal waveform processing unit 13 moves the waveform of the first signal or the waveform of the second signal in a direction approaching each other in the time axis direction by the time difference t ledOnly shift relatively. The signal waveform processing unit 13 may shift either the waveform of the first signal or the waveform of the second signal closer to the other in the time axis direction, or may shift both the waveform of the first signal and the waveform of the second signal closer to each other in the time axis direction. Note that when the time difference t led is 0, there is no need to shift the waveforms of either the first signal or the second signal in the time axis direction.

[0024] FIG. 3 is a diagram showing the processing result of the signal waveform processing unit 13. The timings at which the values of the first signal and the second signal after being shifted by the signal waveform processing unit 13 reach the threshold value should be equal to each other. However, in reality, since the time difference t led obtained by the time difference calculation unit 12 may include an error, the two timings may not be equal to each other. Note that the error included in the time difference t led obtained by the time difference calculation unit 12 is considered to be independent of the γ-ray pair generation position.

[0025] Based on the waveforms (FIG. 3) of the first signal and the second signal after being shifted by the signal waveform processing unit 13, the error estimation unit 14 estimates the error t led included in the time difference t err obtained by the time difference calculation unit 12. When estimating this error t err , a DNN is used, and preferably, a CNN, which is a type of DNN, is used.

[0026] Based on the time difference t led obtained by the time difference calculation unit 12 and the error t err estimated by the error estimation unit 14, the γ-ray pair generation position calculation unit 15 obtains a more accurate time difference t est (=t led -t err ). Then, based on this time difference t est , the γ-ray pair generation position calculation unit 15 obtains the γ-ray pair generation position on the coincidence counting line connecting the two radiation detectors that counted the γ-ray pair simultaneously.

[0027] The image creation unit 16 creates a tomographic image of the subject 2 based on the gamma-ray pair generation positions obtained by the gamma-ray pair generation position calculation unit 15 for each of a large number of gamma-ray pair simultaneous counting events.

[0028] The learning unit 17 learns the DNN in the error estimation unit 14 based on information on a large number of gamma-ray pair simultaneous counting events collected for a positron-emitting radionuclide placed in the measurement space of the PET detector 20 instead of the subject 2. For each of a plurality of gamma-ray pair simultaneous counting events, the learning unit 17 uses the waveforms of the first signal and the second signal after shifting by the signal waveform processing unit 13 (Fig. 3) as input data to the DNN, and the time difference t obtained by the time difference calculation unit 12 led and the difference from the true time difference based on the position of the positron-emitting radionuclide as teacher data to learn the DNN.

[0029] The tomographic image creation method using such a tomographic image creation apparatus 10 includes a signal waveform acquisition step by the signal waveform acquisition unit 11, a time difference calculation step by the time difference calculation unit 12, a signal waveform processing step by the signal waveform processing unit 13, an error estimation step by the error estimation unit 14, a gamma-ray pair generation position calculation step by the gamma-ray pair generation position calculation unit 15, an image creation step by the image creation unit 16, and a learning step by the learning unit 17.

[0030] That is, in the signal waveform acquisition step, pulse signals output from each of a plurality of radiation detectors of the PET detector 20 in response to gamma-ray detection events are input. In the time difference calculation step, for each of a plurality of gamma-ray pair simultaneous counting events, the time difference t between the timings at which the values of the first signal and the second signal output from two radiation detectors that simultaneously counted the gamma-ray pair among the plurality of radiation detectors reach the threshold value led is obtained. In the signal waveform processing step, the waveform of the first signal or the waveform of the second signal is relatively shifted by the time difference t led in a direction approaching each other in the time axis direction.

[0031] In the error estimation step, based on the waveforms of the first signal and the second signal after the shift by the signal waveform processing step, the DNN estimates the time difference t led of the error t err . In the gamma-ray pair generation position calculation step, based on the time difference t led and the error t err , the gamma-ray pair generation position on the coincidence line connecting the two radiation detectors is obtained. In the image creation step, based on the gamma-ray pair generation positions obtained by the gamma-ray pair generation position calculation step for each of the plurality of gamma-ray pair coincidence events of the PET detector 20, a tomographic image of the subject 2 is created.

[0032] In the learning step, the DNN is trained based on the information of a plurality of gamma-ray pair coincidence events collected for the positron-emitting radionuclide placed in the measurement space of the PET detector 20. Note that if the DNN has been trained, the learning step and the learning unit 17 are not necessary. However, even if the DNN has been trained, if further training is performed to enable more accurate estimation, the learning step and the learning unit 17 may be provided.

[0033] Next, the results of the experiment conducted to confirm the effects of the present embodiment in comparison with the comparative example will be described. FIG. 4 is a diagram showing the configuration of the experimental system. In this experiment, positron-emitting radionuclides ( 22 Na) were sequentially placed at seven positions P1 to P7 spaced at 5 mm pitches on the line (corresponding to the coincidence line) connecting the two radiation detectors 21 and 22.

[0034] Each of the radiation detectors 21 and 22 was provided with a LYSO (Cerium Doped Lutetium Yttrium Orthosilicate) scintillator on the light-receiving surface of an MPPC (Multi-Pixel Photon Counter). The MPPC (registered trademark) has a quenching resistor connected to an avalanche photodiode operating in Geiger mode as one pixel, and a plurality of pixels are two-dimensionally arranged, enabling high-speed and high-sensitivity light detection. The size of the light-receiving surface of the MPPC was 3 mm × 3 mm. The size of the LYSO scintillator was 3 mm × 3 mm × 10 mm thick.

[0035] In Comparative Example 1, for each γ-ray pair simultaneous counting event, the time difference t obtained by the time difference calculation unit 12 was used to determine the γ-ray pair generation position. led based on this, the γ-ray pair generation position was determined.

[0036] Both Comparative Example 2A and Comparative Example 2B correspond to the technology (Comparative Example 2) described in Non-Patent Document 1 mentioned above, but they differ in the database used for training the CNN. In Comparative Examples 2A and 2B, for each γ-ray pair simultaneous counting event, based on the waveforms (Figure 2) of the first signal and the second signal respectively acquired by the signal waveform acquisition unit 11, the CNN was used to estimate the time difference between the two signals, and based on this estimated time difference, the γ-ray pair generation position was determined.

[0037] In Comparative Example 2A, when training the CNN, the waveforms (Figure 2) of the first signal and the second signal respectively acquired by the signal waveform acquisition unit 11 when a positron-emitting radionuclide was placed at each of the seven positions P1 to P7 were used as input data to the CNN, and the true time difference based on the position where the positron-emitting radionuclide was placed was used as teacher data.

[0038] In Comparative Example 2B, when training the CNN, the waveforms (Figure 2) of the first signal and the second signal respectively acquired by the signal waveform acquisition unit 11 when a positron-emitting radionuclide was placed at each of the six positions P1 to P4, P6, and P7 excluding position P5 were used as input data to the CNN, and the true time difference based on the position where the positron-emitting radionuclide was placed was used as teacher data.

[0039] In the embodiment, the γ-ray pair generation position was obtained by the tomographic image creation apparatus 10 or the tomographic image creation method of the present embodiment described above. In the embodiment, when training the CNN, the waveforms of the first signal and the second signal (FIG. 3) after being shifted by the signal waveform processing unit 13 when a positron-emitting radionuclide was placed only at the position P4 among the seven positions P1 to P7 were used as input data to the CNN, and the time difference t obtained by the time difference calculation unit 12 led and the difference from the true time difference based on the position P4 of the positron-emitting radionuclide were used as teacher data.

[0040] FIG. 5 is a graph showing the distribution of the γ-ray pair generation positions obtained in Comparative Example 1. FIG. 6 is a graph showing the distribution of the γ-ray pair generation positions obtained in Comparative Example 2A. FIG. 7 is a graph showing the distribution of the γ-ray pair generation positions obtained in Comparative Example 2B. FIG. 8 is a graph showing the distribution of the γ-ray pair generation positions obtained in the embodiment. FIGS. 5 to 8 show the shapes of the distributions of the γ-ray pair generation positions obtained when a positron-emitting radionuclide was placed at each of the seven positions P1 to P7.

[0041] From these figures, the following can be said about the shape of the distribution of the obtained γ-ray pair generation positions. In Comparative Example 1 (FIG. 5) and the embodiment (FIG. 8), the distributions of the obtained γ-ray pair generation positions are substantially symmetric about the peak position. In contrast, in Comparative Example 2A (FIG. 6), the distribution of the γ-ray pair generation positions obtained when a positron-emitting radionuclide was placed at the central position P4 is substantially symmetric about the peak position, but the distribution of the γ-ray pair generation positions obtained when a positron-emitting radionuclide was placed at a position other than the central position P4 is not symmetric about the peak position, and the peak position is biased toward the side far from the central position P4.

[0042] In Comparative Example 2B (Fig. 7), in addition to the above tendency of Comparative Example 2A, the following can be said. In Comparative Example 2B where the data when the positron-emitting nuclide was placed at position P5 was not used for the learning of the CNN, two peaks appear in the distribution of the γ-ray pair generation positions obtained when the positron-emitting nuclide was placed at position P5. Further, in Comparative Example 2B, the distribution of the γ-ray pair generation positions obtained when the positron-emitting nuclide was placed at the central position P4 is not symmetric about the peak position, and the peak position is biased toward the side of position P3.

[0043] Fig. 9 is a table summarizing the peak positions of the distributions of the γ-ray pair generation positions obtained in Comparative Example 1, Comparative Example 2A, Comparative Example 2B, and the Examples respectively. Fig. 10 is a table summarizing the full widths at half maximum of the distributions of the γ-ray pair generation positions obtained in Comparative Example 1, Comparative Example 2A, Comparative Example 2B, and the Examples respectively. Figs. 9 and 10 are obtained from the shapes of the distributions of the γ-ray pair generation positions shown in Figs. 5 to 8, and show the peak positions or the full widths at half maximum of the distributions of the γ-ray pair generation positions obtained when the positron-emitting nuclide was placed at each of the seven positions P1 to P7 in terms of time (unit: ps).

[0044] Since the seven positions P1 to P7 are separated at a 5 mm pitch, ideally, the peak positions of the distributions of the required γ-ray pair generation positions should be separated at a 33 ps pitch. As shown in Fig. 9, in the Examples and Comparative Example 1, the peak positions of the distributions of the γ-ray pair generation positions obtained are separated at approximately 33 ps pitch as ideally expected. In contrast, in Comparative Example 2A and Comparative Example 2B, the pitch of the peak positions of the distributions of the γ-ray pair generation positions obtained is different from the ideal, and particularly, the pitch of the peak positions of the distributions of the γ-ray pair generation positions becomes narrower the farther away from the central position P4.

[0045] The full width at half maximum of the obtained distribution of the γ-ray pair generation positions is, as shown in FIG. 10, the narrowest for Comparative Example 2A and Comparative Example 2B, and the next narrowest for the Examples. That is, the time resolution of the detection of the γ-ray pair generation positions is the highest for Comparative Example 2A and Comparative Example 2B, and the next highest for the Examples. When a positron-emitting nuclide is placed at the central position P4, the full width at half maximum of the distribution of the γ-ray pair generation positions obtained is 175.5 ps for Comparative Example 1, whereas it is 159.2 ps for the Examples, and the time resolution of the Examples is higher than that of Comparative Example 1.

[0046] The following can be said from the experimental results shown in FIGS. 5 to 10. In the present embodiment, as in Comparative Example 1, since the pitch of the peak positions of the distributions of the γ-ray pair generation positions obtained for the positron-emitting nuclides placed at each position with a constant pitch is also substantially constant, a tomographic image with little distortion can be created. In the present embodiment, compared with Comparative Example 1, a tomographic image with high time resolution and high spatial resolution of the detection of the γ-ray pair generation positions can be obtained.

[0047] In Comparative Example 2 (2A, 2B), although the γ-ray pair generation positions can be obtained with high time resolution, since the pitch of the peak positions of the distributions of the γ-ray pair generation positions obtained for the positron-emitting nuclides placed at each position with a constant pitch is not constant, the created tomographic image will be distorted. In Comparative Example 2, it is considered that a tomographic image with little distortion can be created by training a CNN using learning data obtained by densely placing positron-emitting nuclides at a large number of positions over a range wider than the space occupied by the subject (in some cases, a range wider than the measurement space surrounded by a large number of radiation detectors), but it is difficult to prepare such a large amount of learning data and it is also difficult to train the CNN.

[0048] FIG. 11 and FIG. 12 are diagrams showing positions where the positron-emitting nuclide 3 should be placed in the measurement space of the PET detector 20 in order to collect learning data in Comparative Example 2. FIG. 11 shows the case where there is no performance variation among a plurality of radiation detectors of the PET detector 20. In this case, it is necessary to densely place the positron-emitting nuclide 3 at a large number of positions on a straight line extending in the radial direction to collect learning data. FIG. 12 shows the case where there is performance variation among a plurality of radiation detectors of the PET detector 20. In this case, it is necessary to densely place the positron-emitting nuclide 3 at a large number of grid-like positions to collect learning data. In Comparative Example 2, in either case, the field of view of the apparatus will be limited.

[0049] On the other hand, in the present embodiment, since it is only necessary to train the DNN using learning data obtained by placing the positron-emitting nuclide at a smaller number of positions compared to Comparative Example 2 (2A, 2B), the DNN can be easily trained.

[0050] FIG. 13 and FIG. 14 are diagrams showing positions where the positron-emitting nuclide 3 should be placed in the measurement space of the PET detector 20 in order to collect learning data in the present embodiment. FIG. 13 shows the case where there is no performance variation among a plurality of radiation detectors of the PET detector 20. In this case, the learning data may be collected by placing the positron-emitting nuclide 3 at any one position in the measurement space. FIG. 14 shows the case where there is performance variation among a plurality of radiation detectors of the PET detector 20. In this case, for example, the learning data may be collected for all pairs of radiation detectors while rotating the positron-emitting nuclide 3 around the central axis in the measurement space. In the present embodiment, unlike Comparative Example 2, the field of view of the apparatus is not limited.

Description of Reference Numerals

[0051] 1... TOF-PET apparatus, 2... subject, 3... positron-emitting nuclide, 10... tomographic image creation apparatus, 11... signal waveform acquisition unit, 12... time difference calculation unit, 13... signal waveform processing unit, 14... error estimation unit, 15... γ-ray pair generation position calculation unit, 16... image creation unit, 17... learning unit, 20... PET detector, 21, 22... radiation detectors.

Claims

1. An apparatus for creating a tomographic image of a subject based on information on a plurality of pairs of γ-ray coincidence events collected for the subject placed in a measurement space of a PET detector including a plurality of radiation detectors, For each of the plurality of gamma-ray pair simultaneous counting events, a time difference t between the timings at which the values of the first signal and the second signal respectively output from the two radiation detectors that simultaneously counted the gamma-ray pair among the plurality of radiation detectors reach a threshold value led a time difference calculation unit that obtains it, A signal waveform processing unit that relatively shifts the waveform of the first signal or the waveform of the second signal by only the time difference t in a direction approaching each other in the time axis direction led and a signal waveform processing unit that relatively shifts the waveform of the first signal or the waveform of the second signal by only the time difference t in a direction approaching each other in the time axis direction Based on the waveforms of the first signal and the second signal after shifting by the signal waveform processing unit, an error estimation unit that estimates the time difference t led of the error t err using a deep neural network, and the time difference t led and the error t err Based on these, a gamma-ray pair generation position calculation unit that obtains the gamma-ray pair generation position on the coincidence counting line connecting the two radiation detectors to each other an image creation unit that creates a tomographic image of the subject based on the γ-ray pair generation positions obtained by a γ-ray pair generation position calculation unit for each of the plurality of pairs of γ-ray coincidence events; A tomographic image creation apparatus comprising the above.

2. Based on the information of a plurality of γ-ray pair coincidence counting events collected for the positron-emitting radionuclide placed in the measurement space, for each of the plurality of γ-ray pair coincidence counting events, the waveforms of the first signal and the second signal after the shift by the signal waveform processing unit are used as input data to the deep neural network, and the time difference t led and the difference from the true time difference based on the position of the positron-emitting radionuclide are used as teacher data, and further comprises a learning unit for learning the deep neural network. The tomographic image creation apparatus according to Claim 1.

3. A PET detector including a plurality of radiation detectors, The tomographic image creation apparatus according to Claim 1 or 2 that creates a tomographic image of the subject based on information on a plurality of pairs of γ-ray coincidence events collected for the subject placed in the measurement space of the PET detector; A TOF-PET apparatus comprising the above.

4. A method for creating a tomographic image of a subject based on information on a plurality of pairs of γ-ray coincidence events collected for the subject placed in a measurement space of a PET detector including a plurality of radiation detectors, For each of the plurality of gamma-ray pair simultaneous counting events, a time difference t between the timings at which the values of the first signal and the second signal respectively output from the two radiation detectors that simultaneously counted the gamma-ray pair among the plurality of radiation detectors reach a threshold value led a time difference calculation step for obtaining the time difference; A signal waveform processing step of relatively shifting the waveform of the first signal or the waveform of the second signal only by the time difference t in a direction approaching each other in the time axis direction led and a signal waveform processing step of relatively shifting the waveform of the first signal or the waveform of the second signal only by the time difference t in a direction approaching each other in the time axis direction Based on the waveforms of the first signal and the second signal after shifting by the signal waveform processing step, an error estimation step of estimating an error \(t\) of the time difference \(t\) by a deep neural network led of the error \(t\) err and an error estimation step the time difference t led and the error t err Based on these, a gamma-ray pair generation position calculation step for obtaining the gamma-ray pair generation position on the coincidence counting line connecting the two radiation detectors to each other; an image creation step of creating a tomographic image of the subject based on the γ-ray pair generation positions obtained by a γ-ray pair generation position calculation step for each of the plurality of pairs of γ-ray coincidence events; A tomographic image creation method comprising the above.

5. Based on the information of a plurality of γ-ray pair coincidence counting events collected for the positron-emitting radionuclide placed in the measurement space, for each of the plurality of γ-ray pair coincidence counting events, the waveforms of the first signal and the second signal after the shift by the signal waveform processing step are used as input data to the deep neural network, and the time difference t led and the difference from the true time difference based on the position of the positron-emitting radionuclide are used as teacher data, and further includes a learning step of training the deep neural network. The tomographic image creation method according to Claim 4.

Images