JP2021131327A - Detection system and learning method - Google Patents

Abstract

To provide a device which, in a detection system comprising a Cerenkov detector and a signal processing unit that uses a neural network, can easily train the neural network.SOLUTION: A detection system 1 comprises a Cerenkov detector 3 and a signal processing unit 4. The signal processing unit 4 includes an estimation unit 42, a learning data creation unit 43 and a learning unit 44. The estimation unit 42 estimates, for each mutual action event, a mutual action position by a neural network on the basis of the number n of Cerenkov light detection events and the detection potion and detection time of each Cerenkov light detection event. The learning data creation unit 43 generates, for each mutual action event, a plurality of less than n sets of Cerenkov light detection events from among the n Cerenkov light detection events by the Cerenkov detector 3, and uses the detection position and detection time of each Cerenkov light detection event included in each of these plurality of sets, as neural network learning data.SELECTED DRAWING: Figure 1

Description

The present invention relates to a detection system including a Cherenkov detector and a signal processing unit using a neural network, and a method for training the neural network.

When the moving speed of charged particles (for example, electrons, muons) in the radiator is faster than the light velocity in the radiator, light is emitted by the interaction between the charged particles and the radiator. In addition, uncharged particles (for example, γ-rays and neutrinos) may generate charged particles such as electrons in the radiator due to the photoelectric effect, and the moving speed of the charged particles is higher than the light velocity in the radiator. Light is emitted when it is fast. Such a phenomenon is called the Cherenkov effect or Cherenkov radiation, and the emitted light is called Cherenkov radiation or Cherenkov light.

The Cherenkov detector can detect incoming particles by utilizing the Cherenkov effect. The Cherenkov detector includes a radiator that generates Cherenkov light by interacting with incident particles and a photodetector that detects the Cherenkov light generated by the radiator. The Cherenkov detector is used as a radiation detector for detecting γ-rays (energy 511 keV in PET) arriving from a subject in, for example, a PET (Positron Emission Tomography) device or a SPECT (Single Photon Emission Computed Tomography) device.

Non-Patent Document 1 describes a technique of estimating the interaction position in a radiator by a neural network based on a Cherenkov light detection event by a Cherenkov detector photodetector.

Hashimoto, F., et al. “Afeasibility study on 3D interaction position estimation using deep neuralnetwork in Cherenkov-based detector: a Monte Carlo simulation study.” Biomedical Physics & Engineering Express, 5 (2019): 035001.

In the above technique, it is necessary to train a neural network using a large amount of learning data and then estimate the interaction position by the trained neural network. However, it is not easy to learn a neural network because it takes a long time to collect a large amount of learning data used for learning.

The present invention has been made to solve the above problems, and is an apparatus and method capable of easily learning the neural network in a detection system including a Cherenkov detector and a signal processing unit using a neural network. The purpose is to provide.

The detection system of the first aspect of the present invention has (1) a radiator that generates Cherenkov light by interaction with incident particles and a light receiving surface that detects Cherenkov light generated by the radiator on the light receiving surface. A Cherenkov detector that outputs a signal indicating the detection position and detection time of Cherenkov light, and (2) a signal output from the light detector is input, and each interaction event in the radiator Based on the number of Cherenkov light detection events on the light receiving surface and the detection position and detection time of each Cherenkov light detection event, an estimation unit that estimates the interaction position in the radiator by a neural network, and (3) for each interaction event In addition, a plurality of sets of less than n Cherenkov light detection events are generated from the n Cherenkov light detection events by the light detector of the Cherenkov detector, and each Cherenkov light detection event included in each of these multiple sets is generated. It is provided with a learning data creation unit that uses the detection position and detection time of the above as training data, and (4) a learning unit that trains a neural network using the learning data created by the learning data creation unit.

The detection system of the second aspect of the present invention has (1) a radiator that generates Cherenkov light by interaction with incident particles and a light receiving surface that detects Cherenkov light generated by the radiator, and is on the light receiving surface. A first Cherenkov detector and a second Cherenkov detector including, respectively, an optical detector that outputs a signal indicating a Cherenkov light detection position and a detection time, and (2) a first Cherenkov detector and a second Cherenkov detector, respectively. By inputting the signal output from the Cherenkov detector, the Cherenkov on the light receiving surface of each of the first Cherenkov detector and the second Cherenkov detector for each simultaneous counting event of the photon pair generated by the pair annihilation of the positivity and the electron. An estimation unit that estimates the photon pair generation position by a neural network based on the number of light detection events and the detection position and detection time of each Cherenkov light detection event, and (3) the first Cherenkov detector for each simultaneous counting event. A plurality of sets of less than n1 Cherenkov light detection events are generated from the n1 Cherenkov light detection events by the light detector of the above, and the detection position and detection of each Cherenkov light detection event included in each of these plurality of sets. Using the time as the first learning data, a plurality of sets of less than n2 Cherenkov light detection events are generated from the n2 Cherenkov light detection events by the light detector of the second Cherenkov detector, and these plurality of sets are generated. A learning data creation unit that uses the detection position and detection time of each Cherenkov light detection event included in each as the second learning data, and (4) the first learning data and the second learning data created by the learning data creation unit. It includes a learning unit that trains a neural network using training data.

The learning method of the first aspect of the present invention has (1) a radiator that generates Cherenkov light by interaction with incident particles and a light receiving surface that detects Cherenkov light generated by the radiator on the light receiving surface. A Cherenkov detector that outputs a signal indicating the detection position and detection time of Cherenkov light, and (2) a signal output from the light detector is input, and each interaction event in the radiator , A neural network of a detection system comprising an estimation unit that estimates the interaction position in the radiator by a neural network based on the number of Cherenkov light detection events on the light receiving surface and the detection position and detection time of each Cherenkov light detection event. Is a way to learn. The learning method of the first aspect of the present invention is (a) for each interaction event, a plurality of sets of less than n Cherenkov light detection events out of n Cherenkov light detection events by the light detector of the Cherenkov detector. A learning data creation step that is generated and uses the detection position and detection time of each Cherenkov light detection event included in each of these plurality of sets as learning data, and (b) learning created in the learning data creation step. It includes a learning step of training a neural network using the data for training.

The learning method of the second aspect of the present invention has (1) a radiator that generates Cherenkov light by interaction with incident particles and a light receiving surface that detects Cherenkov light generated by the radiator, and is on the light receiving surface. A first Cherenkov detector and a second Cherenkov detector including, respectively, an optical detector that outputs a signal indicating a Cherenkov light detection position and a detection time, and (2) a first Cherenkov detector and a second Cherenkov detector, respectively. By inputting the signal output from the Cherenkov detector, the Cherenkov on the light receiving surface of each of the first Cherenkov detector and the second Cherenkov detector for each simultaneous counting event of photon pairs generated by the pair annihilation of positivity and electrons. This is a method of learning a neural network of a detection system including an estimation unit that estimates a photon pair generation position by a neural network based on the number of light detection events and the detection position and detection time of each Cherenkov light detection event. The learning method of the second aspect of the present invention is (a) a set of less than n1 Cherenkov light detection events out of n1 Cherenkov light detection events by the light detector of the first Cherenkov detector for each simultaneous counting event. Are generated, and the detection position and detection time of each Cherenkov light detection event included in each of these plurality of sets are used as the first learning data, and n2 Cherenkov light detections by the light detector of the second Cherenkov detector. For learning, a plurality of Cherenkov light detection event pairs of less than n2 are generated from the events, and the detection position and detection time of each Cherenkov light detection event included in each of these plurality of sets are used as the second learning data. It includes a data creation step and a learning step in which a neural network is trained using the first training data and the second training data created in (b) the training data creation step.

According to the present invention, the neural network can be easily learned in a detection system including a Cherenkov detector and a signal processing unit using a neural network.

FIG. 1 is a diagram showing a configuration of the detection system 1 of the first embodiment. FIG. 2 is a perspective view showing the configuration of the Cherenkov detector 3. FIG. 3 is a diagram illustrating an example of processing by the estimation unit 42 of the signal processing unit 4 when estimating the interaction position. FIG. 4 is a diagram illustrating an example of processing by the learning data creation unit 43 and the learning unit 44 of the signal processing unit 4 when learning the neural network. FIG. 5 is a diagram showing an example of the configuration of the neural network. FIG. 6 is a diagram showing the results of the simulation. FIG. 7 is a diagram showing a configuration of the detection system 2 of the second embodiment. FIG. 8 is a diagram illustrating an example of processing by the estimation unit 52 of the signal processing unit 5 when estimating the photon pair generation position. FIG. 9 is a diagram illustrating an example of processing by the learning data creation unit 53 and the learning unit 54 of the signal processing unit 5 when learning the neural network. FIG. 10 is a diagram showing an example of the configuration of the neural network. FIG. 11 is a diagram showing the results of the simulation.

Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same elements are designated by the same reference numerals, and duplicate description will be omitted. The present invention is not limited to these examples, and is indicated by the scope of claims, and is intended to include all modifications within the meaning and scope equivalent to the scope of claims.

FIG. 1 is a diagram showing a configuration of the detection system 1 of the first embodiment. The detection system 1 includes a Cherenkov detector 3 and a signal processing unit 4. The Cherenkov detector 3 includes a radiator 10 and a photodetector 20. FIG. 2 is a perspective view showing the configuration of the Cherenkov detector 3.

The radiator 10 generates Cherenkov light by interacting with incident particles (for example, γ-rays). The radiator 10 is made of a monolithic material having a known refractive index and capable of producing a Cherenkov effect. The radiator 10 is preferably made of a material that does not easily generate scintillation light. The material of the radiator 10 is, for example, lead glass (SiO ₂ + PbO) _{, lead fluoride (PbF 2} ), PWO (PbWO ₄ ) and the like. The radiator 10 preferably has a rectangular parallelepiped shape.

The photodetector 20 has a light receiving surface 21 that detects Cherenkov light generated by the radiator 10. The light receiving surface 21 of the photodetector 20 faces one surface of the rectangular parallelepiped radiator 10 and is parallel to that surface. The light receiving surface 21 of the photodetector 20 may be in contact with one surface of the radiator 10. The photodetector 20 outputs a signal representing the detection position (x, y) and the detection time t of the Cherenkov light on the light receiving surface 21 based on the Cherenkov light detection on the light receiving surface 21. The signal can represent the detection time t by the pulse output at the time of Cherenkov light detection.

The photodetector 20 is preferably a high-speed, high-sensitivity photodetector capable of performing a photon counting operation. As the photodetector 20, for example, a multi-anode photomultiplier tube or MPPC® is preferably used. A multi-anode photomultiplier tube has a plurality of anodes as a plurality of pixels and can output a detection signal according to the amount of light received by each anode. An MPPC (Multi-Pixel Photon Counter) is a pixel in which a quenching resistor is connected to an avalanche photodiode operating in a Geiger mode as one pixel, and a plurality of pixels are arranged two-dimensionally. These can perform high-speed and high-sensitivity photodetection.

The signal processing unit 4 inputs a signal output from the photodetector 20. By processing this input signal, the signal processing unit 4 processes the number n of Cherenkov light detection events on the light receiving surface 21 of the photodetector 20 and the detection position of each Cherenkov light detection event for each interaction event in the radiator 10. Based on (x, y) and the detection time t, the interaction position (x, y, z) in the radiator 10 is estimated by the neural network. The x-axis and y-axis of the x, y, z Cartesian coordinate system are parallel to the light receiving surface 21 of the photodetector 20 and are orthogonal to each other. The z-axis is perpendicular to the light receiving surface 21 of the photodetector 20.

Further, the signal processing unit 4 detects the number n of Cherenkov light detection events on the light receiving surface 21 of the photodetector 20 and the detection positions (x, y) of each Cherenkov light detection event for each interaction event in the radiator 10. Based on the time t, training data for training the neural network is created. Then, the signal processing unit 4 trains the neural network using the learning data.

The signal processing unit 4 is composed of, for example, a computer. The signal processing unit 4 may perform processing by a neural network by a CPU (Central Processing Unit), but it is preferable to perform processing by a DSP (Digital Signal Processor) or a GPU (Graphics Processing Unit) capable of higher speed processing. be.

The signal processing unit 4 includes a preprocessing unit 41, an estimation unit 42, a learning data creation unit 43, a learning unit 44, and a storage unit 45.

The preprocessing unit 41 obtains a detection position (x, y) and a detection time t for each Cherenkov light detection event based on the signal output from the photodetector 20. The detection time t can be obtained from the timing of the pulse of the input signal. After that, the preprocessing unit 41 sorts the data including the detection position (x, y) of each Cherenkov light detection event and the detection time t by the detection time t. Then, the preprocessing unit 41 determines that the data in the time window having a certain width among the sorted data is the data of the Cherenkov light detection event caused by the common interaction event.

In this way, the preprocessing unit 41 classifies the data including the detection position (x, y) and the detection time t of each Cherenkov light detection event for each interaction event. Thereby, for each interaction event, the number n of Cherenkov light detection events on the light receiving surface 21, the detection position (x, y) of each Cherenkov light detection event, and the detection time t can be obtained.

Further, the preprocessing unit 41 discards the data of the interaction event in which the number n of the Cherenkov light detection events is less than the threshold th, and the estimation unit 42 discards the data of the interaction event in which the number n of the Cherenkov light detection events is the threshold th or more. It is preferable to selectively estimate the interaction position for the action event. This is because, in the case of an interaction event in which the number n of Cherenkov light detection events is less than the threshold value th, the accuracy of estimating the interaction position by the estimation unit 42 is poor. Similarly, the learning data creation unit 43 preferably creates learning data for interaction events in which the number n of Cherenkov light detection events is equal to or greater than the threshold value th. For example, the threshold th is 3.

The estimation unit 42 inputs the data after the preprocessing by the preprocessing unit 41, and for each interaction event, the number n of Cherenkov light detection events on the light receiving surface 21 of the photodetector 20 and each Cherenkov light. The interaction position (x, y, z) is estimated by a neural network based on the detection position (x, y) of the detection event and the detection time t. The neural network used here is, for example, a multi-layer perceptron (MLP), a convolutional neural network (CNN), a recurrent neural network (RNN), or the like.

The estimation unit 42 may have a plurality of neural networks and estimate the interaction position by the neural network corresponding to the number n of Cherenkov light detection events. That is, for an interaction event in which the number of Cherenkov light detection events is n, the interaction position may be estimated by a neural network provided corresponding to the number n.

The learning data creation unit 43 generates a plurality of sets of less than n Cherenkov light detection events from the n Cherenkov light detection events by the photodetector 20 of the Cherenkov detector 3 for each interaction event. The detection position and detection time of each Cherenkov light detection event included in each of these plurality of sets are used as training data. When n Cherenkov light detection events are included in the interaction event and m pairs of Cherenkov light detection events are generated from them (however, m <n), m out of n sets are selected. The number of combinations to be taken out is _n C _m .

The learning unit 44 trains the neural network using the learning data created by the learning data creation unit 43. The storage unit 45 stores data before, during, and after processing by the preprocessing unit 41, the estimation unit 42, the learning data creation unit 43, and the learning unit 44, respectively.

FIG. 3 is a diagram illustrating an example of processing by the estimation unit 42 of the signal processing unit 4 when estimating the interaction position. The estimation unit 42 inputs data for each interaction event after the preprocessing is performed by the preprocessing unit 41. That is, the estimation unit 42 inputs the data (x, y, t) of each of the n Cherenkov light detection events included in each interaction event. The estimation unit 42 selects the data of five Cherenkov light detection events with early detection timing from the n Cherenkov light detection events included in each interaction event, and the data of the five Cherenkov light detection events ( x, y, t) is input to the trained neural network. As a result, the data output from the neural network represents the result of estimating the interaction position (x, y, z) in the radiator 10.

FIG. 4 is a diagram illustrating an example of processing by the learning data creation unit 43 and the learning unit 44 of the signal processing unit 4 when learning the neural network. The learning data creation unit 43 inputs data for each interaction event after the preprocessing is performed by the preprocessing unit 41. That is, the learning data creation unit 43 inputs the data (x, y, t) of each of the n Cherenkov light detection events included in each interaction event.

The learning data creation unit 43 generates a plurality of sets of five Cherenkov light detection events from the n Cherenkov light detection events included in each interaction event. At this time, a maximum of _{5 sets of n} C can be generated. The learning data creation unit 43 uses the data (x, y, t) of each of the five Cherenkov light detection events included in each set as training data, and causes the learning data to be input to the neural network. As a result, the same number of output data (x, y, z) as the number of pairs can be obtained from the neural network. The learning unit 44 evaluates an error between the data (x, y, z) output from the neural network and the teacher data (X, Y, Z), and uses an error backpropagation method so that the error becomes small. Adjust the parameters of the neural network to train the neural network.

FIG. 5 is a diagram showing an example of the configuration of the neural network. The neural network shown in this figure is a multi-layer perceptron (MLP) having a five-layer structure including an input layer, an output layer and three intermediate layers. This MLP inputs the data (x, y, t) of each of the five Cherenkov light detection events, and outputs the estimation result of the interaction position (x, y, z). The output data from the neural network may represent the position or the certainty of the interaction of each position.

Next, the result of the simulation performed by using the Monte Carlo method for the first embodiment will be described. As the radiator 10, it is assumed that the radiator 10 is composed of PbF ₂ and has a rectangular parallelepiped shape of 40 × 40 × 10 mm ^3. It is assumed that the light receiving surface 21 of the photodetector 20 is in contact with the surface of the radiator 10 having a size of 40 × 40 mm ² . An ideal photodetector 20 is assumed in which both spatial resolution and temporal resolution are infinitesimal. The MLP shown in FIG. 5 was used as a neural network, and data representing the position was output from the MLP.

In the simulation, the interaction of each particle in the radiator is tracked one by one, and the interaction position in the radiator can be known by referring to the history, so this can be used as training data (X, Y, Z). can.

In the case of experiments, for example, collimated gamma rays (pencil beams) are used. The position when the pencil beam is vertically incident on the xy plane is used as a substitute for the interaction position (X, Y). Similarly, when the pencil beam is incident vertically on the xz plane, (X, Z) is known, and when the pencil beam is vertically incident on the yz plane, (Y, Z) is known. In this method, teacher data (X, Y, Z) cannot be obtained in one shot, so the neural network for each plane is learned individually and the predictions are integrated. For example, the average of the outputs of the neural networks for (X, Y), (X, Z), and (Y, Z) can be used as the teacher data (X, Y, Z).

The simulation was performed for each of the first comparative example, the second comparative example, and the first embodiment. In the first comparative example, MLP was trained using only the data in which the number of Cherenkov light detection events included in the interaction event was 5. In the second comparative example, among the data in which the number of Cherenkov light detection events included in the interaction event is 5 or more and 7 or less, the data of 5 Cherenkov light detection events with early detection timing are used as learning data to perform MLP. I learned. In the first embodiment, based on the data in which the number n of Cherenkov light detection events included in the interaction event is 5 or more and 7 or less, the learning data created under the conditions described as an example using FIG. 4 is used. MLP was trained.

The number of learning data created in the first comparative example was 521,460. The number of learning data created in the second comparative example was 875,000, which was 1.68 times that in the case of the first comparative example. The number of learning data created in the first example was 4,170,600, which was eight times that in the case of the first comparative example. As described above, as compared with the first comparative example and the second comparative example, more learning data can be prepared in the first embodiment.

FIG. 6 is a diagram showing the results of the simulation. This figure shows a histogram of the estimation error of the interaction position on the xy plane and a histogram of the estimation error of the interaction position in the z-axis direction for each of the first comparative example, the second comparative example, and the first embodiment. .. The estimation error is the value obtained by subtracting the estimated interaction position from the true interaction position. The horizontal axis represents the error and the vertical axis represents the frequency. Each histogram was approximated by the Lorentz function, and the full width at half maximum (FWHM) of the Lorentz function was obtained.

In the first comparative example, the FWHM of the estimation error in the xy plane was 0.89 mm, and the FWHM of the estimation error in the z-axis direction was 1.57 mm. In the second comparative example, the FWHM of the estimation error in the xy plane was 0.79 mm, and the FWHM of the estimation error in the z-axis direction was 1.42 mm. In the first embodiment, the FWHM of the estimation error in the xy plane was 0.65 mm, and the FWHM of the estimation error in the z-axis direction was 1.40 mm. Compared with the first comparative example and the second comparative example, in the first embodiment, both the estimation error in the axial direction and the estimation error in the xy plane were smaller values. In the first embodiment, it is considered that the accuracy of estimating the interaction position is improved by training the MLP using more learning data.

Next, the second embodiment will be described. FIG. 7 is a diagram showing a configuration of the detection system 2 of the second embodiment. The detection system 2 includes a first Cherenkov detector 3A, a second Cherenkov detector 3B, and a signal processing unit 5. The Cherenkov detectors 3A and 3B have the same configuration as the Cherenkov detector 3 in the first embodiment. The incident surfaces 11 (planes opposite to the photodetector 20 side, see FIG. 2) of the radiators 10 of the Cherenkov detectors 3A and 3B are opposed to each other.

When there is an RI radiation source in the space between the first Cherenkov detector 3A and the second Cherenkov detector 3B, the positrons and electrons emitted from the RI radiation source are pair-annihilated, and the pair annihilation causes photon pairs ( A pair of γ-rays with energy of 511 keV) is generated. The two photons that make up a photon pair fly in opposite directions. By detecting one photon by the first Cherenkov detector 3A and the other photon by the second Cherenkov detector 3B, that is, by simultaneously counting the photon pairs by the two Cherenkov detectors 3A and 3B (coincidence counting). The photon pair generation event (pair annihilation event) can be detected by the counting event).

The signal processing unit 5 inputs signals output from the photodetectors 20 of the Cherenkov detectors 3A and 3B, respectively. By processing these input signals, the signal processing unit 5 processes the number of Cherenkov light detection events and each Cherenkov light detection event on the light receiving surface of each of the Cherenkov detectors 3A and 3B for each simultaneous counting event. The photon pair generation position (pair annihilation position) is estimated by the neural network based on the detection position and the detection time of.

Further, the signal processing unit 5 is based on the number of Cherenkov light detection events on the light receiving surface of each of the Cherenkov detectors 3A and 3B, and the detection position and detection time of each Cherenkov light detection event for each simultaneous counting event. To create training data for training the neural network. Then, the signal processing unit 5 trains the neural network using the learning data.

The signal processing unit 5 is composed of, for example, a computer. The signal processing unit 5 may perform the processing by the neural network by the CPU, but it is preferable to perform the processing by the DSP or GPU capable of higher speed processing.

The signal processing unit 5 includes a preprocessing unit 51, an estimation unit 52, a learning data creation unit 53, a learning unit 54, and a storage unit 55.

The pre-processing unit 51 performs the same processing as the pre-processing unit 41 in the first embodiment for the signal output from the photodetector 20 of the first Cherenkov detector 3A. The preprocessing unit 51 also performs the same processing as the preprocessing unit 41 in the first embodiment for the signal output from the photodetector 20 of the second Cherenkov detector 3B.

The preprocessing unit 51 also performs the following processing. The preprocessing unit 51 includes the interaction event data detected based on the signal output from the photodetector 20 of the first Cherenkov detector 3A and the signal output from the photodetector 20 of the second Cherenkov detector 3B. When the data of the interaction event detected based on the above is in the coincidence counting time window, it is determined that both data are due to the common photon pair generation event. At this time, it may be determined whether or not the representative times of both data are in the coincidence counting time window. The representative time may be the earliest detection time included in each interaction event, or may be the average value of the detection times.

The estimation unit 52 inputs the data after the preprocessing by the preprocessing unit 51, and for each simultaneous counting event, the Cherenkov light detection event on the light receiving surface of the photodetectors 20 of the Cherenkov detectors 3A and 3B, respectively. The photon pair generation position (x, y, z) is estimated by a neural network based on the number of light detection events and the detection position and detection time of each Cherenkov light detection event. The neural network used here is, for example, a multi-layer perceptron (MLP), a convolutional neural network (CNN), a recurrent neural network (RNN), or the like.

The learning data creation unit 53 generates a plurality of sets of less than n1 Cherenkov light detection events from the n1 Cherenkov light detection events by the light detector 20 of the first Cherenkov detector 3A for each simultaneous counting event. Therefore, the detection position and detection time of each Cherenkov light detection event included in each of these plurality of sets are used as the first learning data, and n2 Cherenkov light detections by the light detector 20 of the second Cherenkov detector 3B are used. A plurality of sets of less than n2 Cherenkov light detection events are generated from the events, and the detection position and detection time of each Cherenkov light detection event included in each of the plurality of sets are used as the second learning data.

The learning unit 54 trains the neural network using the first learning data and the second learning data created by the learning data creation unit 53. The storage unit 55 stores data before, during, and after processing by the preprocessing unit 51, the estimation unit 52, the learning data creation unit 53, and the learning unit 54, respectively.

FIG. 8 is a diagram illustrating an example of processing by the estimation unit 52 of the signal processing unit 5 when estimating the photon pair generation position. The estimation unit 52 inputs data for each coincidence counting event after the preprocessing is performed by the preprocessing unit 51. That is, the estimation unit 52 inputs the data (x, y, t) of each of n1 Cherenkov light detection events by the first Cherenkov detector 3A for each simultaneous counting event, and n2 by the second Cherenkov detector 3B. Enter the data (x, y, t) for each of the Cherenkov light detection events. The estimation unit 52 selects the data of five Cherenkov light detection events with early detection timing from the data of each of the n1 Cherenkov light detection events by the first Cherenkov detector 3A, and n2 by the second Cherenkov detector 3B. From the data of each of the Cherenkov light detection events, the data of the five Cherenkov light detection events with the earliest detection timing are selected, and these selected data are input to the trained neural network. As a result, the data output from the neural network represents the result of estimating the photon pair generation position (x, y, z).

FIG. 9 is a diagram illustrating an example of processing by the learning data creation unit 53 and the learning unit 54 of the signal processing unit 5 when learning the neural network. The learning data creation unit 53 inputs data for each coincidence counting event after the preprocessing is performed by the preprocessing unit 51. That is, the learning data creation unit 53 inputs data (x, y, t) for each of n1 Cherenkov light detection events by the first Cherenkov detector 3A for each simultaneous counting event, and the second Cherenkov detector. The data (x, y, t) of each of n2 Cherenkov light detection events by 3B is input.

The learning data creation unit 53 generates a plurality of sets of five Cherenkov light detection events from the n1 Cherenkov light detection events by the first Cherenkov detector 3A. At this time, a maximum of _{5 sets of n1} C can be generated. The learning data creation unit 53 uses the data (x, y, t) of each of the five Cherenkov light detection events included in each set as the first learning data.

The learning data creation unit 53 generates a plurality of sets of five Cherenkov light detection events from the n2 Cherenkov light detection events by the second Cherenkov detector 3B. At this time, a maximum of _{5 sets of n2} C can be generated. The learning data creation unit 53 uses the data (x, y, t) of each of the five Cherenkov light detection events included in each set as the second learning data.

The learning data creation unit 53 causes the first learning data and the second learning data to be input to the neural network. The learning data creation unit 53 _{can input a maximum of n1} C ₅ × _n2 C ₅ sets of learning data to the neural network. As a result, the same number of output data (x, y, z) as the number of pairs can be obtained from the neural network. The learning unit 54 evaluates an error between the data (x, y, z) output from the neural network and the teacher data (X, Y, Z), and uses an error backpropagation method so that the error becomes small. Adjust the parameters of the neural network to train the neural network.

FIG. 10 is a diagram showing an example of the configuration of the neural network. The neural network shown in this figure is a multi-layer perceptron (MLP) having a five-layer structure including an input layer, an output layer and three intermediate layers. In this MLP, the data (x, y, t) of each of the five Cherenkov light detection events selected from the n1 Cherenkov light detection events detected by the first Cherenkov detector 3A are input, and the second Input the data (x, y, t) of each of the five Cherenkov light detection events selected from the n2 Cherenkov light detection events detected by the Cherenkov detector 3B, and enter the photon pair generation position (x, y). , Z) Output the estimation result. The output data from the neural network may represent a position or a certainty of photon pair generation at each position.

Next, the result of the simulation performed by using the Monte Carlo method for the second embodiment will be described. The above-mentioned ones are assumed as the first Cherenkov detector 3A and the second Cherenkov detector 3B, respectively. It is assumed that the midpoint of the line segment connecting the center positions of the incident surfaces 11 of the first Cherenkov detector 3A and the second Cherenkov detector 3B to each other is the origin of the xyz Cartesian coordinate system, and that the line segment is on the z-axis. It is assumed that each side of the incident surface 11 is parallel to the x-axis or the y-axis, and the distance between the two incident surfaces 11 is 50 mm. The MLP shown in FIG. 10 was used as a neural network, and data representing the certainty of photon pair generation at each position was output from the MLP.

As photon pair generation positions (teacher data (X, Y, Z)), 125 (= 5 x 5 x 5) positions are set at 5 mm intervals in the positive and negative directions of the x-axis, y-axis, and z-axis with the origin as the center. I assumed. Considering the symmetry of the three-dimensional space, an RI radiation source is placed in one quadrant (and on the axis of the boundary of the quadrant) to generate a photon pair to collect coincidence counting information, and this coincidence counting is performed. Coincidence counting information was created for other quadrants by using the information.

The simulation was performed for each of the third comparative example, the fourth comparative example, and the second embodiment. In the third comparative example, MLP was trained using only the data in which the number n1 and n2 of the interaction events included in the common photon pair generation event were 5 as learning data. In the fourth comparative example, among the data in which the number n1 and n2 of the number of interaction events included in the common photon pair generation event are both 5 or more and 8 or less, the data of 5 Cherenkov light detection events with early detection timing are used. The MLP was trained using it as training data. In the second embodiment, it was created under the conditions described as an example using FIG. 9 based on the data that the number n1 and n2 of the interaction events included in the common photon pair generation event are both 5 or more and 8 or less. The MLP was trained using the training data.

The number of learning data created in the third comparative example was 43,000. The number of learning data created in the fourth comparative example was 122,125, which was 2.84 times that in the case of the third comparative example. The number of learning data created in the second example was 6,666,375, which was 155 times that in the case of the third comparative example. As described above, as compared with the third comparative example and the fourth comparative example, more learning data can be prepared in the second embodiment.

FIG. 11 is a diagram showing the results of the simulation. This figure shows the accuracy of estimating the photon pair generation position in each of the third comparative example, the fourth comparative example, and the second embodiment. In the third comparative example, the estimation accuracy was 91.45%, in the fourth comparative example, the estimation accuracy was 93.80%, and in the second example, the estimation accuracy was 95.11%. Compared with the third comparative example and the fourth comparative example, in the second embodiment, the accuracy of estimating the photon pair generation position was high. In the second embodiment, it is considered that the accuracy of estimating the photon pair generation position is improved by training the MLP using more learning data.

If the number of learning data is biased depending on the photon pair generation position, it is considered that the estimation of the photon pair generation position by the neural network trained using the biased learning data is also biased. Therefore, the coincidence counting data was collected while checking the number of training data created for each photon pair generation position between the two Cherenkov detectors, and the training data created for a certain photon pair generation position was collected. When the number reaches a certain target value, it is preferable to end the collection of coincidence counting data for the photon pair generation position. In this case, the time required to collect the coincidence counting data for each photon pair generation position may differ depending on the position.

As described above, according to the present embodiment, in a detection system including a Cherenkov detector and a signal processing unit using a neural network, a large amount of learning data used when training the neural network is created in a short time. And the neural network can be easily trained. The learning of the neural network can be easily performed by the manufacturer of the detection system and can be easily performed by the user of the detection system. For example, when it becomes necessary to calibrate the detection system, the user can easily perform re-learning or additional learning of the neural network by himself / herself.

1,2 ... Detection system, 3 ... Cherenkov detector, 3A ... 1st Cherenkov detector, 3B ... 2nd Cherenkov detector, 10 ... Radiant body, 20 ... Optical detector, 4 ... Signal processing unit, 41 ... Preprocessing Unit, 42 ... estimation unit, 43 ... learning data creation unit, 44 ... learning unit, 45 ... storage unit, 5 ... signal processing unit, 51 ... preprocessing unit, 52 ... estimation unit, 53 ... learning data creation unit, 54 ... Learning department, 55 ... Memory department.

Claims (4)

It has a radiator that generates Cherenkov light by interaction with incident particles and a light receiving surface that detects Cherenkov light generated by the radiator, and a signal that represents the detection position and detection time of Cherenkov light on the light receiving surface. An output light detector, including a Cherenkov detector,
A signal output from the photodetector is input, and each interaction event in the radiator is based on the number of Cherenkov light detection events on the light receiving surface, the detection position of each Cherenkov light detection event, and the detection time. With a neural network, an estimation unit that estimates the interaction position in the radiator,
For each interaction event, a plurality of sets of less than n Cherenkov light detection events are generated from the n Cherenkov light detection events by the photodetector of the Cherenkov detector, and are included in each of these plurality of sets. A learning data creation unit that uses the detection position and the detection time of each Cherenkov light detection event as training data,
A learning unit that trains the neural network using the learning data created by the learning data creation unit, and a learning unit.
Detection system with. It has a radiator that generates Cherenkov light by interaction with incident particles and a light receiving surface that detects Cherenkov light generated by the radiator, and a signal that represents the detection position and detection time of Cherenkov light on the light receiving surface. A first Cherenkov detector and a second Cherenkov detector, including an output light detector, respectively.
The signals output from the light detectors of the first Cherenkov detector and the second Cherenkov detector are input, and the photon pair simultaneous counting event generated by the pair annihilation of positrons and electrons is described as the first. Based on the number of Cherenkov light detection events on the light receiving surface of each of the 1 Cherenkov detector and the 2nd Cherenkov detector, the detection position of each Cherenkov light detection event, and the detection time, the photon pair generation position is determined by a neural network. Estimating part to estimate and
For each simultaneous counting event, a plurality of sets of less than n1 Cherenkov light detection events are generated from the n1 Cherenkov light detection events by the light detector of the first Cherenkov detector, and each of these plurality of sets is used. The detection position and the detection time of each Cherenkov light detection event included are used as the first learning data, and less than n2 Cherenkov lights out of the n2 Cherenkov light detection events by the light detector of the second Cherenkov detector. A learning data creation unit that generates a plurality of sets of detection events and uses the detection position and the detection time of each Cherenkov light detection event included in each of the plurality of sets as the second learning data.
A learning unit that trains the neural network using the first learning data and the second learning data created by the learning data creation unit, and a learning unit.
Detection system with. It has a radiator that generates Cherenkov light by interaction with incident particles and a light receiving surface that detects Cherenkov light generated by the radiator, and a signal that represents the detection position and detection time of Cherenkov light on the light receiving surface. An output light detector, including a Cherenkov detector,
A signal output from the photodetector is input, and each interaction event in the radiator is based on the number of Cherenkov light detection events on the light receiving surface, the detection position of each Cherenkov light detection event, and the detection time. With a neural network, an estimation unit that estimates the interaction position in the radiator,
A method of training the neural network of a detection system comprising
For each interaction event, a plurality of sets of less than n Cherenkov light detection events are generated from the n Cherenkov light detection events by the photodetector of the Cherenkov detector, and are included in each of these plurality of sets. A training data creation step in which the detection position and the detection time of each Cherenkov light detection event are used as training data, and
A learning step in which the neural network is trained using the learning data created in the learning data creation step, and
Learning method with. It has a radiator that generates Cherenkov light by interaction with incident particles and a light receiving surface that detects Cherenkov light generated by the radiator, and a signal that represents the detection position and detection time of Cherenkov light on the light receiving surface. A first Cherenkov detector and a second Cherenkov detector, including an output light detector, respectively.
The signals output from the light detectors of the first Cherenkov detector and the second Cherenkov detector are input, and the photon pair simultaneous counting event generated by the pair annihilation of positrons and electrons is described as the first. Based on the number of Cherenkov light detection events on the light receiving surface of each of the 1 Cherenkov detector and the 2nd Cherenkov detector, the detection position of each Cherenkov light detection event, and the detection time, the photon pair generation position is determined by a neural network. Estimating part to estimate and
A method of training the neural network of a detection system comprising
For each simultaneous counting event, a plurality of sets of less than n1 Cherenkov light detection events are generated from the n1 Cherenkov light detection events by the light detector of the first Cherenkov detector, and each of these plurality of sets is used. The detection position and the detection time of each Cherenkov light detection event included are used as the first learning data, and less than n2 Cherenkov lights out of the n2 Cherenkov light detection events by the light detector of the second Cherenkov detector. A learning data creation step in which a plurality of sets of detection events are generated and the detection position and the detection time of each Cherenkov light detection event included in each of the plurality of sets are used as the second learning data.
A learning step in which the neural network is trained using the first learning data and the second learning data created in the learning data creation step, and
Learning method with.

Images