Time-series signal prediction device, radiation therapy device, time-series signal prediction method, program, and storage medium
The time-series signal prediction device addresses the challenge of predicting tumor position during radiation therapy by dividing biological information signals into block signals and using feature vectors, enhancing prediction accuracy and speed to minimize unnecessary irradiation.
Patent Information
- Application Number
- PCT/JP2024/020754
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-06-21
- Filing Date
- 2024-06-06
- Publication Date
- 2024-12-26
AI Technical Summary
Current methods for predicting tumor position during radiation therapy, especially in organs that move due to breathing or heartbeat, face challenges in achieving high-speed and high-accuracy predictions, leading to unnecessary irradiation of normal tissues.
A time-series signal prediction device that acquires biological information signals, divides them into block signals, and uses feature vectors to predict future signal values, employing state space models and neural networks for accurate and efficient prediction, thereby optimizing the timing of radiation delivery.
The solution enables high-speed and highly accurate prediction of tumor position, reducing unnecessary radiation exposure to normal tissues by improving prediction accuracy and reducing calculation time.
Smart Images

Figure JP2024020754_26122024_PF_FP_ABST
Abstract
Description
Time series signal prediction device, radiation therapy device, time series signal prediction method, program, and storage medium
[0001] Embodiments of the present invention relate to a time-series signal prediction device, a radiotherapy device, a time-series signal prediction method, a program, and a storage medium.
[0002] Radiation therapy is a treatment method that irradiates a patient's lesion with radiation to destroy it, but if the radiation is not directed precisely at the location of the lesion, normal tissue will also be destroyed. Therefore, CT scans are performed in advance to determine the three-dimensional location of the lesion inside the patient's body, and the direction and intensity of the radiation are planned to minimize irradiation of normal tissue.
[0003] In radiation therapy, in order to irradiate radiation according to the plan, it is necessary to align the patient's position between the time of treatment planning and the time of treatment. To align the positions of lesions, bones, etc. within the patient's body with those at the time of treatment planning, one method involves comparing a fluoroscopic image of the patient's body while lying on a bed in a treatment room immediately before treatment with a digitally reconstructed radiograph (DRR), which is a virtual fluoroscopic image reconstructed from a 3D CT image taken during treatment planning, to determine the deviation in the patient's position between the images, and then moving the bed based on the deviation. The deviation in the patient's position is determined by searching for the position on the CT image from which the DRR most similar to the fluoroscopic image is reconstructed. Many methods have been proposed to automate this search using a computer, but the search results must be confirmed by a radiation therapist by visually comparing the fluoroscopic image with the DRR.
[0004] When a patient's tumor is located in an organ, such as the lung or liver, that moves with breathing and heartbeat, a respiratory-gated irradiation method is known. This method identifies the tumor's location during irradiation and irradiates it with a treatment beam when the identified location coincides with the location defined in the treatment plan. One method for estimating tumor location involves sensing the movement of the thoracic and abdominal body surface, which expands and contracts with breathing, and estimating the tumor's location based on the sensor values. Another method involves capturing fluoroscopic video of the patient during irradiation and tracking the tumor's position within the fluoroscopic images. Hereinafter, the signals obtained from the body surface movement and the tumor position trajectory are referred to as "time-series signals." While controlling the timing of treatment beam irradiation in this way can reduce unnecessary irradiation of the patient with the treatment beam, the system's response time, including the computational time required for tumor location estimation, necessitates tumor location prediction.
[0005] For example, the method disclosed in Non-Patent Document 1 predicts signal values at a predetermined time by training a recurrent network, LSTM (Long Short Term Memory), using time-series signals from multiple patients. The method disclosed in Non-Patent Document 2, unlike the method of Non-Patent Document 1, uses a feedforward network, CNN (Convolutional Neural Network). This disclosed technology predicts signal values at a future time (T+M, M>0) from a time-series signal up to a certain time (T). This method is expected to achieve higher accuracy as the input time-series signal length increases, but the associated calculation time increases, making it unsuitable for prediction. Furthermore, teaching waveform signals from time T+1 to T+M, rather than just the signal value at time T+M, increases the amount of information used for training, and therefore is expected to improve prediction accuracy.
[0006] “Towards real-time respiratory motion prediction based on long short-term memory neural networks”, Phys Med Biol. 2019 Apr 10; 64(8): 085010. “Respiratory Prediction Based on Multi-Scale Temporal Convolutional Network for Tracking Thoracic Tumor Movement”, Front. Oncol., 27 May 2023
[0007] The problem to be solved by the present invention is to provide a time-series signal prediction device, a radiation therapy device, a time-series signal prediction method, a program, and a storage medium that are capable of quickly and accurately predicting time-series signals related to a patient's biological information.
[0008] A time series signal prediction device according to an embodiment includes a time series signal acquisition unit, a block conversion unit, a feature vector acquisition unit, and a feature vector conversion unit. The time series signal acquisition unit acquires a time series signal related to a patient's biological information. The block conversion unit divides the time series signal into partial signals at predetermined intervals and converts the partial signals into block signals having a shorter time series than the time series signal. The feature vector acquisition unit acquires, from the block signals, a feature vector representing a time series signal at a time later than the time when the time series signal was acquired. The feature vector conversion unit converts the feature vector into a prediction signal related to the patient's biological information.
[0009] According to the present invention, it is possible to provide a time series signal prediction device, a radiotherapy device, a time series signal prediction method, a program, and a storage medium that can quickly and accurately predict time series signals related to a patient's biological information.
[0010] 1 is a block diagram showing a schematic configuration of a treatment system including a time-series signal prediction device according to a first embodiment. FIG. 1 is a block diagram showing a schematic configuration of a time-series signal prediction device 100 according to a first embodiment. FIG. 2 is a diagram showing an example of a waveform signal acquired by a time-series signal acquisition unit 110. FIG. 3 is a diagram showing an example of conversion from a time-series signal to a block signal by a block conversion unit 120. FIG. 4 is a diagram showing another example of conversion from a time-series signal to a block signal by the block conversion unit 120. FIG. 5 is a diagram showing an example of acquisition of a feature vector by a feature vector acquisition unit 130. FIG. 6 is a diagram showing another example of acquisition of a feature vector by the feature vector acquisition unit 130. FIG. 7 is a flowchart showing an example of a processing flow executed by the time-series signal prediction device 100 according to the first embodiment. FIG. 8 is a block diagram showing a schematic configuration of a time-series signal prediction device 200 according to a second embodiment. FIG. 9 is a flowchart showing an example of a processing flow executed by the time-series signal prediction device 200 according to the second embodiment. FIG. 10 is a block diagram showing a schematic configuration of a time-series signal prediction device 300 according to a third embodiment. FIG. 11 is a flowchart showing an example of a processing flow executed by the time-series signal prediction device 300 according to the third embodiment.
[0011] Hereinafter, a time-series signal prediction device, a radiotherapy device, a time-series signal prediction method, a program, and a storage medium according to embodiments will be described with reference to the drawings.
[0012] (First Embodiment) (Overall Configuration) Fig. 1 is a block diagram showing the schematic configuration of a treatment system including a time-series signal prediction device according to the first embodiment. The treatment system 1 includes, for example, a treatment device 10, a sensor S, and a time-series signal prediction device 100. The treatment device 10 includes, for example, a bed 12, a computed tomography (CT) device 14 (hereinafter referred to as the "CT device 14"), and a treatment beam irradiation gate 16.
[0013] The bed 12 is a movable treatment table on which a subject (patient) P undergoing radiation therapy is fixed in a lying position, for example, using a fixation device. The bed 12 moves, with the patient P fixed thereto, into a circular CT imaging device 14 having an opening, under the control of a bed control unit (not shown). The bed control unit controls a translation mechanism and a rotation mechanism provided on the bed 12 in accordance with a movement amount signal to change the direction in which the treatment beam B is irradiated onto the patient P fixed to the bed 12. The translation mechanism can drive the bed 12 in three axial directions, and the rotation mechanism can drive the bed 12 around three axes. Therefore, the bed control unit controls, for example, the translation mechanism and rotation mechanism of the bed 12 to move the bed 12 with six degrees of freedom. The degrees of freedom that the bed control unit controls the bed 12 do not have to be six, but may be fewer than six degrees of freedom (e.g., four degrees of freedom) or more than six degrees of freedom (e.g., eight degrees of freedom).
[0014] The CT imaging device 14 is an imaging device for performing three-dimensional computed tomography. The CT imaging device 14 has multiple radiation sources arranged inside a circular opening, and each radiation source emits radiation for viewing the inside of the patient P's body. In other words, the CT imaging device 14 emits radiation from multiple positions around the patient P. The radiation emitted from each radiation source in the CT imaging device 14 is, for example, X-rays. The CT imaging device 14 uses multiple radiation detectors arranged inside the circular opening to detect radiation emitted from the corresponding radiation source that has passed through the body of the patient P and reached the patient P. The CT imaging device 14 generates a CT image of the inside of the patient P's body based on the magnitude of the radiation energy detected by each radiation detector. The CT image of the patient P generated by the CT imaging device 14 is a three-dimensional digital image in which the magnitude of the radiation energy is expressed as a digital value. The three-dimensional imaging of the inside of the body of the patient P in the CT imaging device 14, i.e., the irradiation of radiation from each radiation source and the generation of CT images based on the radiation detected by each radiation detector, is controlled, for example, by an imaging control unit (not shown).
[0015] The treatment beam irradiation gate 16 irradiates radiation as a treatment beam B to destroy a tumor (lesion), which is a treatment target site present inside the body of the patient P. The treatment beam B may be, for example, an X-ray, a gamma ray, an electron beam, a proton beam, a neutron beam, or a heavy particle beam. The treatment beam B is irradiated linearly from the treatment beam irradiation gate 16 onto the patient P (more specifically, the tumor inside the body of the patient P). The irradiation of the treatment beam B at the treatment beam irradiation gate 16 is controlled, for example, by a treatment beam irradiation controller (not shown).
[0016] As described above, radiation therapy is performed by irradiating a tumor inside the body of patient P with treatment beam B from treatment beam irradiation port 16. However, if the tumor, which is the focus of patient P, is located in an organ that moves with breathing or heartbeat, such as the lungs or liver, it is necessary to irradiate the tumor with X-rays from CT imaging device 14 or irradiate the tumor with treatment beam B from treatment beam irradiation port 16 at a timing when the tumor is within the irradiation range of treatment beam B. However, in conventional technology, for example, due to a gap in machine response, the tumor may be irradiated with X-rays or treatment beam B even when the tumor is outside the irradiation range of treatment beam B, resulting in unnecessary exposure of patient P. In light of this situation, the present invention utilizes sensor S and time-series signal prediction device 100 to quickly and accurately predict the timing when the tumor will be within the irradiation range of treatment beam B.
[0017] The sensor S intermittently senses the movement of the body surface of the thorax and abdomen, which expands and contracts in accordance with the breathing of the patient P, for example, and outputs a one-dimensional waveform signal with the obtained sensor value as the amplitude to the time series signal prediction device 100. The sensor S may be a contact type sensor placed on the upper surface of the abdomen of the patient P lying on the bed 12, or a non-contact type sensor that measures the distance to the thorax and abdomen using a laser.
[0018] FIG. 2 is a block diagram showing a schematic configuration of a time-series signal prediction device 100 according to the first embodiment. The time-series signal prediction device 100 includes, for example, a time-series signal acquisition unit 110, a block conversion unit 120, a feature vector acquisition unit 130, and a feature vector conversion unit 140. Some or all of the components included in the time-series signal prediction device 100 are implemented by a hardware processor, such as a central processing unit (CPU), executing a program (software). Some or all of these components may be implemented by hardware (including circuitry), such as a large-scale integration (LSI), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a graphics processing unit (GPU), or may be implemented by a combination of software and hardware. Some or all of the functions of these components may be implemented by a dedicated LSI. The program may be stored in advance in a storage device (storage device having a non-transitory storage medium) such as a read only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), or a flash memory provided in the time series signal prediction device 100, or may be stored in a removable storage medium (non-transitory storage medium) such as a DVD or a CD-ROM, and installed in the HDD or flash memory provided in the time series signal prediction device 100 by inserting the storage medium into a drive device provided in the time series signal prediction device 100. The program may be downloaded from another computer device via a network and installed in the HDD or flash memory provided in the time series signal prediction device 100.
[0019] The time-series signal acquiring unit 110 acquires a time-series signal representing a waveform from the sensor S, for example, by wireless communication. Fig. 3 is a diagram showing an example of a waveform signal acquired by the time-series signal acquiring unit 110. In the graph of Fig. 3, the horizontal axis represents time and the vertical axis represents amplitude. In the waveform shown in Fig. 3, peak P1, which is a mountain, represents the point at which the patient P's inhalation is at its maximum, and peak P2, which is a valley, represents the point at which the patient P's exhalation is at its maximum. The time-series signal acquiring unit 110 outputs the acquired time-series signal to the block converting unit 120.
[0020] As another example, the time-series signal acquired by the time-series signal acquisition unit 110 may be a trajectory representing the position of a tumor that moves with the breathing of the patient P. Here, the trajectory representing the position of the tumor means a time-series signal representing a change over time in the position of the center of gravity of the tumor in a fluoroscopic moving image of the patient P. In this case, the time-series signal acquisition unit 110 may acquire the time-series signal not from the sensor S but from the CT imaging device 14.
[0021] As another example, the time-series signal acquired by the time-series signal acquisition unit 110 may be a time-series signal representing a time change in the diaphragm of the patient P, or a trajectory of a marker placed near a tumor in the patient P's body that moves with breathing. The marker trajectory is a time-series signal representing a time change in the marker position in the patient's fluoroscopic video images. The position of the tumor, diaphragm, and marker (hereinafter, the tracked target) in the fluoroscopic video images is estimated at each time by, for example, template matching using an image containing the tracked target before tracking as a template. Alternatively, the position of the tracked target is estimated at each time based on, for example, feature amounts obtained by converting an image containing the tracked target using a pre-prepared model. The model in this case can be constructed by machine learning, which learns from multiple images containing the tracked target before tracking.
[0022] The block conversion unit 120 divides the time-series signal acquired by the time-series signal acquisition unit 110 into partial signals at predetermined intervals and converts them into block signals with a time series shorter than the time series of the time-series signal. FIG. 4 is a diagram showing an example of conversion from a time-series signal to a block signal by the block conversion unit 120. As an example, FIG. 4 shows a state in which a 12-sequence time-series signal consisting of 12 signal values y(t-11) to y(t) from time t-11 to time t (i.e., the current time) is converted into a 3-sequence block signal. The partial signals in this case are vector data consisting of signal values y(t-3) to y(t) from time t-3 to time t, signal values y(t-7) to y(t-4) from time t-7 to time t-4, and signal values y(t-11) to y(t-8) from time t-11 to time t-8, respectively.
[0023] 5 is a diagram showing another example of conversion from a time-series signal to a block signal by the block conversion unit 120. As an example, FIG. 5 shows how eight sequences of signal values y(t-7) to y(t), each consisting of eight values from time t-7 to time t, are converted into three sequences of block signals. The partial signals in this case are vector data consisting of signal values y(t-3) to y(t) from time t-3 to time t, signal values y(t-5) to y(t-2) from time t-5 to time t-2, and signal values y(t-7) to y(t-4) from time t-7 to time t-4. In this way, signal values may overlap between sequences without changing the order of the time series.
[0024] The feature vector acquisition unit 130 acquires a feature vector representing a time-series signal at a time later than the time of acquisition of the time-series signal, from the block signal converted by the block conversion unit 120. More specifically, in order to predict a time-series signal at a time later from the input time-series signal, the feature vector acquisition unit 130 is configured in advance as a predictor obtained by statistically modeling signal fluctuations from time-series signals relating to biological information collected from a large number of patients.
[0025] One method for modeling the feature vector acquisition unit 130 is to construct a state space model. Specifically, when a pair of two waveform values y(t) and y(t-δ) at time t and time t-δ is plotted on a two-dimensional coordinate system, the plotted points move elliptically with time because the respiratory waveform is periodic. This characteristic is expressed by a state space model. Because this model is a nonlinear model, an extended Kalman filter (EKF), which is an approximation of a Kalman filter corresponding to a nonlinear state space model, is used. The feature vector acquisition unit 130 is configured by optimizing the parameters of this state space model using the collected time-series signals.
[0026] Another method for modeling the feature vector acquisition unit 130 is to use a recurrent neural network. Here, a recurrent neural network is a neural network model that receives multiple sequences of input data in order, transitions the internal state of the model, and obtains an output. A typical example is LSTM (Long Short Term Memory).
[0027] Another method for modeling the feature vector acquisition unit 130 is to use, for example, a feedforward neural network. Here, a feedforward neural network is a neural network model that inputs input data all at once without dividing it into sequences and obtains an output. For example, if the feedforward network is a convolutional neural network (CNN), when a block signal is input, the block signal is processed as two-dimensional data. Alternatively, the sequence of block signals may be input to the network as data on separate channels.
[0028] When using these neural network models, the feature vector acquisition unit 130 is configured by learning the internal parameters of the model through deep learning using the collected time-series signals. In this case, the collected time-series signals are divided into multiple partial signals, allowing overlaps, and each partial signal is divided into an input teaching block signal and a target feature vector. The data included in the target feature vector consists only of signals at a time later than the teaching block signal.
[0029] The learning process involves adjusting the internal parameters so that the feature vector obtained as the output when a teaching block signal is input to the model approaches the target feature vector. A function that measures the distance between the target feature vector and the feature vector is used as the error function. Optimization techniques such as backpropagation are used to minimize the error function.
[0030] 6 is a diagram showing an example of how the feature vector is acquired by the feature vector acquisition unit 130. As shown in FIG. 6, the feature vector is a vector whose elements are, for example, the amplitude values of the partial signals. In other words, when predicting the signal value at time t+M (M>0, M is a natural number) for the teaching block signal up to time t, the target feature vector is a vector whose elements are the signal values y(t+1) to y(t+M) from time t+1 to time t+M.
[0031] 7 is a diagram showing another example of acquisition of a feature vector by the feature vector acquisition unit 130. As shown in FIG. 7, the feature vector may be, for example, a vector whose elements are the differences in amplitude values of the partial signals. In other words, when predicting a signal value y(t+M) at time t+M (M>0, M is a natural number) for a teaching block signal up to time t, the target feature vector is a vector whose elements are the differences d(t+1) to d(t+M) in signal values from time t+1 to time t+M, starting from the teaching block signal at time t.
[0032] The feature vector conversion unit 140 converts the feature vector acquired by the feature vector acquisition unit 130 into a predicted signal related to the biological information of patient P. More specifically, for example, when the feature vector acquisition unit 130 acquires the feature vector shown in FIG. 6 , the feature vector conversion unit 140 acquires the signal value y(t+M) at time t+M from the feature vector as a predicted signal. Furthermore, for example, when the feature vector acquisition unit 130 acquires the feature vector shown in FIG. 7 , the feature vector conversion unit 140 acquires the signal value y(t)+d(t+1)+...d(t+M) obtained by adding the sum of the elements d(t+1) to d(t+M) of the feature vector from time t+1 to time t+M to the signal value y(t) at time t, as the signal value y(t+M) at time t+M. The feature vector conversion unit 140 outputs the calculated prediction signal to a monitor or the like of the time series signal prediction device 100, and the operator of the treatment device 10 checks the prediction signal output to the monitor or the like, determines the timing of capturing the X-ray video image and the timing of irradiating the treatment beam B, and then performs the capturing of the X-ray video image and the irradiation of the treatment beam B.
[0033] Furthermore, the feature vector conversion unit 140 calculates the reliability of the time-series signal based on the feature vector acquired by the feature vector acquisition unit 130. For example, the feature vector conversion unit 140 calculates a smaller reliability value for a time-series signal with a discontinuous element of the feature vector. Here, discontinuity can be determined, for example, by an approximation error between an approximate curve obtained from multiple elements included in the feature vector using a technique such as spline interpolation and these multiple elements. In this case, the feature vector conversion unit 140 may output the predicted signal and the calculated reliability together to a monitor or the like of the time-series signal prediction device 100. In this case, the operator of the treatment device 10 can use the output predicted signal for capturing X-ray motion images and irradiating the treatment beam B, taking into account the reliability of the predicted signal.
[0034] [Processing Flow] Next, a processing flow executed by the time-series signal prediction device of the first embodiment will be described with reference to Fig. 8. Fig. 8 is a flowchart showing an example of a processing flow executed by the time-series signal prediction device of the first embodiment.
[0035] First, the time-series signal acquisition unit 110 acquires a time-series signal related to the patient P's biological information from the sensor S (step S100). Next, the block conversion unit 120 divides the acquired time-series signal into partial signals and acquires them as block signals (step S102). Next, the feature vector acquisition unit 130 acquires a feature vector representing a time-series signal at a future time from the acquired block signal (step S104). Next, the feature vector conversion unit converts the acquired feature vector into a predicted signal related to the patient P's biological information (step S106). This completes the processing of this flowchart.
[0036] According to the first embodiment described above, by converting an input time-series signal into a block signal, the input length is shortened while maintaining the amount of information in the input data. This makes it possible to reduce the time required for training a recurrent neural network model and calculating a feature vector, particularly compared to conventional techniques in which time-series signals are input sequentially. Furthermore, according to the first embodiment, compared to the technique described in Non-Patent Document 1, which teaches only the signal values to be predicted in model training, the first embodiment increases the amount of information to be learned by training multiple signal values up to the time to be predicted. This makes it possible to improve the accuracy of the feature vector output by the predictor.
[0037] Second Embodiment Overall Configuration In the first embodiment, a method for predicting a time series signal at a future time from a time series signal input related to the patient P's biological information was described. As an example of the patient P's biological information, a respiratory signal tends to have irregular amplitude when the patient P inhales completely because the amount of air intake during inhalation varies even for the same patient P. On the other hand, the amplitude when the patient P exhales completely tends to be regular because it is stable. Taking advantage of this tendency, in radiation therapy, it is preferable to avoid capturing X-ray video images and irradiating the treatment beam B while the patient P is inhaling or exhaling, and to set the position of the tumor when the patient P exhales completely as the irradiation spot. In light of this, the time series signal prediction device 200 according to the second embodiment predicts the time when the tumor will enter the irradiation spot, and makes it possible to avoid capturing X-ray video images and irradiating the treatment beam B when an abnormality occurs during inhalation, exhalation, or the patient P's respiratory state.
[0038] 9 is a block diagram showing a schematic configuration of a time-series signal prediction device 200 according to the second embodiment. The time-series signal prediction device 200 includes, for example, a time-series signal acquisition unit 210, a block conversion unit 220, a feature vector acquisition unit 230, a feature vector conversion unit 240, and a state estimation unit 250. The configurations of the time-series signal acquisition unit 210, the block conversion unit 220, the feature vector acquisition unit 230, and the feature vector conversion unit 240 are similar to the configurations of the time-series signal acquisition unit 110, the block conversion unit 120, the feature vector acquisition unit 130, and the feature vector conversion unit 140 of the time-series signal prediction device 100 according to the first embodiment, and therefore description thereof will be omitted. Furthermore, the configuration of the treatment system 1 shown in FIG. 1 is also applied to the second embodiment.
[0039] The state estimation unit 250 estimates the state of the patient P based on the prediction signal acquired from the feature vector conversion unit 240 and outputs the estimated state to a monitor or the like of the time-series signal prediction device 200. Here, the estimated state of the patient refers to a state (abnormal state) that deviates from normal breathing due to, for example, coughing. Before and after coughing or sneezing, breathing becomes irregular and differs from the normal state, so the difference between the predicted value and the actual measured value becomes large. The state estimation unit 250 may output the estimated state of the patient to a monitor or the like of the treatment device 10.
[0040] For example, the state estimation unit 250 may calculate the absolute value of the difference between a predicted value calculated a predetermined time ago (e.g., 100 milliseconds ago) and the actual measured value at that time, and if the absolute value is greater than a predetermined value, estimate and output that the patient's condition is abnormal. Alternatively, for example, the state estimation unit 250 may calculate the absolute value of the difference between a predicted value calculated a predetermined time ago and the actual measured value at that time, and if the absolute value remains greater than the predetermined value for a predetermined period of time, estimate and output that the patient's condition is abnormal. Alternatively, for example, the state estimation unit 250 may calculate the absolute value of the difference between a predicted value calculated a predetermined time ago and the actual measured value at that time for multiple time windows, and if the calculated absolute value for more than half of the time windows is greater than the predetermined value, estimate and output that the patient's condition is abnormal.
[0041] As another example, the patient's state may refer to, for example, the state of the respiratory phase around the time when the patient's breathing changes from exhalation to inhalation, i.e., around the time when the patient completely exhales. If the amplitude of the respiratory waveform increases during inhalation and decreases during exhalation, the time when the amplitude value becomes smaller than a certain threshold is the respiratory phase. This threshold may be set in advance by, for example, a user monitoring the respiratory waveform displayed on a monitor of the time-series signal prediction device 200 and based on the respiratory waveform of the same patient for several previous breaths. For example, the user may set the threshold to a predetermined level (e.g., 20%) of the maximum amplitude of the time-series signal acquired by the time-series signal acquisition unit 110. The past period refers, for example, to the period from the completion of setup, such as patient positioning, to the start of treatment. Alternatively, the state estimation unit 250 may automatically identify the maximum amplitude of the time-series signal acquired by the time-series signal acquisition unit 110 and set the threshold to a predetermined level of the identified maximum amplitude, without requiring the user to set the threshold. In this case, the state estimation unit 250 functions as a "parameter setting unit."
[0042] When the predicted signal represents the trajectory of a tracked object, such as a tumor, diaphragm, or marker, in a fluoroscopic video image, the state estimation unit 250 may estimate and output the state of the tracked object in the fluoroscopic video image. Tracking of a tracked object, such as a tumor, diaphragm, or marker, in a fluoroscopic video image may be difficult due to deterioration in image quality caused by increased noise in the fluoroscopic image. Furthermore, the tracked object may move irregularly due to some abnormality (e.g., coughing or sneezing) in the patient P. Therefore, the state estimation unit 250 may estimate and output an abnormal state for the tracked object when the difference between the predicted value and the actual measured value of the predicted signal becomes large. Furthermore, for example, similar to state estimation for a respiratory waveform, if the time when the change in amplitude of a time-series signal consisting of the trajectory of the tracked object changes from downward to upward is planned to be the time when the tumor is scheduled to enter the irradiation spot, the state estimation unit 250 may set a threshold value in the same manner as described above, and estimate and output a state in which the tumor is within the irradiation spot when the amplitude of the time-series signal is equal to or less than the threshold value.
[0043] In the above description, the state estimation unit 250 is described as being included in the time-series signal prediction device 200, but the state estimation unit 250 may be included in the treatment device 10. In this case, the time-series signal prediction device 200 may transmit the prediction signal acquired from the feature vector conversion unit 240 to the treatment device 10, and the state estimation unit 250 included in the treatment device 10 may estimate the state of the patient P based on the received prediction signal and display it on a monitor of the time-series signal prediction device 200, for example.
[0044] [Processing Flow] Next, a processing flow executed by the time-series signal prediction device of the second embodiment will be described with reference to Fig. 10. Fig. 10 is a flowchart showing an example of a processing flow executed by the time-series signal prediction device of the second embodiment.
[0045] First, the time-series signal acquisition unit 210 acquires a time-series signal related to the patient P's biological information from the sensor S (step S200). Next, the block conversion unit 220 divides the acquired time-series signal into partial signals and acquires them as block signals (step S202). Next, the feature vector acquisition unit 230 acquires a feature vector representing a time-series signal at a future time from the acquired block signal (step S204). Next, the feature vector conversion unit 240 converts the acquired feature vector into a predicted signal related to the patient P's biological information (step S206). Next, the state estimation unit 250 estimates the patient P's state from the acquired predicted signal and outputs it to a monitor or the like of the time-series signal prediction device 200 (step S208). This completes the processing of this flowchart.
[0046] According to the second embodiment described above, the condition of the patient P is estimated from the prediction signal acquired regarding the biological information of the patient P and output to a monitor or the like of the time-series signal prediction device 200. For example, the operator of the treatment device 10 determines the timing of capturing X-ray moving images and irradiating the treatment beam while checking the condition of the patient P output to a monitor or the like, and performs capturing of the X-ray moving images and irradiation of the treatment beam. This makes it possible to reduce exposure to radiation associated with unnecessary capturing of X-ray moving images and irradiation of the treatment beam.
[0047] Third Embodiment In the second embodiment, a method was described in which the condition of a patient P is estimated from a prediction signal acquired regarding the patient P's biological information and output to a monitor or the like of the time-series signal prediction device 200. In this case, the operator of the treatment device 10 manually captures X-ray moving images and irradiates the treatment beam B while checking the condition of the patient P output to a monitor or the like, but this can lead to human operation errors and discrepancies in the execution timing. In light of this situation, a time-series signal prediction device 300 according to the third embodiment automatically controls the treatment device 10 based on the estimated condition of the patient P.
[0048] 11 is a block diagram showing a schematic configuration of a time-series signal prediction device 300 according to the third embodiment. The time-series signal prediction device 300 includes, for example, a time-series signal acquisition unit 310, a block conversion unit 320, a feature vector acquisition unit 330, a feature vector conversion unit 340, a state estimation unit 350, and a treatment device control unit 360. The configurations of the time-series signal acquisition unit 310, the block conversion unit 320, the feature vector acquisition unit 330, the feature vector conversion unit 340, and the state estimation unit 350 are similar to the configurations of the time-series signal acquisition unit 210, the block conversion unit 220, the feature vector acquisition unit 230, the feature vector conversion unit 240, and the state estimation unit 250 of the time-series signal prediction device 200 according to the second embodiment, and therefore description thereof will be omitted. Furthermore, the configuration of the treatment system 1 shown in FIG. 1 is also applied to the third embodiment.
[0049] The treatment device control unit 360 controls the treatment device 10 based on the state of the patient P estimated by the state estimation unit 350. More specifically, when the state estimation unit 350 determines that the state of the patient P is in a respiratory phase near the time of complete exhalation (in other words, when the amplitude value is smaller than a certain threshold), the treatment device control unit 360 starts capturing X-ray moving images of the patient P and / or preparing to irradiate the treatment beam B from the treatment device 10. Furthermore, for example, when the state estimation unit 350 estimates that a tumor in the fluoroscopic moving images is within the irradiation spot, the treatment device control unit 360 may shorten the capturing interval of the X-ray moving images and start preparing to irradiate the treatment beam B from the treatment device 10.
[0050] On the other hand, if the state estimation unit 350 determines that the state of the patient P is abnormal, the treatment device control unit 360 stops or shortens the interval of capturing X-ray moving images of the patient P and / or irradiating the treatment beam B from the treatment device 10. For example, if the state estimation unit 350 estimates that the patient P has exhaled completely and then inhaled again, the treatment device control unit 360 may stop capturing X-ray moving images of the patient P and / or irradiating the treatment beam B from the treatment device 10. For example, if the state estimation unit 350 estimates that the trajectory of a tracked object, such as a tumor, diaphragm, or marker, in the fluoroscopic moving images is abnormal, the treatment device control unit 360 may stop capturing X-ray moving images of the patient P and / or irradiating the treatment beam B from the treatment device 10.
[0051] In the above description, the treatment device control unit 360 is described as being included in the time-series signal prediction device 300, but the treatment device control unit 360 may be included in the treatment device 10. In this case, the time-series signal prediction device 300 may transmit the condition of the patient P estimated by the condition estimation unit 350 to the treatment device 10, and the treatment device control unit 360 included in the treatment device 10 may control the treatment device 10 based on the received condition of the patient P.
[0052] [Processing Flow] Next, a processing flow executed by the time-series signal prediction device of the third embodiment will be described with reference to Fig. 12. Fig. 12 is a flowchart showing an example of a processing flow executed by the time-series signal prediction device of the third embodiment.
[0053] First, the time-series signal acquisition unit 310 acquires a time-series signal related to the patient P's biological information from the sensor S (step S300). Next, the block conversion unit 320 divides the acquired time-series signal into partial signals and acquires them as block signals (step S302). Next, the feature vector acquisition unit 330 acquires a feature vector representing a time-series signal at a future time from the acquired block signal (step S304). Next, the feature vector conversion unit 340 converts the acquired feature vector into a predicted signal related to the patient P's biological information (step S306). Next, the state estimation unit 350 estimates the patient P's state from the acquired predicted signal (step S308). Next, the treatment device control unit 360 controls the treatment device 10 based on the estimated state of the patient P. This completes the processing of this flowchart.
[0054] According to the third embodiment described above, the treatment device 10 is controlled based on the estimated state of the patient P to perform or stop capturing X-ray moving images of the patient P and / or irradiating the treatment beam B from the treatment device 10. This makes it possible to further reduce exposure to radiation due to unnecessary capturing of X-ray moving images and irradiation of the treatment beam.
[0055] Although several embodiments of the present invention have been described above, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as defined in the claims.
[0056] The time series signal prediction device, radiotherapy device, time series signal prediction method, program, and storage medium according to the present invention employ the following configuration: (1): A time series signal prediction device according to one aspect of the present invention includes a time series signal acquisition unit that acquires a time series signal related to a patient's biological information, a block conversion unit that divides the time series signal into partial signals at predetermined intervals and converts the partial signals into block signals of a shorter time series than the time series of the time series signal, a feature vector acquisition unit that acquires, from the block signals, feature vectors that represent time series signals at times in the future than the time when the time series signal was acquired, and a feature vector conversion unit that converts the feature vectors into prediction signals related to the patient's biological information.
[0057] (2) In the aspect (1) above, the time-series signal prediction device further includes a state estimation unit that predicts the state of the patient from the prediction signal.
[0058] (3): In the aspect (2), the time-series signal prediction device further includes a parameter setting unit that adjusts parameters used for prediction in the state estimation unit based on the time-series signal and the prediction signal.
[0059] (4) In the above aspects (1) to (3), the plurality of block signals include overlapping signals at a certain time.
[0060] (5) In the above aspects (1) to (4), the feature vector conversion unit calculates the reliability of the early prediction signal from the feature vector.
[0061] (6) In the above aspects (1) to (5), the time series signal is a signal that changes according to the breathing of the patient, and the state is an abnormal state of the breathing movement of the patient.
[0062] (7): In the above aspects (1) to (5), the time series signal is a signal that changes in accordance with the breathing of the patient, and the state is the timing at which the breathing of the patient changes from exhalation to inhalation.
[0063] (8): In the above aspects (1) to (5), the time series signal is a signal that changes depending on the position of the tumor in the patient, and the predicted signal is a predicted position of the tumor in the patient.
[0064] (9): A radiation therapy device according to one aspect of the present invention controls the timing of taking fluoroscopic images of the patient based on the state predicted by the time series signal prediction device according to aspect (2) above.
[0065] (10): A radiation therapy device according to one aspect of the present invention controls the timing of irradiation of a treatment beam to the patient based on the state predicted by the time series signal prediction device according to aspect (2) above.
[0066] (11): A time series signal prediction method according to one aspect of the present invention involves a computer acquiring a time series signal relating to a patient's biometric information, dividing the time series signal into partial signals at predetermined intervals, converting the partial signals into block signals of a time series shorter than the time series of the time series signal, acquiring from the block signals a feature vector representing a time series signal at a time in the future relative to the time of acquisition of the time series signal, and converting the feature vector into a prediction signal relating to the patient's biometric information.
[0067] (12): A program according to one aspect of the present invention causes a computer to acquire a time series signal relating to a patient's biometric information, divide the time series signal into partial signals at predetermined intervals, convert the time series signal into block signals of a time series shorter than the time series of the time series signal, acquire from the block signals a feature vector representing a time series signal at a time in the future relative to the time when the time series signal was acquired, and convert the feature vector into a prediction signal relating to the patient's biometric information.
[0068] (13): A storage medium according to one aspect of the present invention stores a program that causes a computer to acquire a time series signal relating to a patient's biometric information, divide the time series signal into partial signals at predetermined intervals, convert the time series signal into block signals of a shorter time series than the time series of the time series signal, acquire from the block signals a feature vector representing a time series signal at a time in the future relative to the time when the time series signal was acquired, and convert the feature vector into a predictive signal relating to the patient's biometric information.
[0069] 1... treatment system, 10... treatment device, 12... bed, 14... CT imaging device, 16... treatment beam irradiation port, 100, 200, 300... time series signal prediction device, 110, 210, 310... time series signal acquisition unit, 120, 220, 320... block conversion unit, 130, 230, 330... feature vector acquisition unit, 140, 240, 340... feature vector conversion unit, 250, 350... state estimation unit, 360... treatment device control unit
Claims
1. A time series signal prediction device comprising: a time series signal acquisition unit that acquires a time series signal related to a patient's bioinformation; a block conversion unit that divides the time series signal into partial signals at predetermined intervals and converts them into block signals of a time series shorter than the time series of the time series signal; a feature vector acquisition unit that acquires, from the block signals, a feature vector that represents a time series signal at a time in the future than the time of acquisition of the time series signal; and a feature vector conversion unit that converts the feature vector into a prediction signal related to the patient's bioinformation.
2. The time-series signal prediction device according to claim 1, further comprising a state estimation unit that predicts a state of the patient from the prediction signal.
3. The time series signal prediction device according to claim 2, further comprising a parameter setting section that adjusts parameters used for prediction in said state estimation section based on said time series signal and said prediction signal.
4. The time series signal prediction device according to claim 1, wherein the plurality of block signals include overlapping signals at a certain time.
5. The time series signal prediction device according to claim 1, wherein the feature vector conversion unit calculates a reliability of the previous predicted signal from the feature vector.
6. The time series signal prediction device according to claim 2 or 3, wherein the time series signal is a signal that changes according to the respiration of the patient, and the state is an abnormal state of the patient's respiratory movement.
7. A time series signal prediction device according to claim 2 or 3, wherein the time series signal is a signal that changes in response to the breathing of the patient, and the state is the timing at which the breathing movement of the patient changes from exhalation to inhalation.
8. A time series signal prediction device according to any one of claims 1 to 5, wherein the time series signal is a signal that changes depending on a position of a tumor in the patient, and the predicted signal is a predicted position of a tumor in the patient.
9. A radiotherapy device which controls the timing of taking fluoroscopic images of the patient based on the state predicted by the time-series signal prediction device according to claim 2.
10. A radiotherapy device which controls the timing of irradiating the patient with a therapeutic beam based on the state predicted by the time-series signal prediction device according to claim 2.
11. A time series signal prediction method, comprising: a computer acquiring a time series signal relating to a patient's biometric information; dividing the time series signal into partial signals at predetermined intervals and converting the partial signals into block signals of a time series shorter than the time series of the time series signal; acquiring from the block signals a feature vector representing a time series signal at a time in the future relative to the time when the time series signal was acquired; and converting the feature vector into a prediction signal relating to the patient's biometric information.
12. A program which causes a computer to acquire a time series signal relating to a patient's biometric information, divide the time series signal into partial signals at predetermined intervals and convert them into block signals of a time series shorter than the time series of the time series signal, acquire from the block signals a feature vector which represents a time series signal at a time in the future relative to the time when the time series signal was acquired, and convert the feature vector into a predictive signal relating to the patient's biometric information.
13. A storage medium storing a program which causes a computer to acquire a time series signal relating to a patient's biometric information, divide the time series signal into partial signals at predetermined intervals and convert them into block signals of a time series shorter than the time series of the time series signal, acquire from the block signals a feature vector which represents a time series signal at a time in the future than the time when the time series signal was acquired, and convert the feature vector into a predictive signal relating to the patient's biometric information.
Citation Information
Patent Citations
Four-dimensional dose calculation method and system and storage medium
CN113096724A
Multi-session breathing guidance
EP4176796A1
Lung volume-gated x-ray imaging system and method
JP2021526949A
Beam path based patient positioning and monitoring
WO2020020623A1
Cited By
Traditional Chinese medicine and western medicine combined whole-course nursing management system for tumor patients
CN120108728A
