JP2021112471A - Image condition output device, and radiotherapy treatment device - Google Patents

Abstract

To reflect the an imaging condition to a learning data and reduce time for an operator to determine an appropriate imaging condition.SOLUTION: An imaging condition output device according to the embodiment includes an obtaining part and a condition output part. The obtaining part obtains the subject information related to the subject in a radiotherapy treatment. The condition output part outputs the imaging condition by inputting the subject information to the leaned model for outputting the imaging condition of a cone beam CT image used for positioning of the subject based on the subject information.SELECTED DRAWING: Figure 2

Description

Embodiments of the present invention relate to an imaging condition output device and a radiotherapy device.

In the field of radiotherapy, a wide range of radiotherapy devices are equipped with a medical linear accelerator mounted on a bed to irradiate a patient and an imaging device used to align the patient before irradiating the patient. Are known. In this type of radiotherapy device, the imaging device is based on the cone-beam CT (Cone-Beam Computed Tomography) imaging condition (hereinafter, also referred to as CBCT imaging condition) set by the operator, and the imaging device is a cone-beam CT image of the patient (hereinafter, also referred to as CBCT imaging condition). Hereinafter, it is also referred to as a CBCT image). The captured CBCT image is compared with a planned CT image, which is a CT image for treatment planning. As a result of the comparison, the positional relationship between the imaging device and the bed is adjusted as appropriate, and the patient is aligned.

From the viewpoint of improving the accuracy of alignment, such a CBCT image preferably has high image quality like the planned CT image. Generally, as a method for improving the image quality of a CBCT image, an image processing technique for reducing noise by using an algorithm such as a neural network is known. For example, during image processing, a low-quality CBCT image due to noise is input to the neural network, and a CBCT image with reduced noise is output from the neural network. Such a neural network can be realized by machine learning learning data including a low-quality CBCT image and a noise-reduced CBCT image in advance.

However, although there is usually no particular problem in this kind of image processing technique, according to the study of the present inventor, there are the following inconveniences. For example, while it takes time and effort for the operator to determine the imaging conditions of the CBCT image constituting the learning data, the operator does not know the appropriate imaging conditions. In addition, the tendency of noise and scattered rays to be generated due to the shooting conditions is latent in the learning data, and the shooting conditions are not reflected in the learning data.

Special Table 2019-506208

Ian Goodfellow, 2 outsiders, "Chapter 9 Convolutional Networks", Deep Learning, MIT Press, 2016, p. 330-372, Internet <URL: http://www.deeplearningbook.org>

The problem to be solved by the present invention is that the shooting conditions can be reflected in the learning data, and the time and effort required for the operator to determine the appropriate shooting conditions can be reduced.

The photographing condition output device according to the embodiment includes an acquisition unit and a condition output unit.
The acquisition unit acquires subject information regarding a subject in radiotherapy.
The condition output unit inputs the subject information to the trained model for outputting the shooting conditions of the cone beam CT image used for aligning the subject based on the subject information, thereby performing the shooting. Output the condition.

FIG. 1 is a block diagram showing a configuration of a radiotherapy system according to the present embodiment. FIG. 2 is a block diagram showing a configuration of a shooting condition output device according to the present embodiment. FIG. 3 is a schematic diagram for explaining the processing circuit and the evaluation function according to the present embodiment. FIG. 4 is a schematic diagram for explaining photography-related information and learning data according to the present embodiment. FIG. 5 is a schematic diagram for explaining the CBCT imaging conditions according to the present embodiment. FIG. 6 is a schematic diagram for explaining the learning data and the trained model according to the present embodiment. FIG. 7 is a schematic diagram for explaining the learning data and the trained model according to the present embodiment. FIG. 8 is a schematic diagram for explaining the learning data and the trained model according to the present embodiment. FIG. 9 is a flowchart for explaining the operation of the learning stage in the present embodiment. FIG. 10 is a schematic diagram for explaining the operation of step ST20 in this embodiment. FIG. 11 is a flowchart for explaining the operation of the operation stage in the present embodiment. FIG. 12 is a schematic diagram for explaining the learning data and the learned model according to the first modification of the present embodiment. FIG. 13 is a schematic diagram for explaining reinforcement learning according to the third modification of the present embodiment.

Hereinafter, embodiments will be described with reference to the drawings. In the following description, the patient during cone beam CT imaging (hereinafter, also referred to as CBCT imaging) is referred to as a subject, but the subject is not limited to this and may be referred to as a patient. Similarly, the information that affects CBCT imaging among the patient information is referred to as subject information, but the subject information is not limited to this and may be referred to as patient information. The term "cone beam CT" is also referred to as "CBCT".

As shown in FIG. 1, the radiotherapy system 100 includes an image storage device 1, an imaging condition output device 2, a treatment planning CT device 3, a treatment planning device 4, and a radiotherapy device 5. The radiotherapy device 5 includes an imaging device 51. The image storage device 1, the imaging condition output device 2, the treatment planning CT device 3, the treatment planning device 4, and the radiotherapy device 5 are communicably connected to each other via a network.

The image storage device 1 stores a plurality of past CBCT images having different imaging conditions from each other. Further, the image storage device 1 transmits the plurality of past CBCT images to the imaging condition output device 2 based on the request from the imaging condition output device 2.

The imaging condition output device 2 is, for example, a device that outputs the imaging conditions of the cone beam CT image used for aligning the subject to the radiotherapy device 5 and the image capturing device 51 before the treatment based on the subject information. .. Supplementally, in the image capturing apparatus 51 for photographing and confirming the irradiation position, important factors that affect the accuracy of treatment are the image quality of the cone beam CT image and the imaging conditions that affect the image quality. Therefore, the shooting condition output device 2 uses a trained model trained by a combination of subject information and CBCT shooting conditions in order to support the optimum shooting conditions of the image shooting device 51. The shooting condition output device 2 may generate a trained model, or may install the trained model from the outside. However, the shooting condition output device 2 is not limited to these examples. For example, the imaging condition output device 2 is not limited to the external device connected to the radiotherapy device 5, and may be provided as a device included in the radiotherapy device 5. The configuration of the shooting condition output device 2 and the like will be described later.

The treatment planning CT device 3 performs CT imaging on the patient to be treated and generates a CT image to be used for the treatment planning. Hereinafter, the CT image to be used for the treatment plan is referred to as a planned CT image. The planned CT image may be a two-dimensional image composed of pixels arranged in a two-dimensional manner, or a three-dimensional image composed of voxels arranged in a three-dimensional manner. The treatment planning CT device 3 may be any modality device capable of generating a planned CT image. Examples of the modality device include an X-ray computed tomography device, a cone beam CT device, an X-ray diagnostic device, and the like. The data of the planned CT image is transmitted to, for example, the treatment planning device 4.

The treatment planning device 4 is a computer including a processor such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit), a memory such as a ROM (Read Only Memory) and a RAM (Random Access Memory), a display, an input interface, and a communication interface. Has. The treatment planning device 4 creates a treatment plan for the patient to be treated by using the planned CT image. The parameters included in the treatment plan include the number of fields (number of irradiation fields), field angle (irradiation angle), radiation intensity, collimator opening, and the like. The treatment planning device 4 identifies the position and shape of the tumor or organ based on the planned CT image, and determines various treatment planning parameters. At this time, a treatment plan is created so that the dose to be irradiated to the tumor is as high as possible and the dose to normal tissue is as small as possible. The treatment plan data is supplied to the radiotherapy device 5 via the imaging condition output device 2.

The radiation therapy device 5 treats the patient by irradiating the patient with radiation according to the treatment plan created by the treatment planning device 4. The radiotherapy device 5 has a medical linear accelerator provided in the treatment room, an imaging device 51, and a treatment bed. The treatment bed moves the top plate so that the treatment site of the patient is approximately aligned with the isocenter. The medical linear accelerator includes a treatment stand that supports the irradiation head portion so as to be rotatable around a rotation axis. The irradiation head unit irradiates radiation according to the treatment plan created by the treatment planning device 4. Specifically, the irradiation head portion forms an irradiation field with a multi-segment diaphragm (multi-leaf collimator), and the irradiation field suppresses irradiation to normal tissue. When the treatment site is irradiated with radiation, the treatment site disappears or shrinks. The imaging device 51 is mounted on the treatment stand. The imaging device 51 is for collecting internal information of the patient for confirming the position at the time of radiotherapy. For example, the image capturing device 51 has an X-ray tube and an X-ray detector arranged so as to face each other with the rotation axis interposed therebetween. With this configuration, the image capturing apparatus 51 can realize cone beam CT and generate three-dimensional CT image data that depicts the morphology of the patient's body. Further, the radiotherapy device 5 may include an irradiation unit that irradiates the subject with radiation, and an imaging condition output device 2 that outputs imaging conditions of a cone beam CT image used for aligning the subject. The irradiation head portion is an example of the irradiation portion.

As shown in FIG. 2, the photographing condition output device 2 includes a processing circuit 21, a memory 22, an input interface 23, a communication interface 24, and a display 25.

The processing circuit 21 has a processor such as a CPU and a GPU. When the processor activates a program installed in the memory 22 or the like, the processor realizes the acquisition function 21a, the evaluation function 21b, the learning function 21c, the condition output function 21d, and the display control function 21e. The functions 21a to 21e are not limited to the case where they are realized by a single processing circuit. A processing circuit may be configured by combining a plurality of independent processors, and each function 21a to 21e may be distributed and realized by executing a program by each processor.

With the realization of the acquisition function 21a, the processing circuit 21 acquires various data. For example, the processing circuit 21 acquires the treatment plan data and patient information received from the treatment planning device 4 via the communication interface 24. Further, for example, the processing circuit 21 acquires subject information regarding a subject in radiotherapy as information selected from patient information and treatment plan data according to the operation of the input interface 23 by the operator. The subject information is information that affects the cone-beam CT imaging conditions in the patient information and the treatment plan data, but is not limited to this, and may include information that does not affect the cone-beam CT imaging conditions. Specifically, for example, the subject information includes the treatment site of the subject, the fixture information, and the physique information. The physique information includes the weight or thickness of the subject, and may further include height, age and gender. Further, the physique information may include an index of physique such as BMI (Body Mass Index) calculated from the height and weight of the subject. The fixture information is information representing a fixture that fixes the subject to the bed, and may be, for example, a product name of the fixture. The fixture may be located on an X-ray path for capturing a CBCT image. In addition, the subject information may further include the therapeutic position, medical history, internal tissue structure, and the like as appropriate. Further, for example, the processing circuit 21 may acquire a plurality of past CBCT images received from the image storage device 1 via the communication interface 24. The plurality of CBCT images in the past are images taken under different shooting conditions. Further, for example, the processing circuit 21 may acquire a planned CT image of the subject received from the treatment planning device 4 via the communication interface 24. Further, for example, the processing circuit 21 may acquire the imaging conditions of the planned CT image of the subject received from the treatment planning device 4 via the communication interface 24. The acquisition function 21a and the processing circuit 21 are examples of the acquisition unit.

With the realization of the evaluation function 21b, the processing circuit 21 obtains the evaluation result of the quality of the cone beam CT image. In this case, as shown in FIG. 3, the processing circuit 21 may obtain the evaluation result d3 of each of the plurality of past CBCT images d1 using the planned CT image d2 as the correct answer data. In this case, the processing circuit 21 calculates an evaluation value from, for example, the similarity between the feature amount of the planned CT image d2 and the feature amount of the CBCT image d1, a histogram reflecting the contrast of each CT image, and the evaluation value. The evaluation result may be determined from. Here, the feature amount of the planned CT image d2 and the feature amount of the CBCT image d1 are, for example, among the image data, the pixel value, the statistical value of the pixel value, and the morphological feature amount of the tumor region to be treated with radiotherapy. At least one of. The statistical value of the pixel value is at least one of the feature amounts related to statistics such as the average value, the maximum value, the minimum value, the intermediate value, the standard deviation, and the dispersion value of the pixel values of the plurality of pixels constituting the tumor region. The morphological feature amount is at least one of the feature amounts related to the morphology such as the volume and surface area of the tumor region, the number of cavities, and the degree of shape. The degree of shape is at least one of the features related to the shape such as the circularity, sharpness, and the number of irregularities of the tumor region.

Further, the processing circuit 21 may obtain the evaluation result d3 of each of the plurality of cone beam CT images d1 according to the operation of the input interface 23 by the operator. The evaluation function 21b and the processing circuit 21 are examples of the evaluation unit.

With the realization of the learning function 21c, the processing circuit 21 creates the learning data d10 from the shooting-related information d0 as shown in FIG. The imaging-related information d0 includes a plurality of CBCT images d1, a planned CT image d2, a plurality of evaluation results d3, subject information d4, and a plurality of CBCT imaging conditions d5. The learning data d10 includes subject information d4 and CBCT imaging condition d5, which is correct answer data, among the imaging-related information d0. However, the present invention is not limited to this, and the learning data d10 may further include the planned CT image d2 and the evaluation result d3. Further, instead of the planned CT image d2, the imaging conditions of the planned CT image d2 may be included in the learning data d10. Further, in FIG. 4, there are n CBCT images d1, evaluation result d3, and CBCT imaging condition d5 with the notation "(1, ..., N)" relating to a plurality of information for each subject information. Is shown. Since the shooting-related information d0 exists for each subject information d4, a plurality of the CBCT image d1, the evaluation result d3, and the CBCT shooting condition d5 also exist for each subject information d4. Since the learning data d10 is created for a plurality of subject information, a plurality of learning data d10 are created as shown in FIG. These plurality of learning data d10 are used in the learning stage. Such a plurality of learning data d10 may be referred to as a learning data set.

Here, the various data shown in FIG. 4 have the following contents.

n CBCT images d1: CBCT images with different shooting conditions.
Plan CT image d2: An image of the treatment site taken in the treatment position at the time of treatment planning (data close to the correct answer).
n evaluation results d3: Results indicating the quality of the generated CBCT image. The evaluation result may be expressed by an alternative such as good (OK) / bad (NG), or may be expressed by a quantitative numerical value such as a value of a plurality of stages such as 10 stages. Further, the evaluation result may be obtained for each viewpoint for evaluating the contrast of each image, the shape of the organ, and the like.
Subject information d4: Treatment site, treatment position, medical history, internal tissue structure, fixture information, physique information.
n CBCT imaging conditions d5: gantry angle, rotation range, rotation direction, aperture type, filter type, imaging range (FOV size), tube voltage, tube current (or tube current time product), frame rate, Shooting conditions such as the distance to the detector (the image shooting device 51 is fixed), the position of the sleeper, and so on. As for the CBCT imaging condition d5, as shown in FIG. 5, each combination of conditions can be used as each protocol. In this case, the number of combinations of the CBCT imaging condition d5 is less than or equal to the number of protocols that can be used. Further, since the elements included in the CBCT imaging condition d5 are the conditions used in the image capturing apparatus 51, they are not necessarily limited to the elements that affect the image quality of the CBCT image. For example, the rotation range and the rotation direction included in the CBCT imaging condition d5 are constraint conditions of the image capturing apparatus 51 and do not affect the image quality of the CBCT image.

Further, in order to determine the correct answer data among the n CBCT imaging conditions d5, for example, there is a method in which the imaging condition of the CBCT image showing a good evaluation result is used as the correct answer data. Further, as the evaluation result for using the CBCT image capturing condition d5 as the correct answer data, for example, an evaluation result that the contrast is good and the shape of the organ can be easily discriminated, an evaluation result that the influence of noise and scattered rays is small, and the like. Can be used as appropriate.

Further, by realizing the learning function 21c, as shown in FIG. 6, the processing circuit 21 performs machine learning on the machine learning model M1 using the created learning data d10, so that the machine learning model M1 is used to learn the trained model M2. To create. In FIG. 6, the learning data d10 includes the subject information d4 in the input data and the CBCT imaging condition d5 in the output data. However, the learning data d10 is not limited to this, and other data may be added as long as the subject information d4 is included in the input data and the CBCT imaging condition d5 is included in the output data. For example, as the learning data d10, as shown in FIG. 7, the evaluation result d3 may be further included in the output data, as compared with the configuration shown in FIG. The evaluation result d3 is not limited to the output data of the learning data d10, and may be included in the input data of the learning data d10. Further, for example, as the learning data d10, as shown in FIG. 8, the planned CT image capturing condition d2a may be further included in the input data as compared with the configuration shown in FIG. Alternatively, as the learning data d10, the planned CT image d2 may be further included in the input data as compared with the configuration shown in FIG. Alternatively, as compared with the configuration shown in FIG. 6, the evaluation result d3 may be further included in the input data or the output data with respect to the learning data d10 in which the planned CT image capturing condition d2a and the planned CT image d2 are further included in the input data. good.

As described above, each learning data d10 has input data and output data. The input data is the data input to the input layer of the machine learning model M1 among the learning data d10. The output data is data other than the input data in the learning data d10.

The machine learning model M1 is a composite function with parameters in which a plurality of functions are synthesized, which takes the input data as an input and outputs the output data. A parameterized composition function is defined by a combination of multiple adjustable functions and parameters. The machine learning model M1 according to the present embodiment may be a composite function with any parameter that satisfies the above requirements, but is a multi-layer network model (hereinafter referred to as a multi-layer network). The trained model M2 using the multi-layer network has at least one layer provided between the input layer for inputting the subject information d4, the output layer for outputting the shooting condition d5 of the CBCT image, and the input layer and the output layer. It has an intermediate layer. As the input layer, the input of the evaluation result d3, the planned CT image d2, and the shooting condition d2a of the planned CT image may be accepted in parallel with the input of the subject information d4. These evaluation results d3, planned CT image d2, and imaging condition d2a of the planned CT image are information related to CBCT imaging, but are not essential and are optional additional items. As such a multi-layer network, for example, a deep neural network (DNN), which is a multi-layer neural network targeted for deep learning, is used. As the DNN, for example, a convolutional neural network (CNN) may be used. This type of CNN is also referred to as DCNN (Deep CNN). In this embodiment, DCNN is used as the trained model. Regarding CNN, for example, "Ian Goodfellow, 2 outsiders," Chapter 9 Convolutional Networks ", Deep Learning, MIT Press, etc. 2016, p. 330-372, Internet <URL: http://www.deeplearningbook.org> ”. The above description of the multi-layer network also applies to all the following machine learning models and trained models.

With the realization of the condition output function 21d, the processing circuit 21 inputs the subject information d4 to the learned model M2 for outputting the shooting condition d5 of the CBCT image used for the alignment of the subject based on the subject information d4. As a result, the shooting condition d5 is output. Here, the processing circuit 21 inputs the subject information d4 to the trained model M2 that outputs the shooting condition d5 of the CBCT image indicating that the evaluation result d3 is good based on the subject information d4, thereby inputting the shooting condition. You may output d5. Further, the processing circuit 21 inputs the subject information d4 and the planned CT image shooting condition d2a to the trained model M2 that outputs the shooting condition d5 of the CBCT image based on the subject information d4 and the planned CT image shooting condition d2a. By doing so, the shooting condition d5 may be output. Further, the processing circuit 21 inputs the subject information d4 and the planned CT image d2 to the trained model M2 that outputs the shooting condition d5 of the CBCT image based on the subject information d4 and the planned CT image d2. The shooting condition d5 may be output. The condition output function 21d and the processing circuit 21 are examples of the condition output unit.

With the realization of the display control function 21e, the processing circuit 21 displays various information on the display 25. For example, the processing circuit 21 controls the display 25 so as to display the above-mentioned shooting-related information, learning data, and the like. The display control function 21e and the processing circuit 21 are examples of the display control unit.

The memory 22 is a storage device such as a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), an SSD (Solid State Drive), or an integrated circuit storage device that stores various information. In addition to the above storage device, the memory 22 is a drive device that reads and writes various information between a portable storage medium such as a CD (Compact Disc), a DVD (Digital Versatile Disc), and a flash memory, and a semiconductor memory element. It may be. Further, the memory 22 may be in another computer connected to the shooting condition output device 2 via a network. Further, the memory 22 stores data such as a program, subject information, treatment plan, imaging-related information, learning data, machine learning model, learned model, and various tables.

Supplementally, the memory 22 stores the machine learning model M1 and the trained model M2 as described above. However, even when the processing circuit 21 that realizes the condition output function 21d uses the trained model M2, the storage of the machine learning model M1 and the learning function 21c are not essential and may be omitted. This is because when the trained model M2 is used, the shooting condition output device 2 does not need to generate the trained model M2 by learning the machine learning model M1, and the trained model acquired from the outside may be used. .. For example, the memory 22 may store a trained model in advance before the shooting condition output device 2 is shipped from the factory, or may store a learned model acquired from a server device (not shown) after the shooting condition output device 2 is shipped from the factory. You may remember. These are the same even when the radiotherapy device 5 includes the imaging condition output device 2. The memory 22 is an example of a storage unit.

The input interface 23 receives various input operations from the operator, converts the received input operations into electric signals, and outputs the received input operations to the processing circuit 21. Specifically, the input interface 23 is connected to an input device such as a mouse, keyboard, trackball, switch, button, joystick, touch pad, and touch panel display. The input interface 23 outputs an electric signal corresponding to an input operation to the input device to the processing circuit 21. Further, the input device connected to the input interface 23 may be an input device provided in another computer connected via a network or the like.

The communication interface 24 is an interface for data communication with other devices included in the radiotherapy system 100. For example, the communication interface 24 receives patient information, treatment plan, and planned CT image data from the treatment planning device 4 via a network. Further, for example, the communication interface 24 receives the data of the past CBCT image from the image storage device 1 via the network.

The display 25 displays various information according to the display control function 21e of the processing circuit 21. For example, the display 25 displays an evaluation screen of a CBCT image. As the display 25, for example, a liquid crystal display (LCD), a CRT (Cathode Ray Tube) display, an organic EL display (OELD), a plasma display, or any other display can be appropriately used. be. The display 25 is an example of a display unit.

Next, the operation of the photographing condition output device 2 configured as described above will be described with reference to the drawings of FIGS. 9 to 11. More specifically, the operation of the learning stage will be described with reference to the flowchart of FIG. 9 and the schematic diagram of FIG. 10, and the operation of the operation stage will be described with reference to the flowchart of FIG.

[Learning stage operation]
For each CBCT image taken before radiotherapy, the processing circuit 21 transfers the past CBCT image to the image storage device 1 via the communication interface 24 in response to the operator's operation on the input interface 23, as shown in FIG. Keep it (step ST10). At this time, the CBCT imaging condition is stored together with the CBCT image as incidental information, for example.

After step ST10, the processing circuit 21 reads out a plurality of past CBCT images from the image storage device 1 via the communication interface 24 in response to the operator's operation on the input interface 23. Each CBCT imaging condition is read out together with a plurality of CBCT images as incidental information.

After that, the processing circuit 21 evaluates each of the plurality of past CBCT images and obtains the evaluation result of each CBCT image (step ST20). At the time of evaluation in step ST20, for example, as shown in FIG. 10, the input screen of the evaluation result d3 may be displayed on the display 25. Specifically, for example, an input screen including the identification number d1a of the past CBCT image d1, the CBCT image d1 corresponding to the identification number d1a, the decrease button bt1 and the increase button bt2 of the identification number d1a, and the evaluation result d3. It may be displayed. While displaying such an input screen, the processing circuit 21 inputs the evaluation result d3 of each CBCT image d1 according to the operation of the operator with respect to the input interface 23.

Alternatively, at the time of evaluation in step ST20, the processing circuit 21 calculates the evaluation value by arithmetic processing using, for example, the feature amount of the planned CT image d2 and the feature amount of the CBCT image d1, and a plurality of evaluation values are used. The evaluation result of the CBCT image may be obtained by classifying into stages. In any case, step ST20 is completed by obtaining the evaluation result of each CBCT image.

After step ST20, the processing circuit 21 creates learning data d10 in response to an operator's operation on the input interface 23 (step ST30). The learning data d10 includes subject information d4 and CBCT imaging condition d5 as illustrated in FIGS. 4, 6, 7, and 8, and further includes an evaluation result d3, a planned CT image d2, and a plan. It is created by appropriately including the imaging conditions d2a of the CT image and the like. If the learning data d10 does not include the evaluation result d3, the above-mentioned step ST20 can be omitted. Further, in step ST30, not only the subject information d4 corresponding to the same patient but also a large number of learning data d10 are created for a large number of subject information d4 corresponding to a large number of patients.

After step ST30, the processing circuit 21 causes the machine learning model M1 to perform machine learning on a large number of learning data d10 created in step ST30. Upon completion of machine learning, the processing circuit 21 generates a trained model M2 for outputting the output data of the learning data d10 based on the input data of the learning data d10 (step ST40).

With the end of step ST40, the operation of the learning stage ends.

[Operation in operation stage]
Prior to the radiotherapy, the processing circuit 21 selects patient information and treatment plan data regarding the patient to be radiotherapy according to the operator's operation on the input interface 23. The patient information and the treatment plan data can be selected and acquired from, for example, the patient information and the treatment plan data in the treatment planning device 4. After that, the processing circuit 21 inputs subject information that affects the CBCT imaging condition among the patient information and the treatment plan data based on the patient information and the treatment plan data (step ST50). The subject information includes, for example, a treatment site of the subject, fixture information, and physique information. Not limited to this, the subject information may further include the therapeutic position, medical history, and internal tissue structure of the subject.

After step ST50, the processing circuit 21 inputs the subject information d4 into the trained model M2, and outputs the shooting condition d5 of the CBCT image used for aligning the subject based on the subject information d4 (step). ST60).

After step ST60, the processing circuit 21 controls the display 25 so as to display the output shooting condition d5. As a result, the processing circuit 21 prompts the operator to confirm the shooting condition d5 (step ST70).

After step ST70, the processing circuit 21 determines whether or not to capture a CBCT image using the displaying imaging condition d5 in response to the operator's operation on the input interface 23 (step ST80). In the case of no (ST80: No), the shooting condition d5 being displayed is modified according to the operation of the operator with respect to the input interface 23, and the process proceeds to step ST90.

After step ST80 or ST81, the processing circuit 21 outputs the confirmed imaging condition d5 to the radiotherapy apparatus 5. As a result, the operation of the shooting condition output device 2 in the operation stage is completed.

Hereinafter, in the radiotherapy device 5, the image capturing device 51 captures a cone beam CT image based on the imaging condition d5 received from the imaging condition output device (step ST90). The captured cone-beam CT image is compared with the planned CT image, which is a CT image for treatment planning. As a result of the comparison, the positional relationship between the image capturing device 51 and the bed is adjusted as appropriate, and the subject is aligned. After the alignment, the subject is subjected to radiation therapy by the radiation therapy device 5. When the radiotherapy is completed, the processing of the radiotherapy device 5 is terminated.

As described above, according to the present embodiment, subject information regarding a subject in radiotherapy is acquired. Further, based on the subject information, the subject information is input to the trained model for outputting the imaging conditions of the cone beam CT image used for the alignment of the subject, and the imaging conditions are output. As a result, appropriate shooting conditions are output from the trained model simply by inputting the subject information. Therefore, the shooting conditions can be reflected in the learning data, and the time and effort required for the operator to determine the appropriate shooting conditions can be reduced. Further, by learning the imaging conditions of high image quality and low dose to generate a trained model, it is possible to apply the imaging conditions of a cone beam CT image that achieves both high image quality and reduction of exposure dose.

Further, according to the present embodiment, the subject information may include the treatment site of the subject, the fixture information, and the physique information. In this case, in addition to the effects described above, subject information that affects the imaging conditions of the cone-beam CT image is used, so that the appropriateness of the imaging conditions output from the trained model can be ensured.

Further, according to the present embodiment, the physique information may include the weight or body thickness of the subject. In this case, in addition to the above-mentioned effects, the physique information that affects the imaging conditions of the cone beam CT image is used among various physique information, so that the appropriateness of the imaging conditions output from the trained model should be further ensured. Can be done.

Further, according to the present embodiment, the fixture information may be information representing a fixture that fixes the subject to the sleeper. The fixture may be located on an X-ray path for capturing a cone-beam CT image. In this case, in addition to the above-mentioned effects, since the information representing the fixture located on the X-ray path for capturing the cone beam CT image is used among various fixture information, the imaging output from the trained model is used. It is possible to further ensure the appropriateness of the conditions.

Further, according to the present embodiment, the evaluation result of the quality of the cone beam CT image may be obtained. Further, the shooting conditions may be output by inputting the subject information to the trained model that outputs the shooting conditions of the cone beam CT image indicating that the evaluation result is good based on the subject information. .. In this case, in addition to the above-mentioned effects, the imaging conditions indicating that the evaluation result of the cone beam CT image is good are output, so that the reliability of the output imaging conditions can be improved.

Further, according to the present embodiment, a plurality of past cone-beam CT images having different imaging conditions and a planned CT image of the subject may be further acquired. Further, the evaluation result of each of the plurality of cone beam CT images may be obtained by using the planned CT image as the correct answer data. In this case, in addition to the above-mentioned effects, since the planned CT image is used as the correct answer data, it can be expected that the imaging conditions of the cone beam CT image having high image quality such as the planned CT image are output.

Further, according to the present embodiment, the planned CT image capturing conditions indicating the capturing conditions of the planned CT image of the subject may be further acquired. Further, the imaging conditions are output by inputting the subject information and the planned CT image imaging conditions to the trained model that outputs the imaging conditions of the cone beam CT image based on the subject information and the planned CT image imaging conditions. , You may do so. In this case, in addition to the effects described above, the planned CT image capturing condition is added to the input to the trained model, so that the trained model can be easily generated, and the output imaging condition is a high-quality cone such as the planned CT image. It can be expected to correspond to the beam CT image. Supplementally, since the planned CT image capturing condition on the input side and the imaging condition of the cone beam CT image on the output side are related to some extent, the trained model can be generated as compared with the case where there is no planned CT image capturing condition on the input side. It is estimated that it will be easier. Similarly, since the planned CT image capturing condition on the input side and the imaging condition of the cone beam CT image on the output side are related to some extent, it is presumed that the output imaging condition corresponds to the high-quality cone beam CT image.

Further, according to the present embodiment, the planned CT image of the subject may be further acquired. Further, the imaging conditions are output by inputting the subject information and the planned CT image to the trained model that outputs the imaging conditions of the cone beam CT image based on the subject information and the planned CT image. May be good. In this case, in addition to the effects described above, the planned CT image is added to the input to the trained model, so that the trained model can be easily generated, and the output imaging conditions are high-quality cone beam CT such as the planned CT image. It can be expected to correspond to the image.

[First modification]
Next, a first modification of the present embodiment will be described. The first modification is a form in which a plurality of imaging conditions in which a plurality of viewpoints of cone beam CT imaging are individually prioritized are output.

Specifically, for example, as shown in FIG. 12, the imaging condition d5 of the CBCT image includes the first imaging condition when high image quality is prioritized and the second imaging condition when low dose is prioritized over high image quality. It is a plurality of shooting conditions d5 including. In FIG. 12, the "CBCT imaging condition (image quality priority)" corresponds to the first imaging condition, and the "CBCT imaging condition (dose priority)" corresponds to the second imaging condition. Along with this, with the realization of the condition output function 21d, the processing circuit 21 inputs the subject information d4 to the trained model M2 for outputting the plurality of shooting conditions d5 based on the subject information d4. , Outputs a plurality of shooting conditions d5. Further, the second imaging condition is not necessarily limited to the case where the image quality is lower than the first imaging condition, but includes the case where the image quality is equivalent to that of the first imaging condition and the dose is lower than that of the first imaging condition. Other configurations are the same as in this embodiment.

Therefore, according to the first modification as described above, in addition to the above-mentioned effects, the first imaging condition when high image quality is prioritized and the second imaging condition when low dose is prioritized over high image quality. Can be output. Further, when the operator selects the first imaging condition or the second imaging condition, the cone beam CT imaging can be performed based on the selected imaging condition. For example, according to the first modification, a plurality of optimum imaging conditions are displayed, and the operator can select an imaging condition that can further suppress the exposure dose from among the imaging conditions having the same high image quality. Therefore, according to the first modification, it is possible to apply the imaging conditions of the cone-beam CT image that achieve both high image quality and reduction of the exposure dose, depending on the operator's choice.

A first trained model M2 that outputs the first shooting condition and a second trained model M2 that outputs the second shooting condition are created, and one of the trained models M2 is selected. Therefore, the shooting conditions from the selected trained model M2 may be output.

[Second modification]
Next, a second modification of the present embodiment will be described. The second modification is a form in which the subject information is updated according to the change in the weight of the subject.

Specifically, for example, a weight measuring unit (weight scale) for measuring the weight of the subject is provided on the bed, and the processing circuit 21 updates the subject information based on the measured weight. Along with this, with the realization of the condition output function 21d, the processing circuit 21 outputs the updated shooting condition d5 by inputting the updated subject information d4 to the learned model M2. Other configurations are the same as in this embodiment.

Therefore, according to the second modification as described above, in addition to the above-mentioned effect, even if the physique of the subject changes during the radiation therapy schedule of about 30 days, the change in physique can be reflected in the imaging condition d5. , Appropriate shooting conditions can be maintained.

[Third variant]
Next, a third modification of the present embodiment will be described. In the third modification, as shown in FIG. 13, the processing circuit 21 performs reinforcement learning on the machine learning model M1. In reinforcement learning, the agent acts on the environment 21c1, and the environment 21c1 returns the state and the reward to the agent according to the action. By repeating this process, the agent learns to act (measure) to maximize the reward received from the environment. Since a detailed theory is known for such reinforcement learning, only the correspondence with the third modification will be described here. In the third modification, the agent corresponds to the machine learning model M1, the environment 21c1 corresponds to the reinforcement learning function of the learning function 21c, the action corresponds to the output of the CBCT imaging condition, and the state corresponds to the subject information d4. The reward corresponds to the evaluation result d3. Each of the agent and the environment 21c1 is realized by the processing circuit 21 executing each program.

Specifically, for example, by realizing the acquisition function 21a, the processing circuit 21 determines the subject information d4 regarding the subject in radiotherapy, the imaging condition d5 of the cone beam CT image used for aligning the subject, and the quality of the cone beam CT image. The evaluation result d3 is acquired.

Further, by realizing the learning function 21c, the processing circuit 21 sets the shooting condition d5 so that the evaluation result d3 as a reward for the action of updating and outputting the shooting condition d5 is good based on the state showing the subject information d4. Reinforcement learning to be updated and output is performed on the machine learning model M1. In reinforcement learning, the machine learning model M1 performs an action of outputting the CBCT imaging condition d5 to the environment 21c1, and the environment 21c1 machine-learns the state showing the subject information d4 and the reward showing the evaluation result d3 according to the action. Return to model M1. By repeating this, the machine learning model M1 learns to act (measure) to maximize (optimize) the evaluation result d3 received from the environment 21c1. As a result, the processing circuit 21 generates a trained model M2 for outputting the shooting condition d5 in which the evaluation result d3 is good, based on the subject information d4. That is, the machine learning model M1 for which reinforcement learning has been completed is designated as the trained model M2. The learning function 21c and the processing circuit 21 are examples of the learning unit. Other configurations are the same as in this embodiment.

According to the third modification as described above, in addition to the above-mentioned effects, even if a sufficient number of CBCT imaging conditions d5 having good evaluation results cannot be prepared, good evaluation results and bad evaluation results are rewarded. A trained model can be generated by the reinforcement learning. Further, since the bad evaluation result can be used for the learning data in addition to the good evaluation result, the number of learning data can be increased, and thus the accuracy of machine learning can be improved.

The third modification may be modified so that the cone beam CT image is further used as a state. That is, in the modified example of the third modified example, the state corresponds to the subject information d4 and the cone beam CT image d1. Other correspondences are the same as in the third modification.

In this case, by realizing the acquisition function 21a, the processing circuit 21 sets the subject information d4 regarding the subject, the cone beam CT image d1 used for aligning the subject, the shooting condition d5 of the cone beam CT image, and the cone beam CT image. The quality evaluation result d3 is acquired.

Further, by realizing the learning function 21c, the processing circuit 21 performs reinforcement learning on the machine learning model M1. Specifically, the processing circuit 21 shows the shooting conditions so that the evaluation result d3 as a reward for the action of updating and outputting the shooting condition d5 based on the state showing the subject information d4 and the cone beam CT image d1 is good. Reinforcement learning that updates and outputs d5 is performed on the machine learning model M1. In reinforcement learning, the machine learning model M1 performs an action of outputting the CBCT imaging condition d5 to the environment 21c1, and the environment 21c1 displays the subject information d4 and the cone beam CT image d1 and the evaluation result d3 according to the action. The shown reward is returned to the machine learning model M1. By repeating this, the machine learning model M1 learns to act (measure) to maximize (optimize) the evaluation result d3 received from the environment 21c1. As a result, the processing circuit 21 generates a trained model M2 for outputting the shooting condition d5 in which the evaluation result d3 is good, based on the subject information d4 and the planned CT image d2 of the subject. At this time, the processing circuit 21 sets the machine learning model M1 for which reinforcement learning has been completed as the trained model M2. In the learning stage, the past cone beam CT image d1 is input to the machine learning model M1, but in the operation stage, since there is no cone beam CT image of the subject, the planned CT image of the subject is input to the trained model M2. .. Supplementally, since the operation stage is a stage in which the learned model M2 for outputting the cone beam CT image capturing conditions is used before the cone beam CT image of the subject is captured, the cone beam CT image of the subject does not exist. Other configurations are the same as in this embodiment.

According to the modified example of the third modified example as described above, in addition to the effect of the third modified example, the state of the cone beam CT image d1 is changed, so that the efficiency of reinforcement learning can be improved.

According to at least one embodiment described above, the shooting conditions can be reflected in the learning data, and the time and effort required for the operator to determine the appropriate shooting conditions can be reduced.

The term "processor" used in the above description means, for example, a circuit such as a CPU, a GPU, an application specific integrated circuit (ASIC), or a programmable logic device. The programmable logic device includes, for example, a circuit such as a simple programmable logic device (SPLD), a complex programmable logic device (CPLD), and a field programmable gate array (FPGA). Can be used as appropriate. The processor realizes the function by reading and executing the program stored in the memory 22 or the like. Instead of storing the program in the memory 22 or the like, the program may be directly incorporated in the circuit of the processor. In this case, the processor realizes the function by reading and executing the program embedded in the circuit. Further, instead of executing the program, the function corresponding to the program may be realized by a combination of logic circuits. It should be noted that each processor of the present embodiment is not limited to the case where each processor is configured as a single circuit, and a plurality of independent circuits may be combined to form one processor to realize its function. good. Further, the plurality of components in FIG. 1 or FIG. 2 may be integrated into one processor to realize the function.

Further, the imaging condition output device or the radiotherapy device as described above may be expressed as shown in the following [1] to [4].

[1] An imaging condition output device that generates a learned model trained by combining subject information and cone beam CT imaging conditions, and uses the learned model to support optimum imaging of an image imaging device.

[2] The apparatus according to the above [1], which automatically discriminates or manually discriminates a cone beam CT image according to the viewpoint of evaluation.

[3] After the subject information is input to the trained model, the cone beam CT imaging conditions used in the image capturing apparatus are operated from a plurality of optimum cone beam CT imaging conditions presented as outputs of the trained model. The device according to the above [1] or [2], which can be selected by a person.

[4] A radiotherapy apparatus provided with the imaging condition output device according to any one of the above [1] to [3].

Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, as well as in the scope of the invention described in the claims and the equivalent scope thereof.

1 Image storage device 2 Imaging condition output device 21 Processing circuit 21a Acquisition function 21b Evaluation function 21c Learning function 21d Condition output function 21e Display control function 22 Memory 23 Input interface 24 Communication interface 25 Display 3 Treatment planning CT device 4 Treatment planning device 5 Radiation therapy device 51 Imaging device 100 Radiation therapy system

Claims (11)

An acquisition unit that acquires subject information about the subject in radiation therapy,
Condition output to output the shooting conditions by inputting the subject information to the trained model for outputting the shooting conditions of the cone beam CT image used for the alignment of the subject based on the subject information. A shooting condition output device that includes a unit. The subject information includes a treatment site, fixture information, and physique information of the subject.
The shooting condition output device according to claim 1. The physique information includes the weight or body thickness of the subject.
The shooting condition output device according to claim 2. The fixture information is information representing a fixture that fixes the subject to the sleeper.
The fixture is located on an X-ray path for capturing the cone-beam CT image.
The shooting condition output device according to claim 2 or 3. An evaluation unit for obtaining the evaluation result of the quality of the cone beam CT image is further provided.
The condition output unit inputs the subject information to the trained model that outputs the shooting conditions of the cone beam CT image indicating that the evaluation result is good based on the subject information, thereby performing the shooting conditions. To output,
The imaging condition output device according to any one of claims 1 to 4. The acquisition unit further acquires a plurality of past cone-beam CT images having different imaging conditions and a planned CT image of the subject.
The evaluation unit obtains the evaluation results of each of the plurality of cone beam CT images using the planned CT image as correct answer data.
The shooting condition output device according to claim 5. The acquisition unit further acquires the planned CT image shooting conditions indicating the shooting conditions of the planned CT image of the subject.
The condition output unit outputs the subject information and the planned CT image shooting condition to the trained model that outputs the shooting condition of the cone beam CT image based on the subject information and the planned CT image shooting condition. By inputting, the shooting conditions are output.
The imaging condition output device according to any one of claims 1 to 4. The acquisition unit further acquires a planned CT image of the subject, and obtains the planned CT image of the subject.
The condition output unit inputs the subject information and the planned CT image to the trained model that outputs the shooting conditions of the cone beam CT image based on the subject information and the planned CT image. , Output the shooting conditions,
The imaging condition output device according to any one of claims 1 to 4. The imaging conditions for the cone beam CT image are a plurality of imaging conditions including a first imaging condition when high image quality is prioritized and a second imaging condition when low dose is prioritized over the high image quality.
The condition output unit outputs the plurality of shooting conditions by inputting the subject information to the trained model for outputting the plurality of shooting conditions based on the subject information.
The imaging condition output device according to any one of claims 1 to 4. An acquisition unit that acquires subject information regarding a subject in radiotherapy, imaging conditions of a cone beam CT image used for aligning the subject, and an evaluation result of quality of the cone beam CT image.
Based on the state showing the subject information, the subject information is obtained by performing reinforcement learning to update and output the shooting conditions so that the evaluation result as a reward for the action of updating and outputting the shooting conditions is good. Based on this, a shooting condition output device including a learning unit that generates a trained model for outputting a shooting condition for which the evaluation result indicates good. Irradiation part that irradiates radiation and
An acquisition unit that acquires subject information about the subject in radiation therapy,
Condition output to output the shooting conditions by inputting the subject information to the trained model for outputting the shooting conditions of the cone beam CT image used for the alignment of the subject based on the subject information. A radiotherapy device including a part.

Images