WO2021237869A1

WO2021237869A1 - Parameter monitoring device and system for proton therapy

Info

Publication number: WO2021237869A1
Application number: PCT/CN2020/099492
Authority: WO
Inventors: 彭浩; 胡宗晟; 张小可
Original assignee: 杭州珞珈质子科技有限公司
Priority date: 2020-05-26
Filing date: 2020-06-30
Publication date: 2021-12-02
Also published as: CN111569279A; CN111569279B

Abstract

The present invention provides a parameter monitoring device and system for proton therapy. The device comprises: a recurrent neural network module, providing a trained recurrent neural network model, wherein the trained recurrent neural network model provides a nonlinear relationship between positron nuclide activity distribution and dose distribution of a proton beam; a PET obtaining module, obtaining, from a positron annihilation tomography system, a PET image of activity distribution of positron nuclides produced by a reaction between protons and tissues; a prediction module, inputting the PET image into the trained recurrent neural network model, and predicting the dose distribution and a Bragg peak position of the proton beam by means of the nonlinear relationship between the positron nuclide activity distribution and the dose distribution; a determination module, determining a positional relationship between the predicted dose distribution and Bragg peak position of the proton beam, and a target region; and an adjustment module, determining, according to a determination result of the determination module, whether a beam outlet parameter of the proton beam needs to be adjusted. The device and system provided in the present invention achieve the parameter monitoring and adjustment of proton therapy.

Description

Parameter monitoring device and system for proton therapy

Technical field

The invention relates to the field of nuclear medicine imaging, in particular to a parameter monitoring device and system for proton therapy.

Background technique

The current methods of treating tumors (cancer) mainly include surgery, chemotherapy and radiotherapy, and more than 70% of tumor patients need to receive radiotherapy alone or combined with radiotherapy. The ideal radiation therapy is to give a lethal dose to tumor cells while minimizing the radiation dose to the surrounding normal organs and tissues. Proton therapy is such a cutting-edge radiotherapy technology for the precise treatment of cancer. The radiation dose curve of the proton beam in tissues (such as humans, animals, etc.) first slowly rises, then gradually becomes faster, until the maximum dose deposition occurs at the Bragg peak, and then rapidly drops to zero. The unique dose-depth characteristic of the proton beam enables the tumor tissue to receive the maximum radiation dose, and at the same time, it can avoid the normal organs behind the tumor tissue from being damaged by radiation and reduce the side effects of treatment.

The advantages of proton therapy are: 1) The dose at the end of the proton Bragg peak is three to four times higher than the dose at the entrance, and the dose after the Bragg peak is almost zero. This feature not only maximizes the dose at the tumor, but also Effectively protect the normal tissues before and after the tumor; 2) The energy of the proton beam can be adjusted to irradiate tumors of different depths, so that proton therapy can adapt to tumors of different depths and different sizes and shapes; 3) Proton transmission has better performance. The small scattering and background make the edge of the illumination field clear, so it can treat tumors close to sensitive organs.

The Bragg peak characteristic of the proton beam is the biggest advantage of proton therapy, but it is also because of the existence of the Bragg peak that the treatment effect is very sensitive to the range. Once the actual treatment range deviates from the treatment plan, it is easy to cause insufficient dose of tumor tissue or over-irradiation of normal tissue, which greatly increases the risk of tumor recurrence and complications of normal organs. Therefore, in the process of proton therapy, the monitoring of the dose of the proton beam in the tissue and the position of the Bragg peak is very important to the effect of proton therapy.

At present, methods for monitoring the dose of proton therapy include prompt gamma measurement (for example, patent document CN 106291656A), acoustic wave monitoring and so on. But these methods have their shortcomings. For example, the prompt gamma method lacks a high-performance detector suitable for the gamma-ray energy spectrum region, and the detection efficiency of the detector is low; the sound signal in the acoustic wave method is very weak, and the signal-to-noise ratio is very low.

In proton therapy, protons will react with tissues to produce positron nuclides (mainly carbon-11 and oxygen-15). The PET system is used to image the activity of positron nuclides.

Therefore, how to collect nuclear medicine image data through computer equipment, and process the collected nuclear medicine image data to realize the parameter monitoring and adjustment of proton therapy.

Summary of the invention

In order to overcome the defects of the above-mentioned related technologies, the present invention provides a parameter monitoring device and system for proton therapy, thereby realizing the monitoring and adjustment of the parameters of proton therapy.

According to one aspect of the present invention, there is provided a parameter monitoring device for proton therapy, including:

A recurrent neural network module, configured to provide a trained recurrent neural network model, the trained recurrent neural network model providing a non-linear relationship between the positron nuclide activity distribution of the proton beam and the dose distribution;

The PET acquisition module is configured to acquire a PET image of the activity distribution of positron nuclides produced by the reaction between protons and tissues from the positron annihilation tomography system;

A prediction module, configured to input the PET image into the trained recurrent neural network model to predict the dose distribution and Bragg peak position of the proton beam;

The judgment module is configured to judge the positional relationship between the Bragg peak position and the target area according to the predicted dose distribution of the proton beam;

The adjustment module is configured to determine whether the beam output parameters of the proton beam need to be adjusted according to the judgment result of the judgment module.

In some embodiments of the present invention, the adjustment module is further configured to:

If the judgment module judges that the predicted dose distribution of the proton beam and the Bragg peak position are within the target area, then the beam output parameters of the proton beam are not adjusted;

If the judgment module judges that the predicted dose distribution of the proton beam and the Bragg peak position are outside the target area, readjust the beam output parameters of the proton beam.

In some embodiments of the present invention, the training samples of the recurrent neural network model during the training process at least include: stopping power, proton beam dose distribution, and positron nuclide activity distribution, wherein the stopping power passes the treatment The CT value of the CT image of the subject is converted, and the dose distribution and the positron nuclide activity distribution are obtained through the tissue simulation constructed in the Monte Carlo system of the three-dimensional phantom of the subject. The model is constructed based on the CT image of the subject.

In some embodiments of the present invention, the dose distribution and the positron nuclide activity distribution simulate different incident energies and/or different incident positions in a tissue constructed in a Monte Carlo system via a three-dimensional phantom of the subject to be treated The proton therapy process is obtained.

In some embodiments of the present invention, the stopping power is transformed according to the following steps:

Extracting the CT value of each pixel on the path of the proton beam from the three-dimensional CT image formed by the CT image of the treatment subject;

The stopping power corresponding to each pixel is calculated according to the beam output parameter of the proton beam and the CT value of each pixel.

In some embodiments of the present invention, the training sample further includes the CT value of the path of the proton beam, in the training sample, the CT value of each pixel on the path of the proton beam, the CT value of the proton beam The stopping power and the distribution of positron nuclide activity corresponding to each pixel on the path are used as the input of the recurrent neural network model, and the dose distribution of the proton beam is used as the output of the recurrent neural network model to determine the Recursive neural network model for training.

In some embodiments of the present invention, the prediction result of the recurrent neural network model adopts the mean square error and/or the mean absolute error for quantitative evaluation.

In some embodiments of the present invention, the prediction result of the recurrent neural network model adopts the following steps to evaluate the generalization ability:

Evaluation of the predicted dose distribution and Bragg peak position at different positions in the same picture and/or the evaluation of the predicted dose distribution and Bragg peak position by replacing different PET activity images.

In some embodiments of the present invention, it further includes:

The CT acquisition module is configured to acquire the CT image of the treatment subject,

Wherein, the prediction module is further configured to input the CT image of the treatment subject acquired by the CT acquisition module and the stopping power obtained based on the conversion of the CT image of the treatment subject into the trained recurrent neural network model.

According to another aspect of the present invention, there is also provided a parameter monitoring system for proton therapy, including:

CT system, configured to provide CT images of the subject;

The Monte Carlo system is configured to construct a treatment object based on a three-dimensional phantom of the treatment object, and to obtain a dose distribution of a proton beam and a positron nuclide activity distribution by simulating, the three-dimensional phantom is constructed based on a CT image of the treatment object

A PET system configured to provide PET images of the activity distribution of positron nuclides produced by the reaction between protons and tissues; and

The parameter monitoring device for proton therapy as described above.

Compared with the prior art, the advantages of the present invention are:

The invention adopts a trained recurrent neural network model, and predicts the corresponding dose distribution of PET image data collected in actual clinical practice by using the recurrent neural network model to realize the use of PET data to monitor the dose distribution and Bragg peak position in proton therapy. The invention greatly improves the prediction accuracy of dose distribution and Bragg peak position through data collection, data processing and machine learning, and shortens the prediction time. Therefore, through the device and system provided by the present invention, it is ensured that the dose of proton beam irradiation to the tumor tissue during the proton therapy process is normal and the position is accurate, and the risk of tumor recurrence and normal organ complications is reduced, thereby ensuring the proton therapy for tumor patients The therapeutic effect.

Description of the drawings

The above-mentioned and other features and advantages of the present invention will become more apparent by describing in detail the exemplary embodiments thereof with reference to the accompanying drawings.

Fig. 1 shows a block diagram of a parameter monitoring device for proton therapy according to an embodiment of the present invention.

Fig. 2 shows a schematic diagram of calculating stopping power according to an embodiment of the present invention.

Fig. 3 shows a schematic diagram of a recurrent neural network model according to an embodiment of the present invention.

Fig. 4 shows a block diagram of a parameter monitoring system for proton therapy according to an embodiment of the present invention.

Fig. 5 shows a schematic diagram of a recurrent neural network training process diagram according to an embodiment of the present invention.

Fig. 6 shows a schematic diagram of result prediction according to an embodiment of the present invention.

Fig. 7 shows a schematic diagram of a data acquisition system according to an embodiment of the present invention.

Fig. 8 shows a schematic diagram of a PET collection module according to an embodiment of the present invention.

Detailed ways

Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms, and should not be construed as being limited to the examples set forth herein; on the contrary, the provision of these embodiments makes the present invention more comprehensive and complete, and fully conveys the concept of the example embodiments To those skilled in the art. The described features, structures or characteristics can be combined in one or more embodiments in any suitable way.

In addition, the drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures denote the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in the form of software, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and/or processor devices and/or microcontroller devices.

The flowchart shown in the drawings is only an exemplary description, and does not necessarily include all the steps. For example, some steps can be decomposed, and some steps can be combined or partially combined. Therefore, the actual execution order may be changed according to actual conditions.

The parameter monitoring device 100 for proton therapy includes a recurrent neural network module 110, a PET acquisition module 120, a prediction module 130, a judgment module 140, and an adjustment module 150.

The recurrent neural network module 110 is configured to provide a trained recurrent neural network model that provides a non-linear relationship between the positron nuclide activity distribution of the proton beam and the dose distribution. Specifically, the recurrent neural network model may be a multi-input recurrent neural network model.

In some embodiments of the present invention, the training samples in the training process of the recurrent neural network model at least include stopping power (SP), the dose distribution of the proton beam, and the positron activity distribution.

The stopping power is obtained by transforming the CT value of the CT image of the subject.

Specifically, the parameter monitoring device 100 for proton therapy may also include a CT acquisition module. The CT acquisition module is configured to acquire a CT image of the treatment target (for example, a CT image of the patient's whole body) from a medical CT system. CT images can be obtained from medical CT systems. A three-dimensional phantom can be constructed based on the CT image of the subject to be treated, so as to determine the tumor focus area and the area of interest. Specifically, a medical CT system can be used to scan the whole body of the subject to obtain CT images, and the obtained multiple frames of CT images can be combined to construct a three-dimensional phantom, outline the location of the tumor lesion area, and appropriately expand the lesion area to obtain an interesting Region (Region of interest, ROI).

The stopping power can be transformed according to the following steps: extract the CT value of each pixel on the path of the proton beam from the three-dimensional CT image composed of the CT image of the treatment subject; according to the beam parameters of the proton beam and the proton beam The CT value on the passing path calculates the stopping power corresponding to each pixel on the path of the proton beam.

The stopping power can be calculated as follows:

According to the incident position of the proton beam, the CT value of each pixel on the path of the proton beam is extracted from the three-dimensional CT image, combined with the incident energy of the proton beam, and the stopping power corresponding to each pixel on the path of the proton beam is calculated. The formula is as follows:

Wherein, the SP _i is the i th pixel of the stopping power, constant k = 51Mev fm ^2, β _i is the ratio of the speed of light and the particle velocity, m _e is the electron mass, ρ _e is the electron density per volume of the substance, I is the compound Average excitation energy,

Is the shell correction item,

Is the polarization effect term, c is the speed of light, and E is the energy remaining after the particle passes through the matter.

Among them, the electron density per volume ρ _e of the substance and the average excitation energy I of the compound can be calculated according to the following formula:

Wherein, N _A is the Avogadro number, [rho] is the density of the material, _i the Z is the atomic number of the element i, A _i mass number of element i, W _i is the weight of element i weighting factor, the average excitation element i I _i Yes, where ρ, Z _i and w _i can be obtained by looking up the table of CT values.

The dose distribution and the positron nuclide activity distribution are obtained through a tissue simulation constructed in a Monte Carlo system with a three-dimensional phantom of the subject to be treated. Specifically, the tissues described in the embodiments of the present invention may include humans and/or animals. Specifically, the dose distribution and the positron nuclide activity distribution are obtained by simulating the proton treatment process of different incident energies and/or different incident positions in the tissue constructed in the Monte Carlo system through the three-dimensional phantom of the subject to be treated. . For example, the parameters of the pencil beam source can be set in the Monte Carlo system, and the positron emission process can be simulated through the program to ensure sufficient beam flow and set the corresponding beam period. The dose distribution information and the activity distribution information of positron emission are saved by the Monte Carlo tool.

In some embodiments of the present invention, the training sample further includes a CT value along the path of the proton beam. In the training sample, the CT value of each pixel on the path of the proton beam, the stopping power corresponding to each pixel on the path of the proton beam, and the positron nuclide activity distribution are used as the recursive The input of the neural network model, and the dose distribution is used as the output of the recurrent neural network model to train the recurrent neural network model to find the non-linear relationship between the positron nuclide activity distribution of the proton beam and the dose distribution .

In some embodiments of the present invention, the prediction result of the recurrent neural network model adopts mean square error (MSE) and/or mean absolute error (MAE) for quantitative evaluation.

In some embodiments of the present invention, the prediction result of the recurrent neural network model adopts the following steps to evaluate the generalization ability: the predicted dose distribution at different positions in the same picture is different from the evaluation and/or replacement of the Bragg peak position The estimated dose distribution and Bragg peak position were evaluated based on the PET activity image.

The PET acquisition module 120 is configured to acquire a PET image of the activity distribution of positron nuclides produced by the reaction between protons and tissues from the positron annihilation tomography system.

Specifically, in clinical proton therapy, after entering the tissue, protons react with the tissue to produce positron nuclides. The medical PET system is used to perform PET scanning on the target to obtain a PET reflecting the distribution of positron nuclides. Activity image.

Specifically, the Monte Carlo system can also be used to simulate clinically treated subjects undergoing PET imaging after receiving proton therapy. Specifically, a corresponding hardware PET system can be set up through simulation software, positron nuclides are selected (carbon-11, oxygen-15), according to imaging needs, in order to obtain sufficient photon data statistics (projection information) , Select the corresponding Monte Carlo time to get enough counts. After sufficient projection information is obtained, the acquired projection is reconstructed, and the single-slice rebinning algorithm (Single-Slice Rebinning, SSRB) and the two-dimensional ordered subset maximum likelihood method (Ordered Subsets Expectation Maximization, OSEM) are used to image Iterative reconstruction to obtain the reconstructed PET activity image. The attenuation correction of CT is completed, and the entire imaging field of the image is uniformized and corrected, and the corresponding PET image reflecting the activity distribution of positron nuclide is obtained.

The prediction module 130 is configured to input the PET image into the trained recurrent neural network model to predict the dose distribution and Bragg peak position of the proton beam.

The prediction module is further configured to input the CT image of the treatment subject acquired by the CT acquisition module and the stopping power obtained based on the conversion of the CT image of the treatment subject into the trained recurrent neural network model.

The determining module 140 is configured to determine the positional relationship between the Bragg peak position and the target area according to the predicted dose distribution of the proton beam.

The adjustment module 150 is configured to determine whether the beam output parameters of the proton beam need to be adjusted according to the judgment result of the judgment module 140.

Specifically, if the proton beam dose distribution does not fall within the tumor lesion area, but deviates from the tumor lesion area, then the irradiation position may be deviated when the proton beam is irradiated, and the irradiation position of the proton beam treatment head needs to be adjusted ; If the proton beam dose distribution does not fall within the tumor focus area, but spreads out of the tumor focus area, it means that the proton beam energy is too large when the proton beam is irradiated, and the dose of the proton beam needs to be adjusted. Therefore, the adjustment module 150 is further configured to:

If the judgment module 140 judges that the predicted dose distribution of the proton beam and the Bragg peak position are within the target area, it judges that the proton beam radiotherapy is accurate, and continues to perform radiotherapy according to the existing proton beam emission parameters, and does not treat the proton beam. The beam parameters of the beam are adjusted until the end of the treatment. The present invention is not limited by this.

If the judgment module 140 judges that the predicted dose distribution of the proton beam and the Bragg peak position are outside the target area, it judges that the proton beam radiotherapy is not accurate, and readjusts the beam parameters of the proton beam until the adjusted proton beam The dose distribution and Bragg peak position are consistent with the radiation treatment plan.

In the parameter monitoring device 100 for proton therapy according to the exemplary embodiment of the present invention, a trained recurrent neural network model is adopted, and the PET image data collected in the actual clinic is used to predict the corresponding dose distribution by using the recurrent neural network model to achieve The use of PET data in proton therapy to monitor dose distribution and Bragg peak position. The invention greatly improves the prediction accuracy of dose distribution and Bragg peak position through data collection, data processing and machine learning, and shortens the prediction time. Therefore, through the device and system provided by the present invention, it is ensured that the dose of proton beam irradiation to the tumor tissue during the proton therapy process is normal and the position is accurate, and the risk of tumor recurrence and normal organ complications is reduced, thereby ensuring the proton therapy for tumor patients The therapeutic effect.

In the parameter monitoring device 100 for proton therapy according to the exemplary embodiment of the present invention, it is verified that the monitoring method for dose and range verification in proton therapy based on the machine learning model of the present invention has a positive effect on the dose distribution, Broad applicability of Bragg peak position monitoring. It ensures that the dose of proton beam irradiation is normal and the position is accurate, and the risk of tumor recurrence and normal organ complications is reduced, thereby ensuring the therapeutic effect of proton therapy for tumor patients.

FIG. 1 only schematically shows the parameter monitoring device 100 for proton therapy provided by the present invention. Without violating the concept of the present invention, the splitting, merging, and adding of modules are all within the protection scope of the present invention. The parameter monitoring device 100 for proton therapy provided by the present invention can be implemented by software, hardware, firmware, plug-ins and any combination between them, and the present invention is not limited to this.

Fig. 2 shows a schematic diagram of calculating stopping power according to an embodiment of the present invention. The stopping power can be calculated according to the following method (see number 10 in Figure 2):

First, according to the CT value of each pixel in the CT image, the stopping power corresponding to each pixel is calculated. The stopping power calculation formula is as follows:

Wherein, the SP _i is the i-th pixel of the stopping power, constant k = 51Mev fm ^2, β _i is the speed of light, the ratio of particle velocity, m _e is the electron mass, ρ _e is the electron density per volume of the substance, I is the compound Average excitation energy.

Is the shell correction item,

Is the polarization effect term, c is the speed of light, and E is the remaining kinetic energy of the particle after passing through the matter.

Among them, the electron density per volume of the substance ρ _e is calculated as follows:

N _A is the Avogadro number, [rho] is the density of the material, _i the Z is the atomic number of the element i, the mass number of the element i A _i, W _i is the weight of element i weighting factor.

Among them, the average excitation energy I of the compound is calculated as follows:

Wherein, N _A is the Avogadro number, [rho] is the density of the material, _i the Z is the atomic number of the element i, A _i mass number of element i, W _i is the weight of element i weighting factor, the average excitation element i I _i can. Among them, ρ, Z _i , and w _i are obtained through the CT value look-up table, where the look-up table is: the corresponding table of the CT value and the composition of different tissue elements.

As the protons continue to pass through the tissue, the energy of the particles is gradually decreasing, and the protons continue to deposit a small amount of energy on the path they pass, until all the energy is released at the end of the proton beam, forming a Bragg peak, as shown in Figure 2. 20. The bethe theory in Figure 2 is the SP calculation formula (Bethe-Block formula) in the text.

Through this embodiment, it is possible to accurately calculate the energy deposited by the proton at the pixel position passed by, and provide accurate training samples for the training of the model; use it as an input to the neural network to provide a priori for the machine learning model, reducing learning Difficulty, to improve the prediction accuracy, anti-noise performance and generalization ability of the model.

Recurrent Neural Network (RNN) is specialized in processing time series data. It converts a time series input into a time series output through a hidden layer state, and continuously trains and learns the mapping relationship between input and output through back propagation and gradient descent methods. . The recurrent neural network model is composed of an input layer (x ₀ ～x _n ), a hidden layer (h ₀ ～h _n ), and an output layer (D ₀ ～D _n ). Wherein, the input layer of the network is a multi-data input, consisting of the CT image of the treatment subject, the stopping power obtained based on the conversion of the CT image of the treatment subject, and the activity value obtained from the PET image, and then the CT value, Three arrays of SP value and positron nuclide distribution activity value are juxtaposed. The output of the recurrent neural network model is the dose distribution, and the maximum position of the dose distribution is the Bragg peak. Further, the input data in the recurrent neural network model has a cascading relationship.

A network model is constructed through this embodiment to predict the proton beam dose distribution and Bragg peak position. After testing, the model has good prediction accuracy, anti-noise performance and generalization ability, and performs well in prediction based on PET images.

Fig. 4 shows a block diagram of a parameter monitoring system for proton therapy according to an embodiment of the present invention. The parameter monitoring system for proton therapy includes a CT system 210, a Monte Carlo system 220, a PET system 230, and the parameter monitoring device 100 for proton therapy as shown in FIG. 1.

The CT system 210 is configured to provide CT images of the treatment subject. Furthermore, a three-dimensional phantom can be constructed to delineate the area of interest for the final proton therapy monitoring terminal. The Monte Carlo system 220 is configured to construct the treatment object based on the three-dimensional phantom of the treatment object, and to obtain the dose distribution of the proton beam and the activity distribution of the positron nuclide by simulating to generate the training set required by the model. The three-dimensional phantom is based on the The construction of CT images of the subject to be treated. The PET system 230 is configured to provide a PET image of the activity distribution of positron nuclides produced by the reaction between protons and tissues. The modules included in the parameter monitoring device 100 have been described above, and will not be repeated here. The parameter monitoring device 100 can also perform a quantitative and qualitative evaluation of the predicted result, and follow up the adjustment of the proton beam.

Fig. 4 only schematically shows the parameter monitoring system of proton therapy provided by the present invention. Without violating the concept of the present invention, the splitting, merging, and adding of modules are all within the protection scope of the present invention. The parameter monitoring system for proton therapy provided by the present invention can be implemented by software, hardware, firmware, plug-ins and any combination between them, and the present invention is not limited to this.

Next, referring to Fig. 5, Fig. 5 shows a schematic diagram of a recurrent neural network training process diagram according to a specific embodiment of the present invention. Figure 5 shows a data collection system, a data preprocessing system, a recurrent neural network RNN model system, and a model evaluation system. The data collection system includes a data collection system, which mainly collects patient clinical data, including anatomical clinical information collected by the CT system, and clinical information collected by the PET system; the data collection system also includes a data simulation system, which is mainly based on CT images, Generate dose distribution diagram and activity distribution diagram in the Monte Carlo system. All the data obtained by the data collection system will be used for subsequent neural network model training and dose verification.

The data preprocessing module is used to convert CT data into stopping power, and to standardize the acquired data for better network training. The shown recursive neural network model selects preprocessed data according to a predetermined batch and inputs it into the neural network model for training, uses the validation set data to evaluate the model, quantitatively analyzes the performance parameters of the model, and quantitatively evaluates the model error. Evaluate whether the plan goal is met, continuously adjust the model parameters until the goal plan is met, and save the model.

Next, referring to FIG. 6, FIG. 6 shows a schematic diagram of result prediction according to an embodiment of the present invention. Figure 6 shows the patient data module, which includes a data acquisition module and a data calculation module. The data acquisition module mainly includes CT data acquisition and PET data acquisition. The data calculation is mainly based on the collected CT data and calculates it as the corresponding stopping power (SP), then input the CT data collected by the CT system, the activity data collected by the PET and the calculated stopping power into the trained RNN model, and output the prediction results.

Next, referring to Fig. 7, Fig. 7 shows a schematic diagram of a data acquisition system according to an embodiment of the present invention. FIG. 7 describes the data collection process and the method of obtaining data in a specific embodiment of the present invention. Figure 7 shows an accelerator system that generates a proton beam and a commonly used clinical data acquisition system. Among them, the accelerator system is mainly used to generate proton beams for proton therapy, and then the proton beams are shot into human tissues for tumor treatment. The hardware data acquisition system in the acquisition system mainly includes a CT system and a PET system. Among them, the CT system uses X-rays to image the structure of the tissue. Among them, the X-rays of the CT system are generated in a high-vacuum X-ray tube. The nuclear electric field acts to form radiation, which produces a continuous beam of X-rays. X-ray scans human tissues, and forms X-ray projection data after tissue attenuation on the X-ray detector, and reconstructs the data to form CT image data. After the proton beam enters the human body, it will react with the nucleus in the human body to produce a positron decay nuclide. The released positron annihilates two 511 keV photons in opposite directions. These annihilation photons are detected by the PET detector and then pass through another one. A series of image reconstruction algorithms form a PET human body activity image.

Next, referring to FIG. 8, FIG. 8 shows a schematic diagram of a PET collection module according to an embodiment of the present invention. It is known that the proton beam passing through the human body will react with the nucleus in the human body to produce positron decay nuclide, and the released positron annihilation produces two opposite 511keV photons, which are detected by the detector through positron annihilation imaging technology. (PET) Construct human activity images. It is necessary to image the positron-induced radionuclide activity signal so that the proton beam dose distribution can be found through the activity distribution. Therefore, it is necessary to build a PET system to collect data and complete the activity imaging. FIG. 8 shows an In-Beam PET system built in a Monte Carlo system according to an embodiment of the present invention, which includes two detector array units 2. After the proton beam 1 enters the patient's body 3, it will excite photons to hit the detector. Through coincidence detection, the activity distribution map is reconstructed to form an activity distribution data set, which is used for data input of the neural network.

The above are only multiple specific implementations of the present invention, and each specific implementation can be implemented individually or in combination, and the present invention is not limited thereto.

Compared with the prior art, the advantages of the present invention are:

Those skilled in the art will easily think of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptive changes of the present invention. These variations, uses, or adaptive changes follow the general principles of the present invention and include common knowledge or conventional technical means in the technical field that are not disclosed by the present invention. . The description and the embodiments are to be regarded as exemplary only, and the true scope and spirit of the present invention are pointed out by the appended claims.

Claims

A parameter monitoring device for proton therapy, which is characterized in that it comprises:

A recurrent neural network module, configured to provide a trained recurrent neural network model, the trained recurrent neural network model providing a non-linear relationship between the positron nuclide activity distribution of the proton beam and the dose distribution;

The PET acquisition module is configured to acquire a PET image of the activity distribution of positron nuclides produced by the reaction between protons and tissues from the positron annihilation tomography system;

A prediction module configured to input the PET image into the trained recurrent neural network model, and the non-linear relationship between the positron nuclide activity distribution of the proton beam and the dose distribution provided by the trained recurrent neural network model , To predict the dose distribution of the proton beam and the position of the Bragg peak;

A judging module, configured to judge the positional relationship between the Bragg peak position and the target area according to the predicted dose distribution of the proton beam;

The adjustment module is configured to determine whether the beam output parameters of the proton beam need to be adjusted according to the judgment result of the judgment module.
The parameter monitoring device for proton therapy according to claim 1, wherein the adjustment module is further configured to:

If the judgment module judges that the predicted dose distribution of the proton beam and the Bragg peak position are within the target area, then the beam output parameters of the proton beam are not adjusted;

If the judgment module judges that the predicted dose distribution of the proton beam and the Bragg peak position are outside the target area, readjust the beam output parameters of the proton beam.
The parameter monitoring device for proton therapy according to claim 1, wherein the training samples of the recurrent neural network model during the training process at least include: stopping power, proton beam dose distribution and positron nuclide activity distribution, Wherein, the stopping power is obtained by transforming the CT value of the CT image of the treatment subject, and the dose distribution and the positron nuclide activity distribution are obtained by the tissue simulation constructed in the Monte Carlo system through the three-dimensional phantom of the treatment subject , The three-dimensional phantom of the treatment target is constructed according to the CT image of the treatment target.
The parameter monitoring device for proton therapy according to claim 3, wherein the dose distribution and the positron nuclide activity distribution are simulated in a tissue constructed in a Monte Carlo system via a three-dimensional phantom of the subject to be treated. Different incident energy and/or different incident positions are obtained during proton therapy.
The parameter monitoring device for proton therapy according to claim 3, wherein the stopping power is transformed according to the following steps:

Extracting the CT value of each pixel on the path of the proton beam from the three-dimensional CT image formed by the CT image of the treatment subject;

The stopping power corresponding to each pixel is calculated according to the beam output parameter of the proton beam and the CT value of each pixel.
The parameter monitoring device for proton therapy according to claim 3, wherein the training sample further includes CT values on the path of the proton beam, and in the training sample, the proton beam passes through each pixel on the path. The CT value of a point, the stopping power corresponding to each pixel on the path of the proton beam, and the distribution of positron nuclide activity are used as the input of the recurrent neural network model, and the dose distribution of the proton beam is used as the recursive nerve The output of the network model is used to train the recurrent neural network model.
The parameter monitoring device for proton therapy according to claim 1, wherein the prediction result of the recurrent neural network model is quantitatively evaluated by means of mean square error and/or mean absolute error.
The parameter monitoring device for proton therapy according to claim 1, wherein the prediction result of the recurrent neural network model adopts the following steps to evaluate the generalization ability:

Evaluation of the predicted dose distribution and Bragg peak position at different positions in the same picture and/or the evaluation of the predicted dose distribution and Bragg peak position by replacing different PET activity images.
The parameter monitoring device for proton therapy according to claim 1, further comprising:

The CT acquisition module is configured to acquire the CT image of the treatment subject,

Wherein, the prediction module is further configured to input the CT image of the treatment subject acquired by the CT acquisition module and the stopping power obtained based on the conversion of the CT image of the treatment subject into the trained recurrent neural network model.
A parameter monitoring system for proton therapy, which is characterized in that it comprises:

CT system, configured to provide CT images of the subject;

The Monte Carlo system is configured to construct a treatment object based on a three-dimensional phantom of the treatment object, and to obtain a dose distribution of a proton beam and a positron nuclide activity distribution by simulating, the three-dimensional phantom is constructed based on a CT image of the treatment object;

A PET system configured to provide PET images of the activity distribution of positron nuclides produced by the reaction between protons and tissues; and

The parameter monitoring device for proton therapy according to any one of claims 1 to 9.