Dose determination method and apparatus based on time series signals, device, and storage medium

Abstract

The application provides a dose determination method and device based on time series signals, equipment and storage medium, relating to the fields of physics and computers. The method comprises: acquiring a plurality of sample CT images and determining dose distribution information and nuclide distribution information of each sample CT image; for each sample CT image, generating a PET cumulative signal image corresponding to each signal statistical window based on the nuclide distribution information; taking the sample CT image and the PET cumulative signal image corresponding to the sample CT image as a first training data set to train a first initial model to obtain a first target model; taking the dose distribution information, an output result of the first target model and part of training data in the first training data set as a second training set to train a second initial model to obtain a second target model; and inputting a target CT image and a PET cumulative signal image of a target time window into the first target model and the second target model in sequence to obtain dose information output by the second target model.

Description

Technical Field

[0001] This invention relates to the fields of physics and computer science, specifically to the application of physics and computer technology in the field of radiotherapy, and more specifically to a method, apparatus, device, and storage medium for dose determination based on time-series signals. Background Technology

[0002] Radiotherapy is one of the important means of cancer treatment. Proton and heavy ion therapy (collectively known as particle therapy) can achieve high-dose deposition in the target area and significantly reduce the radiation dose to surrounding normal tissues by utilizing the superior Bragg peak dose distribution characteristics. This results in higher tumor control rates and lower adverse reactions. The accuracy of dose delivery in particle therapy directly affects the effectiveness of particle therapy.

[0003] Currently, commonly used clinical dose verification methods mainly include offline physical dosimetry, CT-based range reconstruction, and radiochemical tracing. However, these methods have limitations, such as being unable to reflect the real-time location of the particle beam within the body and the actual dose distribution, and being significantly affected by factors such as tissue heterogeneity, scattering, and nuclear reactions. Summary of the Invention

[0004] The present invention provides a dose determination method, apparatus, device and storage medium based on time series signals, which is used to at least partially solve one of the above-mentioned technical problems.

[0005] According to a first aspect of the present invention, a dose determination method based on time-series signals is provided, comprising: acquiring multiple sample CT images and determining dose distribution information and nuclide distribution information corresponding to each sample CT image through Monte Carlo particle transport simulation; for each sample CT image, using multiple preset time points as signal statistical windows, and generating PET cumulative signal images corresponding to each signal statistical window based on the nuclide distribution information in each signal statistical window; training a first initial model using the multiple sample CT images and the PET cumulative signal images corresponding to the multiple sample CT images as a first training dataset to obtain a first target model for predicting PET cumulative signals corresponding to different time windows; training a second initial model using the dose distribution information, the output results of the first target model, and a portion of the training data in the first training dataset as a second training set to obtain a second target model for predicting dose distribution; and sequentially inputting the target CT image and the PET cumulative signal image of the target time window into the first target model and the second target model to obtain dose information output by the second target model.

[0006] According to an embodiment of the present invention, multiple sample CT images are acquired and the dose distribution information and nuclide distribution information corresponding to each sample CT image are determined by Monte Carlo particle transport simulation, including: simulating the particle transport process in tissue under different energy, energy divergence and broadening conditions based on Monte Carlo particle transport simulation technology; for each simulation scenario, calculating the dose distribution information and nuclide distribution information corresponding to the sample CT image under that simulation scenario.

[0007] According to an embodiment of the present invention, multiple preset time points are used as signal statistical windows, and a PET cumulative signal image corresponding to each signal statistical window is generated based on the nuclide distribution information in each signal statistical window. The method includes: setting multiple time points as signal statistical windows according to the nuclide half-life; calculating signal attenuation based on the number of remaining nuclides at each time point to obtain the number of photon pairs detected at each time point; determining the detection signal based on the number of photon pairs and generating a PET signal image based on the detection signal; integrating the PET signal images at multiple times in each time window to generate a PET cumulative signal image corresponding to each signal statistical window.

[0008] According to an embodiment of the present invention, the method further includes: for each simulated scenario, using sample CT images, PET cumulative signal images of a first time window, and beam information as input data, and using PET cumulative signal images of the remaining time window as output data to construct a sample pair corresponding to the simulated scenario; the first training dataset includes multiple sample pairs, and the beam parameters include incident energy, beam intensity, and beam broadening information.

[0009] According to an embodiment of the present invention, a first initial model is trained using CT images and PET cumulative signal images as a first training dataset to obtain a first target model for predicting PET cumulative signals corresponding to different time windows. The training includes: taking CT images, PET cumulative signal images of the first time window, and beam parameters as inputs, and taking PET cumulative signal images of the remaining time windows as outputs, training the first initial model through supervised learning to obtain a first target model capable of predicting the temporal evolution of subsequent PET signals.

[0010] According to an embodiment of the present invention, a second initial model is trained based on a second training dataset to obtain a second target model capable of predicting dose distribution. This includes: taking CT images, PET cumulative signals in a first time window, PET cumulative signals in the remaining time window predicted by the first target model, and beam information as inputs, and dose distribution information as outputs, training the second initial model through supervised learning, so that the second initial model learns the nonlinear mapping relationship between the temporal evolution characteristics of PET signals, tissue density characteristics, and the spatial distribution of dose deposition, thereby obtaining a second target model for predicting dose distribution.

[0011] According to an embodiment of the present invention, the target CT image and the PET cumulative signal image of the target time window are sequentially input into a first target model and a second target model to obtain dose information output by the second target model. This includes: inputting the target CT image, the PET cumulative signal image of the target time window, and beam information as first input data into the first target model so that the first target model can predict the respective PET cumulative signal images of future time windows based on the first input data; and inputting the PET cumulative signal image output by the first target model and the first input data as second input data into the second target model so that the second target model can determine the target dose distribution information based on the second input data.

[0012] According to a second aspect of the present invention, a dose determination device based on time-series signals is provided, comprising: a determination module, configured to acquire multiple sample CT images and determine the dose distribution information and nuclide distribution information corresponding to each sample CT image through Monte Carlo particle transport simulation; a generation module, configured to, for each sample CT image, use multiple preset time points as signal statistical windows, and generate PET cumulative signal images corresponding to each signal statistical window based on the nuclide distribution information in each signal statistical window; a first training module, configured to train a first initial model using multiple sample CT images and the PET cumulative signal images corresponding to multiple sample CT images as a first training dataset to obtain a first target model for predicting PET cumulative signals corresponding to different time windows; a second training module, configured to train a second initial model using dose absorption information, the output result of the first target model, and a portion of the training data in the first training dataset as a second training set to obtain a second target model for predicting dose distribution; and an acquisition module, configured to sequentially input the target CT image and the PET cumulative signal image of the target time window into the first target model and the second target model to obtain dose information output by the second target model.

[0013] According to a third aspect of the present invention, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0014] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0015] The above and other objects, features and advantages of the present invention will become more apparent from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0016] Figure 1 A flowchart illustrating a dose determination method based on time-series signals according to an embodiment of the present invention is shown schematically.

[0017] Figure 2 This schematically illustrates a flowchart of acquiring multiple sample CT images and determining the dose distribution information and radionuclide distribution information corresponding to each sample CT image through Monte Carlo particle transport simulation according to an embodiment of the present invention.

[0018] Figure 3 This illustration illustrates a flowchart of how, according to an embodiment of the present invention, multiple preset time points are used as signal statistics windows, and a PET cumulative signal image corresponding to each signal statistics window is generated based on the nuclide distribution information in each signal statistics window.

[0019] Figure 4 The schematic diagram illustrates the principle of training a first initial model according to an embodiment of the present invention;

[0020] Figure 5 The schematic diagram illustrates the principle of training a second initial model according to an embodiment of the present invention;

[0021] Figure 6 The flowchart illustrates, according to an embodiment of the present invention, the sequential input of a target CT image and a target time window PET cumulative signal image into a first target model and a second target model to obtain dose information output by the second target model;

[0022] Figure 7 A schematic diagram illustrating the structure of a dose determination device based on time-series signals according to an embodiment of the present invention is shown.

[0023] Figure 8 A block diagram of an electronic device for a dose determination method based on time-series signals according to an embodiment of the present invention is shown schematically. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of a feature, step, operation, and / or component, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a connection that allows communication between them; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0027] In the description of this invention, it should be understood that the terms "longitudinal", "length", "circumferential", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the subsystem or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0028] Throughout the accompanying drawings, identical elements are represented by the same or similar reference numerals. Conventional structures or configurations may be omitted where they might cause confusion in understanding the invention. Furthermore, the shapes, dimensions, and positional relationships of the components in the drawings do not reflect actual size, scale, or actual positional relationships. Additionally, any reference symbols placed within parentheses should not be construed as limiting.

[0029] Similarly, to simplify the invention and aid in understanding one or more of the various disclosed aspects, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together into a single embodiment, figure, or description thereof. The use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicates that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0030] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0031] This invention provides a method, apparatus, device, and storage medium for dose determination based on time-series signals. Before introducing the technical solutions provided by this invention, the related technologies involved in this invention will be described first.

[0032] In recent years, online or near-real-time particle beam monitoring technology based on positron emission tomography (PET) has gradually attracted attention. Its principle is based on utilizing the β⁺ radioactive nuclides (such as...) produced by the interaction between particles and tissue nuclei. 11 C 15 O、 13 The spatial distribution of these radioactive signals is detected by a PET system, thereby indirectly reflecting the particle beam energy deposition and range information.

[0033] However, traditional PET monitoring has the following problems: First, the amount of radionuclide generated is affected by the complex influence of beam energy, tissue composition and nuclear reaction cross section, resulting in a non-linear correspondence between signal intensity and actual dose; second, the acquisition time is long, the image noise is high and the spatiotemporal resolution is limited, making it difficult to achieve high-precision dynamic dose mapping.

[0034] With the rapid development of artificial intelligence technology, deep learning-based methods for radiotherapy dose prediction and monitoring have gradually become a research hotspot in the field of particle therapy. Traditional physical and empirical models require complex Monte Carlo simulations and tissue parameter assumptions in dose calculation, resulting in high computational costs and high sensitivity to input accuracy. In recent years, deep neural networks, with their powerful capabilities in high-dimensional nonlinear mapping modeling, have been widely used to directly learn the particle beam energy deposition patterns from multimodal image data, achieving end-to-end prediction from image to dose distribution.

[0035] In particle therapy, researchers have attempted to use multi-source data, including pre-treatment CT images, stopping power ratio (SPR) distribution, and PET signals acquired during treatment, as network inputs. They have employed structures such as deep convolutional networks (CNNs), 3D U-Nets, or spatiotemporal attention networks to learn the complex correspondence between PET activity distribution and actual dose deposition distribution. While these methods overcome the limitation of traditional models that can only qualitatively estimate range deviation based on static activity maps, their accuracy in predicting 3D dose distribution remains insufficient.

[0036] Current PET dose verification methods typically infer dose distribution based solely on radioactivity images (i.e., PET signals) acquired at a fixed time point (such as at the end of treatment). On one hand, the energy deposition of the particle beam in tissue changes dynamically over time, and single-time-point data only reflects an "instantaneous state," failing to fully describe this process and leading to inaccurate dose estimation. On the other hand, obtaining more comprehensive data requires multiple PET signal acquisitions, resulting in prolonged acquisition time, increased patient radiation exposure, and low clinical efficiency.

[0037] This invention proposes a dose determination method based on time-series signals, comprising: acquiring multiple sample CT images and determining the dose distribution information and nuclide distribution information corresponding to each sample CT image through Monte Carlo particle transport simulation; for each sample CT image, using multiple preset time points as signal statistical windows, and generating PET cumulative signal images corresponding to each signal statistical window based on the nuclide distribution information in each signal statistical window; training a first initial model using multiple sample CT images and the corresponding PET cumulative signal images as a first training dataset to obtain a first target model for predicting PET cumulative signals corresponding to different time windows; training a second initial model using the dose distribution information, the output results of the first target model, and a portion of the training data in the first training dataset as a second training set to obtain a second target model for predicting dose distribution; and sequentially inputting the target CT image and the PET cumulative signal image of the target time window into the first target model and the second target model to obtain the dose information output by the second target model.

[0038] This invention introduces a first target model that uses time-series prediction technology to predict the cumulative signal at subsequent time points (e.g., 5 minutes, 10 minutes, 30 minutes) using early PET signals (e.g., at 2 minutes), reducing acquisition time and radiation burden. By predicting subsequent signals, the cascade distribution can be dynamically calculated based on PET signals at different times, improving the accuracy of dose estimation. The second target model integrates CT images, PET signals, and beam parameter information to achieve fusion learning of tissue structure characteristics, radioactivity distribution characteristics, and particle transport characteristics. Through multi-source data joint modeling, the second target model can capture the differences in PET signals caused by different tissue densities and nuclear reactions, thereby accurately reconstructing the true energy deposition distribution. This effectively improves the dose spatial resolution and numerical accuracy of the dose estimation results.

[0039] Figure 1 A flowchart illustrating a dose determination method based on time-series signals according to an embodiment of the present invention is shown.

[0040] like Figure 1As shown, the dose determination method based on time series signals in this embodiment includes operations S110 to S150.

[0041] In operation S110, multiple sample CT images are acquired and the dose distribution information and nuclide distribution information corresponding to each sample CT image are determined through Monte Carlo particle transport simulation.

[0042] In some embodiments, the sample CT images can be image data obtained through actual scans or simulated images generated using computer software and algorithms. The sample CT images can include images of different types and tissue structures. Each sample CT image can be simulated using a Monte Carlo particle transport simulation algorithm. By simulating the proton or heavy ion incident process under different energies, energy divergences, and broadening conditions, the dose distribution information and nuclide distribution information corresponding to each sample CT image can be accurately determined to simulate the particle transport and energy deposition process in tissues during actual treatment.

[0043] In operation S120, for each sample CT image, multiple preset time points are used as signal statistics windows, and PET cumulative signal images corresponding to each signal statistics window are generated based on the nuclide distribution information in each signal statistics window.

[0044] In some embodiments, multiple preset time points are set as signal statistical windows for each sample CT image. Based on the nuclide distribution information, half-life, and decay characteristics of the nuclides in each signal statistical window, the cumulative PET signal image corresponding to each signal statistical window is statistically analyzed and generated. For example, different time points such as 2 minutes, 5 minutes, 10 minutes, and 30 minutes can be set as statistical windows, and the cumulative PET signal within these windows can be calculated respectively.

[0045] In operation S130, multiple sample CT images and the corresponding PET cumulative signal images of multiple sample CT images are used as the first training dataset to train the first initial model, thereby obtaining the first target model for predicting the PET cumulative signal corresponding to different time windows.

[0046] In some embodiments, multiple sample CT images and their corresponding PET cumulative signal images are combined to form a first training dataset. A first initial model is trained using the first training dataset. The first initial model can be a deep neural network model, such as U-Net, ResNet, etc. Through supervised learning, the model learns the mapping relationship from CT images and PET cumulative signal images of a specific time window to PET cumulative signals of other time windows, resulting in a first target model for predicting PET cumulative signals corresponding to different time windows.

[0047] For example, sample CT images, 2-minute accumulated PET signals, and beam parameters (such as incident energy, beam intensity, beam broadening information, etc.) can be used as input data, and the accumulated PET signals at 5 minutes, 10 minutes, and 30 minutes can be used as output for supervised learning to train the first target model, thereby obtaining the accumulated PET signals at 5 minutes, 10 minutes, and 30 minutes predicted by the first target model.

[0048] In operation S140, the dose distribution information, the output of the first target model, and a portion of the training data in the first training dataset are used as the second training set to train the second initial model, thereby obtaining the second target model for predicting the dose distribution.

[0049] In some embodiments, the dose distribution information obtained from Monte Carlo particle transport simulation, the 2-minute cumulative PET signal, the output of the first target model (cumulative PET signals at 5 minutes, 10 minutes, and 30 minutes), and the beam information are used as the second training set to train the second initial model. The second initial model is supervised learning based on the dose distribution information obtained from Monte Carlo particle transport simulation as the output, and learns the mapping relationship from the dose distribution information, the output of the first target model, and some original training data to the real dose distribution, thus seeing the second target model used to predict the dose distribution.

[0050] In operation S150, the target CT image and the cumulative PET signal image of the target time window are sequentially input into the first target model and the second target model to obtain the dose information output by the second target model.

[0051] In some embodiments, the target CT image and the cumulative PET image of the target time window are sequentially input into a first target model and a second target model. The first target model predicts the cumulative PET signal of other time windows based on the input target CT image and the cumulative PET signal image of the target time window. The cumulative PET signals of multiple time windows predicted based on the first target model, the target CT image, and the cumulative PET image of the target time window are jointly input into the second target model to obtain the dose information predicted by the second target model.

[0052] The dose determination method provided in this invention comprehensively utilizes CT images, PET signals, and beam parameter information for dose distribution prediction, achieving fusion learning of tissue structure characteristics, radioactivity distribution characteristics, and particle transport characteristics. The model obtained through multi-source data joint modeling can capture the differences in PET signals induced by different tissue densities and nuclear reactions, thereby accurately reconstructing the true energy deposition distribution and effectively improving the physical consistency and accuracy of the predicted distribution.

[0053] The dose determination method provided in this invention predicts dose based on cumulative PET signals over multiple time windows, effectively improving the accuracy and flexibility of dose prediction. The trained first target model can quickly predict the cumulative PET signals of other time windows based on the input CT image and the cumulative PET signal image of a specific time window. This provides effective support for the second target model to predict dose based on the changes in PET signals across different time windows, improving the accuracy and flexibility of dose information determination. The phased training approach for the first and second target models allows each model to focus on a specific task, avoiding the problems of training difficulties and inaccurate predictions caused by overly complex individual models, further improving the accuracy of model predictions.

[0054] Figure 2 The flowchart illustrating the acquisition of multiple sample CT images and the determination of dose distribution information and nuclide distribution information corresponding to each sample CT image through Monte Carlo particle transport simulation according to an embodiment of the present invention is shown.

[0055] like Figure 2 As shown, this embodiment acquires multiple sample CT images and determines the dose distribution information and nuclide distribution information corresponding to each sample CT image through Monte Carlo particle transport simulation, including operations S210 to S220.

[0056] In operation S210, the transport process of particles in tissues under different energies, energy divergences and broadening conditions was simulated using Monte Carlo particle transport simulation technology.

[0057] In operation S220, for each simulated scenario, the dose distribution information and nuclide distribution information corresponding to the sample CT image under that simulated scenario are calculated.

[0058] In some embodiments, the sample CT images can cover samples of different tissue types and structures. Energy divergence reflects the degree of dispersion of particle energy in the particle beam, and broadening describes the spatial diffusion of the particle beam. These parameters can be selected according to common situations in actual treatment to simulate the real particle transport environment.

[0059] Monte Carlo particle transport simulation technology can be used to simulate the transport process of particles in tissue for each sample CT image and a set combination of simulation parameters. The Monte Carlo method tracks the trajectory of each particle through random sampling, considering various interactions between particles and tissue atoms, such as elastic scattering, inelastic scattering, and nuclear reactions. The total energy of particle deposition within each voxel can be calculated from the simulation results, and the dose distribution can be calculated by combining this with tissue density. Furthermore, the nuclide distribution information can be obtained based on the simulated nuclide decay.

[0060] Figure 3 The illustration shows a flowchart of generating a PET cumulative signal image corresponding to a signal statistics window based on the nuclide distribution information in each signal statistics window, using multiple preset time points as signal statistics windows according to an embodiment of the present invention.

[0061] like Figure 3 As shown, this embodiment uses multiple preset time points as signal statistics windows, and generates a PET cumulative signal image corresponding to the signal statistics window based on the nuclide distribution information in each signal statistics window, including operations S310 to S340.

[0062] When operating S310, multiple time points are set as signal statistics windows based on the nuclide half-life.

[0063] In some embodiments, the half-life of a nuclide is an inherent characteristic of the nuclide, and different nuclides have different half-lives. Setting time points based on the half-life of the nuclide allows for the reasonable division of different stages, thereby enabling the study of the nuclide's changes. For example, for nuclides with shorter half-lives, the time point intervals can be relatively shorter; for nuclides with longer half-lives, the time point intervals can be appropriately extended.

[0064] In operation S320, signal attenuation is calculated based on the number of nuclides remaining at each time point to obtain the number of photon pairs detected at each time point.

[0065] In some embodiments, for each set time point, signal attenuation calculations are performed based on the number of nuclides remaining at that time point. Nuclides decay over time, and their numbers gradually decrease. Based on the decay patterns of nuclides, the number of nuclides remaining at each time point can be calculated, and then the number of photon pairs detected at each time point can be obtained according to relevant physical models.

[0066] In operation of S330, the detection signal is determined based on the number of photon pairs and a PET signal image is generated based on the detection signal.

[0067] In some embodiments, the detection signal is determined based on the calculated number of photon pairs. Different numbers of photon pairs result in different detected signal intensities. Based on these detection signals, imaging techniques are used to generate PET signal images, which reflect the signal intensity distribution corresponding to the nuclide distribution at a specific time point.

[0068] In operation S340, the PET signal images at multiple times in each time window are integrated to generate the PET cumulative signal image corresponding to each time window.

[0069] In some embodiments, integration refers to accumulating signal images from multiple time points to generate a cumulative PET signal image corresponding to each time window. The cumulative PET signal image can reflect the decay process and signal changes of the nuclide within the current time window, containing the dynamic characteristics of the nuclide.

[0070] This invention calculates the cumulative PET signal image for each time window based on multiple time points, which can reduce the impact of random errors and noise to a certain extent, and improve the stability and reliability of the signal. The cumulative PET image contains signal change information within the time window. By analyzing the cumulative PET signal image, the distribution of nuclides in different regions and the deposition location and intensity of energy during decay can be determined more accurately, providing a more accurate physical basis for dose prediction, making the dose prediction more consistent with reality, and improving the accuracy and reliability of the prediction.

[0071] According to an embodiment of the present invention, the dose determination method further includes: for each simulated scenario, using sample CT images, PET cumulative signal images of a first time window, and beam information as input data, and using PET cumulative signal images of the remaining time window as output data to construct a sample pair corresponding to the simulated scenario.

[0072] In some embodiments, the first training dataset includes multiple sample pairs, and the beam parameters include incident energy, beam intensity, and beam broadening information. This dataset covers PET cumulative signal images at different time windows under different tissue structures and beam parameter conditions, providing ample data support for subsequent model training.

[0073] In some embodiments, the dose distribution information of each simulation case can be paired with multi-time point PET cumulative signal images (e.g., 2 minutes, 5 minutes, 10 minutes, 30 minutes). The spatial resolution of the dose distribution map and the PET cumulative signal map is consistent, and the data obtained after pairing can reflect the dose distribution and nuclide distribution of the same site after radiation irradiation.

[0074] This invention integrates previously independent information by constructing sample pairs of accumulated PET signal images at multiple time points. This provides subsequent model training with PET signal data under different time windows, helping the model better understand the variation patterns of PET signals under different conditions. The first training dataset contains multi-dimensional data, which can more comprehensively describe the characteristics of the simulated scene, enabling the model to capture more complex physical relationships and patterns during the learning process, thereby improving the accuracy of model training.

[0075] Figure 4 The schematic diagram illustrates the principle of training a first initial model according to an embodiment of the present invention.

[0076] like Figure 4 As shown, according to an embodiment of the present invention, CT images and PET cumulative signal images are used as a first training dataset to train a first initial model to obtain a first target model for predicting PET cumulative signals corresponding to different time windows. The method includes: taking CT images, PET cumulative signal images of the first time window, and beam parameters as inputs, and taking PET cumulative signal images of the remaining time windows as outputs, training the first initial model through supervised learning to obtain a first target model with the ability to predict the temporal evolution of subsequent PET signals.

[0077] In some embodiments, a suitable deep neural network structure can be selected as the first initial model according to the task requirements. The first initial model can be U-Net, ResNet, RNN, or a Transformer-based spatiotemporal feature extraction network, etc.

[0078] The model takes CT images, PET cumulative signal images from the first time window, and beam parameters as inputs. The CT images and PET cumulative signal images from the first time window can be stitched together or input as separate channels. The beam parameters can be used as additional feature vectors fused with the image features. The model outputs PET cumulative signal images from the remaining time windows; that is, the first target model needs to predict the PET cumulative signal for subsequent time windows.

[0079] The training set data is input into the first initial model, and the prediction results are calculated through forward propagation. Then, the loss value is calculated according to the loss function, and the model parameters are updated through backpropagation. This process is repeated multiple times until the model's performance on the validation set no longer significantly improves or reaches the preset number of training rounds. The model performance is evaluated and optimized according to appropriate evaluation metrics to obtain the first target model capable of predicting the temporal evolution of subsequent PET signals.

[0080] The PET signal input to the first initial model can also be a signal incorporating Time-of-Flight (ToF) PET information. ToF-PET technology measures the time difference between the arrival of gamma photon pairs at the detector, utilizing the speed of light propagation in a medium to more accurately locate radionuclides within the body, thereby improving the spatial resolution of the image. Conventional PET signals can be combined with ToF-PET information, allowing the input of a PET signal incorporating more precise location information obtained through ToF technology into the first initial model for training.

[0081] ToF-PET technology can more accurately locate the position of radionuclides in the patient's body, so that the first target model can more accurately understand the spatial distribution of different nuclides, thereby better distinguishing the signals of different nuclides, reducing misjudgments and interference, and more accurately decomposing the signal proportion map of each nuclide.

[0082] This invention employs CT images, PET cumulative signal images within a first time window, and beam parameters as input, and uses the PET cumulative signal from the remaining time window as output for training. This allows for the comprehensive capture of the dynamic characteristics of PET signal decay, diffusion, and transformation over time. CT images provide tissue structure and density information, the PET signal from the first time window reflects the initial nuclide distribution, and beam parameters influence the generation and distribution of nuclides. Using this information, the model can more accurately learn the changing patterns of the PET signal at different time points, improving the accuracy of PET cumulative signal prediction. This, in turn, helps the second target model more accurately calculate the radiation dose received by the tissue.

[0083] Figure 5 The schematic diagram illustrates the principle of training a second initial model according to an embodiment of the present invention.

[0084] like Figure 5 As shown, according to an embodiment of the present invention, a second initial model is trained based on a second training dataset to obtain a second target model capable of predicting dose distribution. This includes: taking sample CT images, PET cumulative signals in a first time window, PET cumulative signals in the remaining time window predicted by the first target model, and beam information as inputs, and dose distribution information as outputs, training the second initial model through supervised learning, so that the second initial model learns the nonlinear mapping relationship between the temporal evolution characteristics of PET signals, tissue density characteristics, and the spatial distribution of dose deposition, thereby obtaining a second target model for predicting dose distribution.

[0085] In some embodiments, a suitable deep learning architecture can be selected to construct a second initial model for predicting three-dimensional dose distribution information. For example, a 3D convolutional neural network, a U-Net architecture, or a Transformer architecture can be used. Exemplarily, the second initial model can employ a multi-branch input structure, processing input data of different dimensions based on different branches, fusing features extracted from different branches through a feature fusion layer, and performing nonlinear regression through fully connected layers or convolutional layers to output the predicted dose distribution.

[0086] Using sample CT images, PET cumulative signals in the first time window, PET cumulative signals in the remaining time window predicted by the first target model, and beam information from the first training dataset as inputs to the second initial model, and using three-dimensional dose distribution information as outputs to define the learning objective of the model, the second initial model is trained.

[0087] Supervised learning can be used to train the second initial model. Input data is fed into the second initial model, which performs forward propagation to obtain the predicted 3D dose distribution information. The loss function value between the predicted and actual dose distribution information is calculated; commonly used loss functions include mean squared error (MSE) and mean absolute error (MAE), which measure the difference between the predicted and actual values. The loss value is then backpropagated into the model using a backpropagation algorithm. The model parameters are updated according to optimization algorithms (such as Adam, SGD, etc.), adjusting the model's weights and biases to gradually reduce the loss function value. The forward and backpropagation processes are repeated for multiple iterations until the model's performance on the validation set no longer significantly improves or reaches a predetermined number of training epochs, resulting in the second target model used for predicting the dose distribution.

[0088] This invention trains a second initial model using data from multiple dimensions. These data complement each other, providing the model with more comprehensive and richer information, which helps improve the accuracy of dose prediction. This invention first trains a first initial model to decompose nuclide signals, then uses the nuclide signal distribution as input to the second model, achieving hierarchical data processing. This allows each model to focus on a specific task: the first target network focuses on nuclide signal decomposition, and the second target network focuses on dose prediction, improving the model's learning efficiency and performance while reducing data processing complexity.

[0089] In some embodiments, the first target model and the second target model can replace deep models with different structures according to task requirements, data characteristics, or computing resources. For example, they can include 3D U-Net, DenseNet, Temporal Convolutional Network (TCN), or hybrid network models that combine convolutional and recurrent units. Those skilled in the art can select models with appropriate structures according to actual conditions, and this invention does not limit them.

[0090] Figure 6 The flowchart illustrates, according to an embodiment of the present invention, a target CT image and a target time window PET cumulative signal image are sequentially input into a first target model and a second target model to obtain dose information output by the second target model.

[0091] like Figure 6As shown, in this embodiment, the target CT image and the cumulative PET signal image of the target time window are sequentially input into the first target model and the second target model to obtain the dose information output by the second target model, including operations S610 to S620.

[0092] In operation S610, the target CT image, the PET cumulative signal image of the target time window, and the beam information are input as the first input data to the first target model, so that the first target model can predict the PET cumulative signal images of the remaining time windows based on the first input data.

[0093] In operation S620, the PET cumulative signal image output by the first target model and the first input data are input as the second input data to the second target model so that the second target model can determine the target dose distribution information based on the second input data.

[0094] In some embodiments, the target time window is the time window corresponding to the first time point, for example, a time window within 2 minutes. The PET cumulative signal, target CT image and beam information acquired within 2 minutes are input into the first target model, which predicts the PET signal distribution at future times based on the input data and outputs the PET cumulative signal images for each future time window.

[0095] The PET cumulative signal image output by the first target model and the first input data (CT image, PET cumulative signal acquired within 2 minutes and beam information) are used as the second input data and input into the second target model. The second target model quickly reconstructs the three-dimensional dose distribution of particle therapy based on these data, thereby realizing real-time or near-real-time assessment of the treatment dose.

[0096] This invention provides a dose distribution prediction method that uses the collaborative work of a first target model and a second target model. By using PET cumulative signals as intermediate stage information, the three-dimensional dose distribution of ion therapy can be quickly reconstructed. This improves the accuracy of dose distribution prediction, effectively shortens the prediction time, and achieves real-time prediction of treatment dose.

[0097] The training data in this embodiment of the invention comes from standardized Monte Carlo simulations, and the network structure is modularly designed, allowing for flexible adaptation to proton and heavy ion therapy systems with different energy ranges and beam patterns to meet the needs of various application scenarios. For different brands or models of PET / CT equipment, cross-device transfer can be achieved simply by fine-tuning the network parameters with a limited number of samples. Furthermore, the model can be further extended to time-dependent PET, ToF-PET, or prompt-γ signal fusion tasks, providing a unified framework for multimodal dose validation.

[0098] The embodiments of this invention employ a dual-model deep learning structure, with a clear division of labor between the time-series prediction model (first target model) and the dose prediction model (second target model), avoiding redundant calculations, accelerating the inference process, significantly reducing the computational burden, and providing real-time feedback capabilities. It can perform rapid inference on GPU hardware, with a prediction time of only a few seconds to tens of seconds. It can be directly embedded into the online monitoring module of a particle therapy system to assist doctors in judging treatment range deviations and verifying the execution of planned doses, further improving treatment safety and the level of individualized control.

[0099] Figure 7 A schematic block diagram of a dose determination device based on a time-series signal according to an embodiment of the present invention is shown.

[0100] like Figure 7 As shown, the dose determination device 700 based on time series signals in this embodiment includes a determination module 710, a generation module 720, a first training module 730, a second training module 740, and an acquisition module 750.

[0101] The determination module 710 is used to acquire multiple sample CT images and determine the dose distribution information and nuclide distribution information corresponding to each sample CT image through Monte Carlo particle transport simulation. In one embodiment, the determination module 710 can be used to perform the operation S110 described above, which will not be repeated here.

[0102] The generation module 720 is used to generate a PET cumulative signal image corresponding to each signal statistical window for each sample CT image, using multiple preset time points as signal statistical windows and based on the nuclide distribution information in each signal statistical window. In one embodiment, the generation module 720 can be used to perform the operation S120 described above, which will not be repeated here.

[0103] The first training module 730 is used to train the first initial model using multiple sample CT images and corresponding PET cumulative signal images as a first training dataset, to obtain a first target model for predicting PET cumulative signals corresponding to different time windows. In one embodiment, the first training module 730 can be used to perform the operation S130 described above, which will not be repeated here.

[0104] The second training module 740 is used to train the second initial model using dose absorption information, the output of the first target model, and a portion of the training data in the first training dataset as a second training set, to obtain a second target model for predicting dose distribution. In one embodiment, the second training module 740 can be used to perform the operation S140 described above, which will not be repeated here.

[0105] The acquisition module 750 is used to sequentially input the target CT image and the PET cumulative signal image of the target time window into the first target model and the second target model to obtain the dose information output by the second target model. In one embodiment, the first training module 750 can be used to perform the operation S150 described above, which will not be repeated here.

[0106] According to embodiments of the present invention, any plurality of modules among the determining module 710, generating module 720, first training module 730, second training module 740, and obtaining module 750 may be combined into one module, or any one of these modules may be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules may be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of the present invention, at least one of the determining module 710, generating module 720, first training module 730, second training module 740, and obtaining module 750 may be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in hardware or firmware, or in any one of software, hardware, and firmware implementations, or in a suitable combination of any of these. Alternatively, at least one of the determining module 710, generating module 720, first training module 730, second training module 740, and obtaining module 750 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0107] Figure 8 A block diagram of an electronic device for a dose determination method based on time-series signals according to an embodiment of the present invention is shown schematically.

[0108] like Figure 8 As shown, an electronic device 800 according to an embodiment of the present invention includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage portion 808 into a random access memory (RAM) 803. The processor 801 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 801 may also include onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0109] RAM 803 stores various programs and data required for the operation of electronic device 800. Processor 801, ROM 802, and RAM 803 are interconnected via bus 804. Processor 801 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 802 and / or RAM 803. It should be noted that the programs may also be stored in one or more memories other than ROM 802 and RAM 803. Processor 801 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0110] According to an embodiment of the present invention, the electronic device 800 may further include an input / output (I / O) interface 805, which is also connected to a bus 804. The electronic device 800 may also include one or more of the following components connected to the input / output (I / O) interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output (I / O) interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 810 as needed so that computer programs read from it can be installed into the storage section 808 as needed.

[0111] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0112] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, a computer-readable storage medium may include ROM 802 and / or RAM 803 and / or one or more memories other than ROM 802 and RAM 803 described above.

[0113] Embodiments of the present invention also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the dose determination method based on time-series signals provided in the embodiments of the present invention.

[0114] When the computer program is executed by the processor 801, it performs the functions defined in the system / apparatus of this invention. According to embodiments of the invention, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0115] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 809, and / or installed from a removable medium 811. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0116] In such an embodiment, the computer program can be downloaded and installed from a network via communication section 809, and / or installed from removable medium 811. When the computer program is executed by processor 801, it performs the functions defined in the system of this embodiment of the invention. According to embodiments of the invention, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0117] According to embodiments of the present invention, program code for executing the computer programs provided in the embodiments of the present invention can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0118] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0119] Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

[0120] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A dose determination method based on time-series signals, characterized in that, include: Multiple sample CT images were acquired, and the dose distribution information and radionuclide distribution information corresponding to each sample CT image were determined through Monte Carlo particle transport simulation. For each CT image sample, multiple preset time points are used as signal statistics windows, and PET cumulative signal images corresponding to each signal statistics window are generated based on the nuclide distribution information in each signal statistics window. The first initial model is trained using the multiple sample CT images and the corresponding PET cumulative signal images as the first training dataset to obtain a first target model for predicting PET cumulative signal images corresponding to different time windows. The dose distribution information, the output of the first target model, and a portion of the training data in the first training dataset are used as the second training set to train the second initial model, thereby obtaining the second target model for predicting the dose distribution. The target CT image and the cumulative PET signal image of the target time window are sequentially input into the first target model and the second target model to obtain the dose information output by the second target model.

2. The method according to claim 1, characterized in that, The process of acquiring multiple sample CT images and determining the dose distribution information and radionuclide distribution information corresponding to each sample CT image through Monte Carlo particle transport simulation includes: Monte Carlo particle transport simulation technology was used to simulate the transport process of particles in tissues under different energy, energy divergence and broadening conditions. For each simulated scenario, calculate the dose distribution information and radionuclide distribution information corresponding to the sample CT image in that simulated scenario.

3. The method according to claim 1, characterized in that, The step of using multiple preset time points as signal statistical windows and generating PET cumulative signal images corresponding to each signal statistical window based on the nuclide distribution information in each signal statistical window includes: Multiple time points are set as signal statistics windows based on the half-life of the nuclide; The signal attenuation is calculated based on the number of nuclides remaining at each time point to obtain the number of photon pairs detected at each time point; The detection signal is determined based on the number of photon pairs, and a PET signal image is generated based on the detection signal. Integrate the PET signal images at multiple times within each time window to generate the PET cumulative signal image corresponding to each signal statistical window.

4. The method according to claim 2, characterized in that, The method further includes: For each simulation scenario, the sample CT image, the PET cumulative signal image of the first time window, and the beam information are used as input data, and the PET cumulative signal image of the remaining time window is used as output data to construct the sample pair corresponding to the simulation scenario; the first training dataset includes multiple sample pairs, and the beam information includes incident energy, beam intensity, and beam broadening information.

5. The method according to claim 4, characterized in that, The first initial model is trained using the plurality of sample CT images and the PET cumulative signal images as a first training dataset to obtain a first target model for predicting PET cumulative signal images corresponding to different time windows, including: The sample CT image, the PET cumulative signal image of the first time window, and the beam information are used as inputs, and the PET cumulative signal image of the remaining time window is used as output. The first initial model is trained through supervised learning to obtain the first target model with the ability to predict the temporal evolution of subsequent PET cumulative signal images.

6. The method according to claim 5, characterized in that, The step of using the dose distribution information, the output of the first target model, and a portion of the training data from the first training dataset as a second training set to train the second initial model includes: Using sample CT images, PET cumulative signal images of the first time window, PET cumulative signal images of the remaining time window predicted by the first target model, and beam information as inputs, and the dose distribution information as outputs, the second initial model is trained through supervised learning. This allows the second initial model to learn the nonlinear mapping relationship between the temporal evolution characteristics of the PET cumulative signal images, tissue density characteristics, and the spatial distribution of dose deposition, thus obtaining the second target model for predicting dose distribution.

7. The method according to claim 1, characterized in that, The step of sequentially inputting the target CT image and the PET cumulative signal image of the target time window into the first target model and the second target model to obtain the dose information output by the second target model includes: The target CT image, the PET cumulative signal image of the target time window, and the beam information are input as the first input data to the first target model, so that the first target model can predict the PET cumulative signal images of each future time window based on the first input data. The PET cumulative signal image output by the first target model and the first input data are used as the second input data and input to the second target model so that the second target model can determine the target dose distribution information based on the second input data.

8. A dose determination device based on time-series signals, characterized in that, include: The determination module is used to acquire multiple sample CT images and determine the dose distribution information and nuclide distribution information of each sample CT image through Monte Carlo particle transport simulation. The generation module is used to generate PET cumulative signal images corresponding to each signal statistical window based on the nuclide distribution information in each signal statistical window for each sample CT image, using multiple preset time points as signal statistical windows. The first training module is used to train the first initial model using the multiple sample CT images and the corresponding PET cumulative signal images as the first training dataset to obtain a first target model for predicting PET cumulative signal images corresponding to different time windows. The second training module is used to train the second initial model using the dose distribution information, the output of the first target model, and a portion of the training data in the first training dataset as the second training set, so as to obtain a second target model for predicting the dose distribution. The acquisition module is used to sequentially input the target CT image and the cumulative PET signal image of the target time window into the first target model and the second target model to obtain the dose information output by the second target model.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The method is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

Images