Evaluation method and system of proton therapy and storage medium

Abstract

The invention discloses a proton therapy evaluation method and system, a storage medium and a computer program product, and relates to the field of data processing. The evaluation method comprises the following steps: forming a data set by data pairs of a historical proton beam delivery dose distribution image and a historical positron nuclide distribution image; constructing a first model, and training the first model by using the data set to obtain a second model; inputting the positron nuclide distribution image in the actual proton therapy into the second model to obtain a proton beam delivery dose distribution image in the actual proton therapy; and extracting the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image in the actual proton therapy to obtain the range information of the proton beam in the actual proton therapy. According to the method, the delivery condition of the planned dose can be directly reflected, Monte Carlo simulation does not need to be carried out on different patients to obtain reference positron nuclide distribution images, time and computing resources are saved, and each treatment process can be verified.

Description

Technical Field

[0001] The present application relates to the field of data processing, and specifically to a proton therapy evaluation method, system, storage medium and computer program product. Background Art

[0002] During proton therapy, a proton beam forms a Bragg peak at the end of its range when it passes through matter to deposit energy. By utilizing the characteristics of the Bragg peak dose of the proton beam, the energy of the proton beam can be adjusted so that its Bragg peak falls within the tumor target area, and the main dose is deposited at the tumor site to reduce damage to normal tissue. However, the Bragg peak dose characteristics of the proton beam also make proton therapy more sensitive to the uncertainty of its various links. The Bragg peak erroneously falling outside the tumor target area will lead to insufficient dose at the tumor site and increased dose to normal tissue. Therefore, the current clinical practice uses the method of setting a safety margin to reduce the impact of proton therapy uncertainty, but this fails to fully utilize the dosimetric advantages of proton therapy. Therefore, it is necessary to provide the actual delivered dose distribution through in vivo monitoring technology to ensure the precise implementation of proton therapy. On the one hand, successful in vivo monitoring can confirm the correct delivery of doses in tumor conformal therapy and improve treatment confidence. On the other hand, in vivo monitoring can also promptly identify the error between the planned dose and the actual dose delivery, which can promptly prevent the wrong dose delivery during treatment and support the adaptive treatment of dose-guided radiotherapy.

[0003] It is known that during proton therapy, the nuclear reaction between the proton beam and biological tissue will produce a series of secondary particles, such as positron nuclides (such as radioactive isotopes of oxygen, carbon, and nitrogen). The positron nuclides decay and then produce positrons, and the positron annihilation produces a pair of gamma photons. In clinical practice, PET (Position Emission Tomography, referred to as positron emission computed tomography) can image the positron nuclides induced by protons, thereby achieving the purpose of proton therapy monitoring.

[0004] In the prior art, the range verification of proton therapy is often performed by comparing the positron nuclide distribution image measured by PET with the reference positron nuclide distribution image. According to different methods for obtaining the reference distribution, such range verification methods can be subdivided into two types: the first is to perform Monte Carlo simulation or analytical calculation of an ideal positron nuclide distribution image corresponding to the treatment plan based on the treatment plan, and use this ideal nuclide distribution as a reference distribution to compare with the measured positron nuclide distribution image, which can indirectly verify whether the range of the proton beam meets expectations. The second is to compare the positron nuclide distribution image measured by the first PET with the positron nuclide distribution image measured by PET during subsequent treatments for fractionated treatments to verify the consistency of the proton beam range during multiple treatments.

[0005] Although the first method mentioned above has made good progress by comparing the positron nuclide distribution images measured by PET, this comparison method requires a known reference positron nuclide distribution image, and the results obtained are not intuitive enough and cannot directly reflect the delivery of the planned dose, which is not convenient for doctors and physicists to use as a reference for the next treatment. In addition, it takes a lot of time and computing resources to perform a specific Monte Carlo simulation on each patient to obtain a reference positron nuclide distribution image. The second method mentioned above can only compare whether the treatments are consistent, but cannot verify whether each treatment is correct.

[0006] The content of the background technology description is only for facilitating understanding of the relevant technology in this field and is not regarded as an admission of the prior art. Summary of the invention

[0007] Therefore, the present application intends to provide a proton therapy evaluation method, system, storage medium and computer program product, which can directly reflect the delivery of the planned dose, so that doctors and physicists can use it as a reference for the next treatment. There is no need to perform Monte Carlo simulation to obtain reference positron nuclide distribution images for different patients, which saves time and computing resources, and can verify each treatment process.

[0008] In a first aspect, a proton therapy evaluation method is provided, the evaluation method comprising:

[0009] The data pairs of the historical proton beam delivery dose distribution image and the historical positron nuclide distribution image are formed into a data set;

[0010] Constructing a first model, and using the data set to train the first model to obtain a second model;

[0011] Inputting the positron nuclide distribution image in actual proton therapy into the second model to obtain the proton beam delivery dose distribution image in actual proton therapy;

[0012] The falling edge at the far end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy is extracted to obtain the range information of the proton beam in actual proton therapy.

[0013] In an embodiment of the present application, the step of using the data set to train the first model to obtain the second model includes:

[0014] Extracting a plurality of one-dimensional curves from the historical positron nuclide distribution image, converting the one-dimensional curves into a first vector and using the first vector as an input of the first model;

[0015] Converting the first vector into a plurality of sequence feature vectors having a first preset length through a first convolutional layer of the first model;

[0016] Adding position information to the sequence feature vector by a first encoder of the first model;

[0017] The first decoder of the first model successively performs: restoring the length of the sequence feature vector to be the same as that of the first vector, reducing the number of channels of the sequence feature vector, concatenating the sequence feature vectors from different layers, and fusing the sequence feature vectors from different layers to obtain an output proton beam delivery dose distribution image;

[0018] The parameters of the first model are corrected according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

[0019] In an embodiment of the present application, the fusing of the sequence feature vectors of different layers performed by the first decoder includes: fusing the sequence feature vectors of different layers by adopting a top-down fusion mechanism through the first decoder.

[0020] In an embodiment of the present application, the first decoder is a decoder based on CNN and Tansformer.

[0021] In an embodiment of the present application, the step of using the data set to train the first model to obtain the second model includes:

[0022] Inputting the historical positron nuclide distribution image into the first model;

[0023] extracting three-dimensional image features of the historical positron nuclide distribution image through a second encoder of the first model, gradually reducing the spatial size of the three-dimensional image features, and increasing the abstract representation of the three-dimensional image features;

[0024] Restoring the spatial resolution of the three-dimensional image features by a second decoder of the first model, and combining the feature map of the second encoder by a splicing operation during decoding to obtain an output proton beam delivery dose distribution image;

[0025] The parameters of the first model are corrected according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

[0026] In an embodiment of the present application, the historical proton beam delivery dose distribution image is obtained by Monte Carlo simulation; and / or,

[0027] The historical positron nuclide distribution image is obtained by PET measurement.

[0028] In an embodiment of the present application, inputting the positron nuclide distribution image in actual proton therapy into the second model to obtain the proton beam delivery dose distribution image in actual proton therapy includes:

[0029] At the beginning of the actual proton therapy, PET is used to collect positron nuclide signals, and positron nuclide signal processing and image reconstruction are performed to obtain a positron nuclide distribution image in the actual proton therapy.

[0030] In the embodiment of the present application, the step of extracting the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy to obtain the range information of the proton beam in actual proton therapy includes:

[0031] The depth corresponding to any point in the range of 50% to 80% of the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image in the actual proton therapy is used as the range information of the proton beam in the actual proton therapy.

[0032] In a second aspect, a proton therapy evaluation system is provided, the evaluation system comprising:

[0033] A data set construction unit configured to form a data set from a data pair of a historical proton beam delivery dose distribution image and a historical positron nuclide distribution image;

[0034] A model building and training unit, configured to build a first model, and use the data set to train the first model to obtain a second model;

[0035] a proton beam delivery dose estimation unit configured to input the positron nuclide distribution image in actual proton therapy into the second model to obtain the proton beam delivery dose distribution image in actual proton therapy;

[0036] The proton beam range estimation unit is configured to extract the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy to obtain the range information of the proton beam in actual proton therapy.

[0037] In an embodiment of the present application, the first model includes a one-dimensional model, and the one-dimensional model includes a first convolutional layer, a first encoder, and a first decoder;

[0038] The model building and training unit is configured to extract a plurality of one-dimensional curves from the historical positron nuclide distribution image, convert the one-dimensional curves into a first vector and use the first vector as an input of the first model;

[0039] The first convolutional layer is configured to convert the first vector into a plurality of sequence feature vectors having a first preset length;

[0040] The first encoder is configured to add position information to the sequence feature vector;

[0041] The first decoder is configured to successively perform: restoring the length of the sequence feature vector to be the same as that of the first vector, reducing the number of channels of the sequence feature vector, concatenating the sequence feature vectors from different layers, and fusing the sequence feature vectors from different layers to obtain an output proton beam delivery dose distribution image;

[0042] The model building and training unit is further configured to correct the parameters of the first model according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

[0043] In an embodiment of the present application, the first decoder is configured to fuse the sequence feature vectors of different layers using a top-down fusion mechanism.

[0044] In an embodiment of the present application, the first decoder is a decoder based on CNN and Tansformer.

[0045] In an embodiment of the present application, the first model includes a second encoder and a second decoder;

[0046] The model building and training unit is configured to extract three-dimensional image features from historical positron nuclide distribution images and input the three-dimensional image features into the first model;

[0047] The second encoder is configured to gradually reduce the spatial size of the three-dimensional image features and increase the abstract representation of the three-dimensional image features;

[0048] The second decoder is configured to restore the spatial resolution of the image features and combine the feature map of the second encoder through a splicing operation during decoding to obtain an output proton beam delivery dose distribution image;

[0049] The model building and training unit is also configured to correct the parameters of the first model according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

[0050] In an embodiment of the present application, the evaluation system further includes: a data acquisition and generation unit, configured to use PET to acquire positron nuclide signals at the beginning of the actual proton therapy, and to process and reconstruct the positron nuclide signals to obtain a positron nuclide distribution image in the actual proton therapy.

[0051] In an embodiment of the present application, the proton beam range estimation unit is configured to use the depth corresponding to any point between 50% and 80% of the falling edge of the distal end of the Bragg peak of the proton beam delivery dose distribution image in the actual proton therapy as the range information of the proton beam in the actual proton therapy.

[0052] According to a third aspect, a storage medium is provided, wherein the storage medium stores a computer program, wherein the computer program is configured to implement the proton therapy evaluation method according to the first aspect when executed.

[0053] In a fourth aspect, a computer program product is provided, comprising a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the proton therapy evaluation method described in the first aspect is implemented.

[0054] The proton therapy evaluation method provided in the present application utilizes historical data to train a model. The trained model can directly estimate the proton beam delivery dose distribution image by relying solely on the positron nuclide distribution image measured in actual proton therapy, and thereby estimate the range information of the proton beam. The result is very intuitive and can directly reflect the delivery of the planned dose, so that doctors and physicists can use it as a reference for the next step of treatment. There is no need to perform Monte Carlo simulation to obtain reference positron nuclide distribution images for different patients, which saves time and computing resources, and can verify each treatment process.

[0055] Part of the optional features and other effects of the embodiments of the present application are described below, and part of them can be understood by reading this document. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The present application will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same reference numerals represent the same structure, wherein:

[0057] Figure 1 A flow chart of a proton therapy evaluation method according to an embodiment of the present application is shown;

[0058] Figure 2 Another flow chart of a proton therapy evaluation method according to a specific embodiment of the present application is shown;

[0059] Figure 3 A schematic diagram of a one-dimensional neural network model used in a proton therapy evaluation method according to an embodiment of the present application is shown;

[0060] Figure 4 A one-dimensional neural network model architecture diagram used in a proton therapy evaluation method according to a specific embodiment of the present application is shown;

[0061] Figure 5 An architectural diagram of a first decoder of a one-dimensional neural network model used in a proton therapy evaluation method according to an embodiment of the present application is shown;

[0062] Figure 6 Another flow chart of a proton therapy evaluation method according to an embodiment of the present application is shown;

[0063] Figure 7 A schematic diagram of a three-dimensional neural network model used in a proton therapy evaluation method according to a specific embodiment of the present application is shown;

[0064] Figure 8 A three-dimensional neural network model architecture diagram used in a proton therapy evaluation method according to a specific embodiment of the present application is shown;

[0065] Fig. 9 A one-dimensional curve showing the distribution of positron nuclides used in a proton therapy evaluation method according to a specific example of the present application is shown;

[0066] Fig.10 The figure shows a proton beam delivery dose distribution curve obtained by a proton therapy evaluation method according to a specific example of the present application;

[0067] Fig.11 A one-dimensional curve showing the distribution of positron nuclides used in a proton therapy evaluation method according to another specific example of the present application;

[0068] Fig.12 is a proton beam delivery dose distribution curve obtained according to a proton therapy evaluation method of another specific example of the present application;

[0069] Fig.13 A block diagram of a proton therapy evaluation system according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0070] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present application, so the present application is not limited by the specific embodiments disclosed below.

[0071] It should be noted that when an element is referred to as "fixed to" another element, it can be directly on another element or there can also be a centered element. When an element is considered to be "connected" to another element, it can be directly connected to another element or there may be a centered element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are only for illustrative purposes. The described features, structures or characteristics can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided so as to give a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical solution of the present application can be practiced in the absence of some specific details, or the technical solution of the present application can be practiced in other ways, components, materials, devices or operations. In these cases, known structures, methods, devices, implementations, materials or operations will not be shown or described in detail.

[0072] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0073] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The term "and / or" or "and / or" includes any and all combinations of one or more related listed items.

[0074] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art. The terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present application.

[0075] As mentioned above, in the technologies known to the inventors, the two existing in vivo monitoring methods in proton therapy have the following problems: the results are not intuitive enough, cannot directly reflect the delivery of the planned dose, and are not convenient for doctors and physicists to use as a reference for the next treatment; or, cannot verify whether each treatment is correct. In this regard, the present application provides a proton therapy evaluation method, which is an in vivo proton therapy monitoring technology. By online estimating the delivery dose distribution and range of the proton beam in the human body, the accuracy of the proton therapy plan is verified, and the problem of inaccurate tumor deposition dose caused by the uncertainty of the proton beam range in proton therapy can be solved.

[0076] Figure 1 A flow chart of the proton therapy evaluation method according to an embodiment of the present application is shown. In order to more clearly demonstrate the proton therapy evaluation method provided by the present application, Figure 1 The proton therapy evaluation method generally includes steps S100 to S400, and the specific details are as follows.

[0077] Step S100: forming a data set from a data pair of a historical proton beam delivery dose distribution image and a historical positron nuclide distribution image.

[0078] The data set includes a plurality of data pairs, each of which includes a historical proton beam delivery dose distribution image and a historical positron nuclide distribution image, and the historical proton beam delivery dose distribution image and the historical positron nuclide distribution image are in a one-to-one correspondence.

[0079] Wherein, the historical proton beam delivery dose distribution image is obtained through Monte Carlo simulation.

[0080] The historical positron nuclide distribution image is obtained by PET measurement, and the PET may be a flat PET, an arc PET or an annular opening PET.

[0081] Step S200: construct a first model, and use the data set to train the first model to obtain a second model.

[0082] Step S200 involves the construction and training of the first model, wherein the first model may be a neural network model having a strong learning ability, and the intrinsic connection between the proton beam delivery dose distribution image and the positron nuclide distribution image may be learned through the training of the data set.

[0083] In the examples of the present application, the first model may be a one-dimensional model or a three-dimensional model. The one-dimensional model and the three-dimensional model will be respectively introduced with two specific examples below.

[0084] Figure 2Another flow chart of the proton therapy evaluation method according to a specific embodiment of the present application is shown. Figure 2 As shown, the first model is a one-dimensional model, and the step S200 can be further broken down into steps S211 to S215, as described below.

[0085] Step S211: extracting a plurality of one-dimensional curves from the historical positron nuclide distribution image, converting the one-dimensional curves into a first vector and using the first vector as an input of the first model.

[0086] Figure 3 A schematic diagram of a one-dimensional neural network model used in a proton therapy evaluation method according to an embodiment of the present application is shown. Figure 4 The one-dimensional neural network model architecture diagram used in the proton therapy evaluation method according to a specific embodiment of the present application is shown. Figure 3 , extract information from the three-dimensional, historical positron nuclide distribution image to obtain a plurality of the one-dimensional curves, and use the one-dimensional curves as the first model ( Figure 3 The input of the neural network in . Figure 4 , exemplarily, the one-dimensional curve ( Figure 4 The 1D positron nuclide distribution curve in ( ) is input into the first model as a 204×1 first vector, which is represented as a vertical black line.

[0087] Step S212: converting the first vector into a plurality of sequence feature vectors having a first preset length through the first convolutional layer of the first model.

[0088] Continue to see Figure 4 , exemplarily, the convolution kernel size and step size of the first convolution layer 410 are both 3, and the 204×1 first vector is divided into sequence feature vectors with a length of 68 (the first preset length), such as Figure 4 The sequence feature vectors are shown represented as 68×128 boxes.

[0089] Step S213: adding position information to the sequence feature vector through the first encoder of the first model.

[0090] The position information may be a vector, and a vector is added to the sequence feature vector. The vectors representing the position information added to the sequence feature vectors of different lengths may be different, thereby realizing position encoding.

[0091] For example, the number of the first encoders 420 is 6, and the 6 first encoders 420 are connected in series, and the output of the previous first encoder 420 is used as the input of the next first encoder 420, and finally the last first encoder 420 combines the outputs of all the first encoders 420 into a single pixel. 1 ~z 6 Output all.

[0092] Step S214: The first decoder of the first model successively executes: restoring the length of the sequence feature vector to the same as that of the first vector, reducing the number of channels of the sequence feature vector, splicing the sequence feature vectors from different layers, and fusing the sequence feature vectors from different layers to obtain an output proton beam delivery dose distribution image.

[0093] Continue to see Figure 4 , the above z 1 ~z 6 As input to the first decoder 430 .

[0094] Figure 5 FIG. 1 shows an architecture diagram of a first decoder of a one-dimensional neural network model used in a proton therapy evaluation method according to an embodiment of the present application. For example, Figure 5 As shown, corresponding to the first encoder 420, the number of the first decoders 430 is also 6. The first 5 first decoders 430 include a deconvolution layer 431, a splicing layer 432, a second convolution layer 433 and a rectified linear unit layer 434, and the last first decoder 430 includes a deconvolution layer. The convolution kernel size of the deconvolution layer 431 is 3. The deconvolution layer 431 is used to restore the length of the sequence feature vector to the same as the first vector, and at the same time, the number of channels of the sequence feature vector is reduced to half of the original number to ensure that the number of channels of the sequence feature vectors from different layers remains unchanged after splicing. The sequence feature vectors from different layers are spliced ​​through the splicing layer 432. The convolution kernel size of the second convolution layer 433 is 3. The spliced ​​sequence feature vectors are fused through the second convolution layer 433 and the rectified linear unit layer 434. As shown Figure 4 As shown, the rectified linear unit layer 434 outputs a 204×64 vector, which is a square box.

[0095] Exemplarily, the first model also includes a third convolution layer 440 with a step size of 1 and a convolution kernel of 3. The third convolution layer 440 processes the output of the first decoder to obtain a one-dimensional output proton beam delivery dose distribution image of 204×1.

[0096] Preferably, the fusing of the sequence feature vectors of different layers performed by the first decoder includes: fusing the sequence feature vectors of different layers by adopting a top-down fusion mechanism through the first decoder, which can improve the ability to obtain local information and reduce the network scale.

[0097] Specifically, in order to enhance the decoder's ability to integrate features of different depths, a top-down feature fusion mechanism is designed. For example, Figure 5 The top-down fusion mechanism is shown. Figure 5 As shown, feature fusion is performed from the sixth first decoder 430 to the first first decoder 430. 6 After the deconvolution layer 431 in the sixth first decoder, a 204×64 feature vector is output. Similarly, Z5 after the deconvolution layer 431 in the fifth first decoder outputs a 204×64 feature vector. The above two 204×64 feature vectors are concatenated by the concatenation layer 432 to output a 204×128 feature vector. This 204×128 feature vector is passed through the second convolution layer 433 and the rectified linear unit layer 434 to output a 204×64 feature vector. This 204×64 feature vector is the same as Z 4 The 204×64 feature vector output by the deconvolution layer 431 in the fourth first decoder is passed through the concatenation layer 432 to output a 204×128 feature vector. The same process is repeated in sequence, and finally a 204×64 feature vector is obtained. 1 After the deconvolution layer 431 in the first decoder outputs a 204×64 feature vector, after the concatenation layer 432 outputs a 204×128 feature vector, and then after the second convolution layer 433 and the rectified linear unit layer 434 outputs a 204×64 feature vector.

[0098] Step S215: Correcting the parameters of the first model according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

[0099] Taking the historical proton beam delivery dose distribution image corresponding to the historical positron nuclide distribution image as a benchmark, the difference between the output proton beam delivery dose distribution image correction and the historical proton beam delivery dose distribution image is analyzed, and the parameters of the first model are adjusted according to the difference to obtain the second model.

[0100] Figure 6 Another flow chart of the proton therapy evaluation method according to an embodiment of the present application is shown. Figure 6 The first model is a three-dimensional model, and the step S200 can be further broken down into steps S221 to S224, as described below.

[0101] Step S221: inputting the historical positron nuclide distribution image into the first model.

[0102] Figure 7 A schematic diagram of a three-dimensional neural network model used in a proton therapy evaluation method according to a specific embodiment of the present application is shown. Figure 8 The following is a diagram showing the architecture of a three-dimensional neural network model used in a proton therapy evaluation method according to a specific embodiment of the present application. Figure 7 and Figure 8 , the first model can be a three-dimensional U-Net neural network model, which uses the three-dimensional image features of the positron nuclide distribution image as the neural network input and outputs a three-dimensional proton beam delivery dose distribution image. The three-dimensional U-Net neural network model mainly includes a second encoder 810 and a second decoder 820. Exemplarily, the three-dimensional U-Net neural network model includes 3 layers, and the number of channels in each layer is 64, 128, and 256 respectively. Of course, it can be understood that the number of layers of the three-dimensional U-Net neural network model can also be 4 layers, 5 layers, or other layers, and the number of channels can also be other.

[0103] Step S222: extracting the three-dimensional image features of the historical positron nuclide distribution image through the second encoder of the first model, gradually reducing the spatial size of the three-dimensional image features, and increasing the abstract representation of the three-dimensional image features.

[0104] Exemplarily, the second encoder includes multiple fourth convolutional layers and multiple pooling layers, and the spatial size of the three-dimensional image features is gradually reduced and the abstract representation of the three-dimensional image features is increased through the multiple fourth convolutional layers and the multiple pooling layers.

[0105] Step S223: The spatial resolution of the three-dimensional image features is restored by the second decoder of the first model, and the feature map of the second encoder is combined by a splicing operation during the decoding process to obtain an output proton beam delivery dose distribution image.

[0106] Exemplarily, the second decoder includes an upsampling layer, through which the spatial resolution of the three-dimensional image features is restored, and during the decoding process, the feature map of the second encoder is combined through a splicing operation to retain high-resolution information.

[0107] Step S224: Correct the parameters of the first model according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

[0108] More specifically, Figure 8 For example, It means that the feature map of the previous level is processed by a three-dimensional convolution layer (3Dconv) and a rectified linear unit layer (ReLU). It means that the feature map of the previous level is processed by a 3D MaxPooling layer. It means that the feature map of the previous level is processed by an upsampling layer (UP-conv). It means that the feature map of the previous level and the feature map output by the upsampling layer (UP-conv) are concatenated through the concatenation layer (Concatenate). The following is an example to elaborate on the whole process of the three-dimensional model taking the positron nuclide distribution image as input and the proton beam delivery dose distribution image as output.

[0109] Step 1, assuming that the image size of the historical, three-dimensional (3D) positron nuclide distribution image is 128×128×128, and the initial channel is 1;

[0110] Step 2: The positron nuclide distribution image is processed by a three-dimensional convolution layer (3D conv) and a rectified linear unit layer (ReLU) to be transformed into a 128×128×128×64 feature map, that is, a 128×128×128 feature map with 64 channels.

[0111] Step 3: After being processed by the 3D MaxPooling layer, it is transformed into a 64×64×64×64 feature map, that is, a 64×64×64 feature map with 64 channels, where the pooling parameters of the 3D MaxPooling layer are 2,2,2;

[0112] Step 4: After further processing through a 3D convolution layer (3D conv) and a rectified linear unit layer (ReLU), the resulting image is transformed into a 64×64×64×128 feature map.

[0113] Step 5: After being processed by the 3D MaxPooling layer, it is transformed into a 32×32×32×128 feature map;

[0114] In step 6, the image is further processed by a 3D convolution layer (3D conv) and a rectified linear unit layer (ReLU) to transform it into a 32×32×32×256 feature map.

[0115] Step 7: After an upsampling layer (UP-conv), it is transformed into a 64×64×64×128 feature map. The upsampling layer increases the size of the feature map, making each dimension twice as large and halving the number of channels.

[0116] In step 8, the 64×64×64×128 feature map output by the upsampling layer and the 64×64×64×128 feature map output by step 4 are concatenated through a concatenation layer to form a 64×64×64×256 feature map.

[0117] In step 9, the image is further processed by a 3D convolution layer (3D conv) and a rectified linear unit layer (ReLU) to transform it into a 64×64×64×128 feature map.

[0118] Step 10: After an upsampling layer (UP-conv), it is transformed into a 128×128×128×64 feature map;

[0119] In step 11, the 128×128×128×64 feature map output by the upsampling layer and the 128×128×128×64 feature map output by step 2 are concatenated through a concatenation layer to form a 128×128×128×128 feature map.

[0120] In step 12, the image is further processed by a 3D convolution layer (3D conv) and a rectified linear unit layer (ReLU) to transform it into a 128×128×128×64 feature map.

[0121] In step 13, the image is finally transformed into a 128×128×128×1 feature map after being processed by a three-dimensional convolution layer (3D conv) and a rectified linear unit layer (ReLU), which is the output of the three-dimensional model - the proton beam delivery dose distribution image.

[0122] As described above, the first model constructed in the present application can be a one-dimensional model or a three-dimensional model. When using a one-dimensional model, a one-dimensional neural network model based on CNN and Transformer can be selected, with the Transformer architecture as the main body, and the decoder therein is replaced with a decoder under the CNN architecture, which can improve the ability to obtain local information and reduce the network scale, and adopt a top-down feature fusion mechanism in the decoder, which can enhance the ability to integrate features of different depths. When using a three-dimensional model, a three-dimensional U-Net neural network model can be selected. The three-dimensional model can be used to directly process the three-dimensional positron nuclide distribution image without extracting it into one-dimensional data, and a three-dimensional proton beam delivery dose distribution image can be directly obtained. There is no need to correct and integrate the one-dimensional proton beam delivery dose distribution image into a three-dimensional one, which reduces the amount of data processing.

[0123] Step S300: inputting the positron nuclide distribution image in actual proton therapy into the second model to obtain the proton beam delivery dose distribution image in actual proton therapy.

[0124] Step S300 involves the application of the second model. Specifically, PET is used to collect positron nuclide signals at the beginning of the actual proton therapy, and positron nuclide signal processing and image reconstruction are performed to obtain a positron nuclide distribution image in the actual proton therapy.

[0125] It is understandable that step S300 may also be performed by using PET to collect positron nuclide signals after the actual proton therapy is completed, and performing positron nuclide signal processing and image reconstruction to obtain a positron nuclide distribution image in the actual proton therapy.

[0126] Step S400: extracting the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy to obtain the range information of the proton beam in actual proton therapy.

[0127] Specifically, step S400 uses the depth corresponding to any point in the range of 50% to 80% of the falling edge of the distal end of the Bragg peak of the proton beam delivery dose distribution image in the actual proton therapy as the range information of the proton beam in the actual proton therapy. Preferably, the depth corresponding to the 80% point of the falling edge of the distal end of the Bragg peak of the proton beam delivery dose distribution image in the actual proton therapy is used as the range information of the proton beam in the actual proton therapy.

[0128] It is understandable that the positron nuclide distribution image and the range information of the proton beam in the actual proton therapy are compared with the proton therapy plan to verify their accuracy.

[0129] Fig. 9 A one-dimensional curve of positron nuclide distribution used in a proton therapy evaluation method according to a specific example of the present application is shown. Fig.10 The figure shows a proton beam delivery dose distribution curve obtained by a proton therapy evaluation method according to a specific example of the present application. Fig.11 A one-dimensional curve of positron nuclide distribution used in a proton therapy evaluation method according to another specific example of the present application is shown. Fig.12 It is a proton beam delivery dose distribution curve obtained according to a proton therapy evaluation method according to another specific example of the present application. Fig. 9 and Fig.10 Correspondingly, Fig.11 and Fig.12 Corresponding. Fig. 9 and Fig.11 The graphs show the relationship between the positron activity (Activity), proton beam delivery depth (Depth), and density value (HU) measured by PET-CT after a proton beam is irradiated into the human body. Fig.10 and Fig.12The Bragg peak curves of a proton beam delivery dose distribution image after a proton beam is irradiated on a human body are shown respectively, wherein the blue curve is obtained by the second model, the red curve is calculated by Monte Carlo (real), the orange curve is the error curve between the proton beam delivery dose distribution obtained by the second model and the Monte Carlo calculation, MRE (mean relative error) refers to the error value between the proton beam delivery dose distribution obtained by the second model and the Monte Carlo calculation, ARE (absolute range error) refers to the absolute value of the error between the proton beam range obtained by the second model and the Monte Carlo calculation, the blue triangle is the 80% position of the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image obtained according to the second model, the red asterisk is the 80% position of the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image calculated by Monte Carlo, and it can be seen that the red asterisk and the blue triangle basically coincide, and the red asterisk is covered by the blue triangle.

[0130] The proton therapy evaluation method provided in the present application utilizes historical data to train a model. The trained model can directly estimate the proton beam delivery dose distribution image by relying solely on the positron nuclide distribution image measured in actual proton therapy, and thereby estimate the range information of the proton beam. The result is very intuitive and can directly reflect the delivery of the planned dose to provide feedback to doctors and physicists as a reference for the next step of treatment. There is no need to perform Monte Carlo simulation to obtain reference positron nuclide distribution images for different patients, which saves time and computing resources, and can verify each treatment process.

[0131] Corresponding to the above-mentioned proton therapy evaluation method, the present application also provides a proton therapy evaluation system. Fig.13 FIG. 4 is a block diagram of a proton therapy evaluation system according to an embodiment of the present application. Fig.13 As shown, the evaluation system includes: a data set construction unit 1310, a model construction and training unit 1320, a proton beam delivery dose estimation unit 1330 and a proton beam range estimation unit 1340.

[0132] Specifically, the data set construction unit configuration 1310 is configured to form a data set with a data pair of a historical proton beam delivery dose distribution image and a historical positron nuclide distribution image. The data set includes a plurality of data pairs, each of which includes a historical proton beam delivery dose distribution image and a historical positron nuclide distribution image, and the historical proton beam delivery dose distribution image and the historical positron nuclide distribution image are in a one-to-one correspondence. The historical proton beam delivery dose distribution image is obtained by Monte Carlo simulation. The historical positron nuclide distribution image is obtained by PET measurement, and the PET can be a flat PET, an arc PET, or a ring-shaped opening PET.

[0133] Specifically, the model building and training unit 1320 is configured to build a first model, and use the data set to train the first model to obtain a second model. The first model may be a neural network model, which has a strong learning ability, and can learn the intrinsic relationship between the proton beam delivery dose distribution image and the positron nuclide distribution image through the training of the data set.

[0134] In the examples of the present application, the first model may be a one-dimensional model or a three-dimensional model. The one-dimensional model and the three-dimensional model will be respectively introduced with two specific examples below.

[0135] In one example, the first model includes a one-dimensional model, and the one-dimensional model includes a first convolution layer, a first encoder and a first decoder. The model building and training unit is configured to extract multiple one-dimensional curves from the historical positron nuclide distribution image, convert the one-dimensional curves into a first vector and use it as the input of the first model. The first convolution layer is configured to convert the first vector into multiple sequence feature vectors with a first preset length. The first encoder is configured to add position information to the sequence feature vector. Among them, the position information can be a vector, and a vector is added to the sequence feature vector. The vector representing the position information added to the sequence feature vectors of different lengths can be different, thereby realizing position encoding. The first decoder is configured to successively execute: restore the length of the sequence feature vector to the same as that of the first vector, reduce the number of channels of the sequence feature vector, splice the sequence feature vectors from different layers, and fuse the sequence feature vectors of different layers to obtain the output proton beam delivery dose distribution image. The model building and training unit is also configured to correct the parameters of the first model according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

[0136] Figure 3 A schematic diagram of a one-dimensional neural network model used in a proton therapy evaluation method according to an embodiment of the present application is shown. Figure 4The one-dimensional neural network model architecture diagram used in the proton therapy evaluation method according to a specific embodiment of the present application is shown. Figure 3 , extract information from the three-dimensional, historical positron nuclide distribution image to obtain a plurality of the one-dimensional curves, and use the one-dimensional curves as the first model ( Figure 3 The input of the neural network in . Figure 4 , exemplarily, the one-dimensional curve ( Figure 4 The 1D positron nuclide distribution curve in ( ) is input into the first model as a 204×1 first vector, which is represented as a vertical black line.

[0137] Continue to see Figure 4 , exemplarily, the convolution kernel size and step size of the first convolution layer 410 are both 3, and the 204×1 first vector is divided into sequence feature vectors with a length of 68 (the first preset length), such as Figure 4 The sequence feature vectors are shown represented as 68×128 boxes.

[0138] Exemplarily, the number of the first encoders 420 is 6, each of which may include 8 attention mechanism units arranged in parallel, and the 6 first encoders 420 are connected in series, and the output of the previous first encoder 420 is used as the input of the next first encoder 420, and finally the last first encoder 420 will combine the outputs of all the first encoders 420 into a single unit. 1 ~z 6 Output all.

[0139] Continue to see Figure 4 , the above z 1 ~z 6 As input to the first decoder 430 .

[0140] Figure 5 FIG. 1 shows an architecture diagram of a first decoder of a one-dimensional neural network model used in a proton therapy evaluation method according to an embodiment of the present application. For example, Figure 5As shown, corresponding to the first encoder 420, the number of the first decoders 430 is also 6. The first five first decoders 430 include a deconvolution layer 431, a splicing layer 432, a second convolution layer 433 and a rectified linear unit layer 434, and the last first decoder 430 includes a deconvolution layer. The convolution kernel size of the deconvolution layer 431 is 3. The deconvolution layer 431 is used to restore the length of the sequence feature vector to the same as the first vector, and at the same time, the number of channels of the sequence feature vector is reduced to half of the original number to ensure that the number of channels of the sequence feature vectors from different layers remains unchanged after splicing. The sequence feature vectors from different layers are spliced ​​through the splicing layer 432. The convolution kernel size of the second convolution layer 433 is 3. The spliced ​​sequence feature vectors are fused through the second convolution layer 433 and the rectified linear unit layer 434. As shown Figure 4 As shown, the rectified linear unit layer 434 outputs a 204×64 vector, which is represented as a square box.

[0141] Exemplarily, the first model also includes a third convolution layer 440 with a step size of 1 and a convolution kernel of 3. The third convolution layer 440 processes the output of the first decoder to obtain a one-dimensional output proton beam delivery dose distribution image of 204×1.

[0142] Preferably, the first decoder is configured to fuse the sequence feature vectors of different layers using a top-down fusion mechanism.

[0143] For example, Figure 5 The top-down fusion mechanism is shown. Specifically, the first decoder 430 in the front processes z first. 6 , and outputs the processing result to a subsequent first decoder 430, the subsequent first decoder 430 processes z5 and the result of the previous first decoder 430 together and then outputs the result to a subsequent first decoder 430, and so on, until the last first decoder 430 outputs the final result.

[0144] In another example, the first model is a three-dimensional model, which includes a second encoder and a second decoder. The model building and training unit is configured to extract three-dimensional image features from historical positron nuclide distribution images and input the three-dimensional image features into the first model. The second encoder is configured to gradually reduce the spatial size of the three-dimensional image features and increase the abstract representation of the three-dimensional image features. The second decoder is configured to restore the spatial resolution of the image features and combine the feature map of the second encoder through a splicing operation during the decoding process to obtain an output proton beam delivery dose distribution image. The model building and training unit is also configured to correct the parameters of the first model according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

[0145] Exemplarily, the second encoder includes multiple fourth convolutional layers and multiple pooling layers, and the spatial size of the three-dimensional image features is gradually reduced and the abstract representation of the three-dimensional image features is increased through the multiple fourth convolutional layers and the multiple pooling layers.

[0146] Exemplarily, the second decoder includes an upsampling layer, through which the spatial resolution of the three-dimensional image features is restored, and during the decoding process, the feature map of the second encoder is combined through a splicing operation to retain high-resolution information.

[0147] Figure 7 A schematic diagram of a three-dimensional neural network model used in a proton therapy evaluation method according to a specific embodiment of the present application is shown. Figure 8 The following is a diagram showing the architecture of a three-dimensional neural network model used in a proton therapy evaluation method according to a specific embodiment of the present application. Figure 7 and Figure 8 , the first model can be a three-dimensional U-Net neural network model, which uses the three-dimensional image features of the positron nuclide distribution image as the neural network input and outputs a three-dimensional proton beam delivery dose distribution image. The three-dimensional U-Net neural network model mainly includes a second encoder 810 and a second decoder 820. Exemplarily, the three-dimensional U-Net neural network model includes 3 layers, and the number of channels in each layer is 64, 128, and 256 respectively. Of course, it can be understood that the number of layers of the three-dimensional U-Net neural network model can also be 4 layers, 5 layers, or other layers, and the number of channels can also be other.

[0148] For a more detailed introduction to the three-dimensional U-Net neural network model, please refer to the example of the estimation method, which will not be repeated here.

[0149] As described above, the first model constructed in the present application can be a one-dimensional model or a three-dimensional model. When using a one-dimensional model, a one-dimensional neural network model based on CNN and Transformer can be selected, with the Transformer architecture as the main body, and the decoder therein is replaced with a decoder under the CNN architecture, which can improve the ability to obtain local information and reduce the network scale, and adopt a top-down feature fusion mechanism in the decoder, which can enhance the ability to integrate features of different depths. When using a three-dimensional model, a three-dimensional U-Net neural network model can be selected. The three-dimensional model can be used to directly process the three-dimensional positron nuclide distribution image without extracting it into one-dimensional data, and a three-dimensional proton beam delivery dose distribution image can be directly obtained. There is no need to correct and integrate the one-dimensional proton beam delivery dose distribution image into a three-dimensional one, which reduces the amount of data processing.

[0150] Specifically, the proton beam delivery dose estimation unit 1330 is configured to input the positron nuclide distribution image in the actual proton therapy into the second model to obtain the proton beam delivery dose distribution image in the actual proton therapy. More specifically, the evaluation system also includes: a data acquisition and generation unit, configured to use PET to collect positron nuclide signals at the beginning of the actual proton therapy, and perform processing and image reconstruction of the positron nuclide signals to obtain the positron nuclide distribution image in the actual proton therapy. It can be understood that the data acquisition and generation unit can also use PET to collect positron nuclide signals after the actual proton therapy ends, and perform positron nuclide signal processing and image reconstruction to obtain the positron nuclide distribution image in the actual proton therapy.

[0151] Specifically, the proton beam range estimation unit configuration 1340 is configured to extract the falling edge of the distal end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy to obtain the range information of the proton beam in actual proton therapy. More specifically, the proton beam range estimation unit is configured to use the depth corresponding to any site in the range of 50% to 80% of the falling edge of the distal end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy as the range information of the proton beam in actual proton therapy. Preferably, the proton beam range estimation unit 1340 is configured to use the depth corresponding to the 80% site of the falling edge of the distal end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy as the range information of the proton beam in actual proton therapy.

[0152] The proton therapy evaluation system provided in the present application utilizes historical data to train a model. The trained model can directly estimate the proton beam delivery dose distribution image by relying solely on the positron nuclide distribution image measured in actual proton therapy, and thereby estimate the range information of the proton beam. The result is very intuitive and can directly reflect the delivery of the planned dose to provide feedback to doctors and physicists as a reference for the next step of treatment. There is no need to perform Monte Carlo simulation to obtain reference positron nuclide distribution images for different patients, which saves time and computing resources, and can verify each treatment process.

[0153] Although not shown, in some embodiments, a computer-readable storage medium is also provided, storing a computer program, which is configured to execute the relevant steps of any method of the embodiment of the present application when it is run. The computer program includes various program modules / units constituting the device according to the embodiment of the present application, and the computer program composed of various program modules / units can realize the functions corresponding to the various steps in the method described in the above embodiment when it is executed. The computer program can also be run on an electronic device as described in the embodiment of the present application.

[0154] It should be understood that the units or modules of the embodiments of the present application can be implemented using semiconductors, especially integrated circuits. For example, the embodiments of the present application may also involve an integrated circuit including any unit or device according to the embodiments of the present application, which falls within the scope of the present application.

[0155] In some embodiments, the system and its modules of the present application can be implemented by hardware, software or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or a dedicated design hardware. Those skilled in the art will appreciate that the above methods and systems can be implemented using computer executable instructions and / or included in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The system and its modules of the present application can not only be implemented by hardware circuits such as ultra-large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., but can also be implemented by software such as executed by various types of processors, and can also be implemented by a combination of the above hardware circuits and software (e.g., firmware).

[0156] It should be noted that the above description of the modules is only for convenience of description and does not limit the present application to the scope of the embodiments. It is understandable that, after understanding the principle of the system, those skilled in the art may arbitrarily combine the modules or form a subsystem connected to other modules without deviating from the principle.

[0157] The basic concepts have been described herein. Obviously, for those skilled in the art, the above detailed disclosure is only for example and does not constitute a limitation of the present application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements and amendments to the present application. Such modifications, improvements and amendments are suggested in the present application, so such modifications, improvements and amendments still belong to the spirit and scope of the exemplary embodiments of the present application.

[0158] At the same time, the present application uses specific words to describe the embodiments of the present application. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of the present application. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or multiple times in different positions in the present application does not necessarily refer to the same embodiment. In addition, some features, structures or characteristics in one or more embodiments of the present application can be appropriately combined.

[0159] In addition, it will be appreciated by those skilled in the art that various aspects of the present application may be illustrated and described by a number of patentable categories or situations, including any new and useful process, machine, product or combination of substances, or any new and useful improvements thereto. Accordingly, various aspects of the present application may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may all be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". In addition, various aspects of the present application may be represented as a computer product located in one or more computer-readable media, which includes computer-readable program code.

[0160] A computer storage medium may include a propagated data signal containing computer program code, for example, in baseband or as part of a carrier wave. The propagated signal may be in a variety of forms, including electromagnetic, optical, etc., or a suitable combination. A computer storage medium may be any computer-readable medium other than a computer-readable storage medium, which can be connected to an instruction execution system, device or apparatus to communicate, propagate or transmit the program for use. The program code on the computer storage medium may be transmitted via any suitable medium, including radio, cable, fiber optic cable, RF, or similar media, or any combination of the above media.

[0161] The computer program coding required for the operation of each part of the application can be written in any one or more programming languages, including object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages ​​such as C language, Visual Basic, Fortran 2003, Perl, COBOL 2002, PHP, ABAP, dynamic programming languages ​​such as Python, Ruby and Groovy, or other programming languages, etc. The program coding can be run completely on the user's computer, or run on the user's computer as an independent software package, or run partly on the user's computer and partly on the remote computer, or run completely on the remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any network form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., via the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0162] In addition, unless explicitly stated in the claims, the order of the processing elements and sequences described in this application, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this application. Although the above disclosure discusses some invention embodiments that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this application. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only by software solutions, such as installing the described system on an existing server or mobile device.

[0163] Similarly, it should be noted that in order to simplify the description of the disclosure of this application and thus help understand one or more embodiments of the invention, in the above description of the embodiments of this application, multiple features are sometimes combined into one embodiment, figure or description thereof. However, this disclosure method does not mean that the features required by the object of this application are more than the features mentioned in the claims. In fact, the features of the embodiments are less than all the features of the single embodiment disclosed above.

[0164] In some embodiments, numbers describing the number of components and attributes are used. It should be understood that such numbers used in the description of the embodiments are modified by the modifiers "about", "approximately" or "substantially" in some examples. Unless otherwise specified, "about", "approximately" or "substantially" indicate that the numbers are allowed to vary by ±10%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may change according to the required features of individual embodiments. In some embodiments, the numerical parameters should take into account the specified significant digits and adopt the general method of retaining digits. Although the numerical domains and parameters used to confirm the breadth of their range in some embodiments of the present application are approximate values, in specific embodiments, the setting of such numerical values ​​is as accurate as possible within the feasible range.

[0165] Each patent, patent application, patent application disclosure, and other materials, such as articles, books, instructions, publications, documents, etc., cited in this application are hereby incorporated by reference in their entirety. Except for application history documents that are inconsistent with or conflicting with the content of this application, documents that limit the broadest scope of the claims of this application (currently or later attached to this application) are also excluded. It should be noted that if the descriptions, definitions, and / or use of terms in the attached materials of this application are inconsistent or conflicting with the content described in this application, the descriptions, definitions, and / or use of terms in this application shall prevail.

[0166] Finally, it should be understood that the embodiments described in this application are only used to illustrate the principles of the embodiments of the present application. Other variations may also fall within the scope of the present application. Therefore, as an example and not a limitation, the alternative configurations of the embodiments of the present application may be considered to be consistent with the teachings of the present application. Accordingly, the embodiments of the present application are not limited to the embodiments explicitly introduced and described in the present application.

Claims

1. A method for evaluating proton therapy, characterized in that: The evaluation methods include: The data pairs of the historical proton beam delivery dose distribution image and the historical positron nuclide distribution image are formed into a data set; Constructing a first model, and using the data set to train the first model to obtain a second model; Inputting the positron nuclide distribution image in actual proton therapy into the second model to obtain the proton beam delivery dose distribution image in actual proton therapy; The falling edge at the far end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy is extracted to obtain the range information of the proton beam in actual proton therapy.

2. The evaluation method according to claim 1, characterized in that: The using the data set to train the first model to obtain a second model comprises: Extracting a plurality of one-dimensional curves from the historical positron nuclide distribution image, converting the one-dimensional curves into a first vector and using the first vector as an input of the first model; Converting the first vector into a plurality of sequence feature vectors having a first preset length through a first convolutional layer of the first model; Adding position information to the sequence feature vector by a first encoder of the first model; The first decoder of the first model successively performs: restoring the length of the sequence feature vector to be the same as that of the first vector, reducing the number of channels of the sequence feature vector, concatenating the sequence feature vectors from different layers, and fusing the sequence feature vectors from different layers to obtain an output proton beam delivery dose distribution image; The parameters of the first model are corrected according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

3. The evaluation method according to claim 2, characterized in that: The fusing of the sequence feature vectors of different layers performed by the first decoder includes: fusing the sequence feature vectors of different layers by adopting a top-down fusion mechanism through the first decoder.

4. The evaluation method according to claim 2, characterized in that: The first decoder is a decoder based on CNN and Tansformer.

5. The evaluation method according to claim 1, characterized in that: The using the data set to train the first model to obtain a second model comprises: Inputting the historical positron nuclide distribution image into the first model; extracting three-dimensional image features of the historical positron nuclide distribution image through a second encoder of the first model, gradually reducing the spatial size of the three-dimensional image features, and increasing the abstract representation of the three-dimensional image features; Restoring the spatial resolution of the three-dimensional image features by a second decoder of the first model, and combining the feature map of the second encoder by a splicing operation during decoding to obtain an output proton beam delivery dose distribution image; The parameters of the first model are corrected according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

6. The evaluation method according to claim 1, characterized in that: The historical proton beam delivery dose distribution image is obtained by Monte Carlo simulation; and / or, The historical positron nuclide distribution image is obtained by PET measurement.

7. The evaluation method according to claim 1, characterized in that: The step of inputting the actual positron nuclide distribution image in proton therapy into the second model to obtain the actual proton beam delivery dose distribution image in proton therapy includes: At the beginning of the actual proton therapy, PET is used to collect positron nuclide signals, and positron nuclide signal processing and image reconstruction are performed to obtain a positron nuclide distribution image in the actual proton therapy.

8. The evaluation method according to claim 1, characterized in that: The step of extracting the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy to obtain the range information of the proton beam in actual proton therapy includes: The depth corresponding to any point in the range of 50% to 80% of the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image in the actual proton therapy is used as the range information of the proton beam in the actual proton therapy.

9. A proton therapy evaluation system, characterized in that: The evaluation system comprises: A data set construction unit configured to form a data set from a data pair of a historical proton beam delivery dose distribution image and a historical positron nuclide distribution image; A model building and training unit, configured to build a first model, and use the data set to train the first model to obtain a second model; a proton beam delivery dose estimation unit configured to input the positron nuclide distribution image in actual proton therapy into the second model to obtain the proton beam delivery dose distribution image in actual proton therapy; The proton beam range estimation unit is configured to extract the falling edge of the far end of the Bragg peak of the proton beam delivery dose distribution image in actual proton therapy to obtain the range information of the proton beam in actual proton therapy.

10. The evaluation system according to claim 9, characterized in that The first model comprises a one-dimensional model, wherein the one-dimensional model comprises a first convolutional layer, a first encoder, and a first decoder; The model building and training unit is configured to extract a plurality of one-dimensional curves from the historical positron nuclide distribution image, convert the one-dimensional curves into a first vector and use the first vector as an input of the first model; The first convolutional layer is configured to convert the first vector into a plurality of sequence feature vectors having a first preset length; The first encoder is configured to add position information to the sequence feature vector; The first decoder is configured to successively perform: restoring the length of the sequence feature vector to be the same as that of the first vector, reducing the number of channels of the sequence feature vector, concatenating the sequence feature vectors from different layers, and fusing the sequence feature vectors from different layers to obtain an output proton beam delivery dose distribution image; The model building and training unit is further configured to correct the parameters of the first model according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

11. The evaluation system according to claim 10, characterized in that: The first decoder is configured to fuse the sequence feature vectors of different layers using a top-down fusion mechanism.

12. The evaluation system according to claim 10, characterized in that: The first decoder is a decoder based on CNN and Tansformer.

13. The evaluation system according to claim 9, characterized in that: The first model includes a second encoder and a second decoder; The model building and training unit is configured to extract three-dimensional image features from historical positron nuclide distribution images and input the three-dimensional image features into the first model; The second encoder is configured to gradually reduce the spatial size of the three-dimensional image features and increase the abstract representation of the three-dimensional image features; The second decoder is configured to restore the spatial resolution of the image features and combine the feature map of the second encoder through a splicing operation during decoding to obtain an output proton beam delivery dose distribution image; The model building and training unit is also configured to correct the parameters of the first model according to the historical proton beam delivery dose distribution image and the output proton beam delivery dose distribution image to obtain the second model.

14. The evaluation system according to claim 9, characterized in that: The evaluation system also includes: a data acquisition and generation unit, which is configured to use PET to acquire positron nuclide signals when the actual proton therapy begins, and to process and reconstruct the positron nuclide signals to obtain a positron nuclide distribution image in the actual proton therapy.

15. The evaluation system according to claim 9, characterized in that The proton beam range estimation unit is configured to use the depth corresponding to any point between 50% and 80% of the falling edge of the distal end of the Bragg peak of the proton beam delivery dose distribution image in the actual proton therapy as the range information of the proton beam in the actual proton therapy.

16. A storage medium, characterized in that: The storage medium stores a computer program configured to implement the proton therapy evaluation method according to any one of claims 1 to 8 when the computer program is executed.

17. A computer program product, characterized in that The method comprises a computer program or an instruction, wherein when the computer program or the instruction is executed by a processor, the method for evaluating proton therapy according to any one of claims 1 to 8 is implemented.

Images