Radiotherapy positioning model construction method and prediction method and device based on image analysis

Abstract

The invention discloses a radiotherapy positioning model construction method and prediction method and device based on image analysis, and belongs to the technical field of radiotherapy. The radiotherapy positioning model construction method based on image analysis comprises the steps of collecting basic clinical data, constructing a primary generic neural network prediction model, collecting thoracic information data, constructing an intermediate individualized neural network prediction model, and constructing an individualized dynamic neural network prediction model to obtain the radiotherapy positioning model based on image analysis. According to the model, on one hand, a patient does not need to be repositioned for many times, manpower and material resources are greatly saved, efficiency is improved, and time and energy of a doctor are saved, so that the doctor can concentrate on other important clinical work; and on the other hand, the efficiency, accuracy and consistency of radiotherapy positioning are greatly improved, so that the work does not excessively depend on the technical level, experience, endurance degree and the like of doctors any more.

Description

Technical Field

[0001] The present application relates to a method and a prediction method and device for constructing a radiotherapy positioning model based on image analysis, belonging to the technical field of radiotherapy. Background Art

[0002] Radiotherapy is an indispensable and important treatment method. By irradiating radioactive elements to the target area of a patient, radiotherapy can kill cancer cells, improve the local control rate of cancer, reduce local recurrence, and improve long-term survival. In existing radiotherapy systems, generally, after a patient determines the target area of the tumor through PET or CT, CT registration is performed before radiotherapy to formulate a radiotherapy plan.

[0003] A radiotherapy course usually takes several months or longer. Before the start of radiotherapy, a radiotherapy plan is usually generated based on the initial positioning image of the patient. However, during the entire radiotherapy process, as the radiotherapy plan is implemented, the position and shape of the patient's tumor usually change, and during radiotherapy and chemotherapy, the patient is prone to problems such as loss of appetite and nausea, resulting in a sudden drop in weight, which causes changes in the radiation area and radiation dose. Therefore, the problem of deviation in radiation dose is likely to occur during radiotherapy, which will not only affect the radiotherapy effect, but is more likely to affect the normal organ tissues of the patient, leading to the occurrence of complications.

[0004] To solve this problem, in the prior art, the patient is usually repositioned after a period of time from the start of radiotherapy, and then the latest positioning image of the patient is obtained, and the doctor redraws the image and formulates a radiotherapy plan based on the latest positioning image. However, this process requires a large amount of manpower and material resources, has high requirements for the doctor's patience and technical level, and the doctor's level, lying position, etc. all affect the accuracy, and it is very time-consuming, resulting in a significant reduction in the radiotherapy efficiency. Summary of the Invention

[0005] One or more embodiments of the present specification provide a method and a prediction method and device for constructing a radiotherapy positioning model based on image analysis, which are used to solve the technical problems proposed in the background art.

[0006] One or more embodiments of the present specification adopt the following technical solutions:

[0007] The method for constructing a radiotherapy positioning model based on image analysis provided by one or more embodiments of the present specification is characterized by including:

[0008] Basic clinical data collection: Obtain the human parameter data of several patients in the early and late stages of the treatment course, and obtain the medical image data of several patients in the early and late stages of the treatment course for preprocessing to obtain a human parameter data set and a first medical image data set;

[0009] Construction of primary general population neural network prediction model: Classify the human parameter data to obtain different human parameter data sets, and process, learn, and train the human parameter data sets and the first medical image data sets to obtain a tumor size and location prediction model that establishes a time series based on the first medical image data sets and the human parameter data sets;

[0010] Collection of thoracic cavity information data: Obtain the thoracic cavity information data of the patient during radiotherapy in the above data sets and the corresponding medical image data, that is, the second medical image data, and process the data to obtain a thoracic cavity information data set and a second medical image data set;

[0011] Construction of intermediate individualized neural network prediction model: Learn and train the thoracic cavity information data set and the corresponding second medical image data set to obtain a tumor size and location prediction model that establishes a time series based on the thoracic cavity information data, the first medical image data set, the second medical image data set, and the human parameter data set, so as to correct the primary general population neural network prediction model;

[0012] Radiotherapy positioning model based on image analysis: Divide the radiotherapy cycle of the patient into several stages according to the above prediction model, and use the medical image data of the nth stage to verify the prediction results of the (n - 1)th stage and dynamically correct the model parameters of the intermediate individualized neural network prediction model to obtain a radiotherapy positioning model based on image analysis.

[0013] Specifically, correcting the primary general population neural network prediction model and dynamically correcting the model parameters of the intermediate individualized neural network prediction model include modifying the network layer weights and bias parameters.

[0014] It should be noted that through the above content, the embodiments of this specification have the following beneficial effects:

[0015] Improve efficiency: Through the radiotherapy positioning model based on image analysis, it is possible to predict the changes in the tumor position, shape, and size of the patient as the radiotherapy plan is implemented, without having to reposition the patient multiple times, greatly saving manpower and material resources and improving efficiency; on the other hand, it saves the time and energy of doctors, facilitating their focus on other important clinical work.

[0016] Improve accuracy and consistency: By predicting through the radiotherapy positioning model based on image analysis, the dependence on the doctor's technical level, experience, patience, etc. is reduced, the positioning differences caused by doctor differences are reduced, and the accuracy and consistency of positioning are improved.

[0017] Personalized treatment plan: This radiotherapy positioning model based on image analysis allows for the formulation of personalized radiotherapy plans according to the specific conditions of each patient, such as weight loss, etc., as well as tumor characteristics, which helps to maximize the effectiveness of treatment.

[0018] Furthermore, the human parameter data includes age, gender, BMI, tumor type, tumor location, number of radiotherapy sessions, radiotherapy time, and radiotherapy dose; the first medical image data includes CT images, MRI images, PET images, or ultrasonic images; the thoracic cavity information data includes the gas volume and pressure data of the body film airbag; the second medical image data includes CT images, MRI images, PET images, or ultrasonic images.

[0019] Furthermore, the preprocessing includes gray-scale inhomogeneity correction, noise removal, contrast enhancement, unifying the format and scale of the images, and segmentation to obtain tumor feature information.

[0020] Furthermore, in the construction of the primary general population neural network prediction model, the processing, learning, and training of the human parameter data set and the first medical image data set, and in the construction of the intermediate individual neural network prediction model, the learning and training of the thoracic cavity information data set and the corresponding second medical image data set include:

[0021] Dividing the preprocessed human parameter data set, first medical image data set, or thoracic cavity information data set, second medical image data set into data sets, specifically into training set, validation set, and test set;

[0022] Loading the training set into the neural network prediction model for training;

[0023] Saving the trained and optimal model weights;

[0024] Loading the model weights and performing actual tests on the test set to obtain the final result.

[0025] Furthermore, loading the training set into the neural network prediction model for training specifically includes:

[0026] Loading the training set into the neural network prediction model and initializing the parameters of the neural network prediction model;

[0027] Using the stochastic gradient descent algorithm to perform forward and backward propagation on the feature vectors learned by the neural network prediction model,

[0028] Continuously updating the learning parameters of the neural network prediction model;

[0029] Calculating the cross-entropy loss function between the prediction results after each iterative training and the real data until the loss value reaches the global optimal solution;

[0030] Record the training and validation losses for each iteration of training;

[0031] When both the training and validation losses are decreasing synchronously, the neural network prediction model continues to be trained;

[0032] When the training loss is decreasing while the validation loss is increasing, the neural network prediction model stops training;

[0033] Train the neural network prediction model for a specified number of iterations and then stop training.

[0034] A prediction method for a radiotherapy positioning model based on image analysis provided by one or more embodiments of this specification; input the human parameter data, first medical image data, thoracic cavity information data, and second medical image data of the current case into the trained radiotherapy positioning model based on image analysis to obtain the prediction result of the patient of this case.

[0035] Specifically, first classify it into a suitable data set according to its human parameter data, and based on the human parameter data, first medical image data, thoracic cavity information data, and second medical image data of the current case, the radiotherapy positioning model based on image analysis gives the prediction result.

[0036] A radiotherapy positioning device based on image analysis provided by one or more embodiments of this specification, including:

[0037] A first data acquisition module for acquiring human parameter data;

[0038] A second data acquisition module for acquiring first medical image data and second medical image data;

[0039] A third data acquisition module for acquiring thoracic cavity information data;

[0040] An image processing module for preprocessing the first medical image data and the second medical image data to obtain a first medical image data set and a second medical image data set;

[0041] A calculation module, in which a trained radiotherapy positioning model based on image analysis is configured, and input the human parameter data, first medical image data, and thoracic cavity information data during radiotherapy of the current case into the trained neural network prediction model for calculation to obtain the predicted radiotherapy area positioning.

[0042] Specifically, a model training unit is provided in the calculation module for training the neural network prediction model according to the human parameter information, medical image data, and thoracic cavity information data of several patients.

[0043] Furthermore, the third data acquisition module includes:

[0044] A body film, which is covered outside the chest and abdomen of a patient;

[0045] A body film airbag, which is arranged between the chest and abdomen of the patient and the body film, and both sides of the body film airbag are in contact with the body film and the chest and abdomen of the patient respectively;

[0046] An air outlet device, which is arranged on the body film airbag to discharge the gas in the body film airbag and measure the exhaust gas volume;

[0047] An air inlet device, which is arranged on the body film airbag to input gas into the body film airbag and measure the intake gas volume;

[0048] A gas pressure sensor, which is used to detect the pressure in the body film airbag;

[0049] A controller, which is signal-connected to the air inlet device, the air outlet device and the gas pressure sensor respectively.

[0050] Specifically, with the implementation of the radiotherapy plan, due to reasons such as a sudden drop in the patient's weight, a space is generated between the patient's chest cavity and the body film, and the patient's chest cavity and the body film cannot be in complete contact, which affects the accuracy of positioning. In this application, a body film airbag is arranged in the space between the patient's chest cavity and the body film. The air inlet device inputs gas into the body film airbag, so that one side of the body film airbag is in contact with the patient's chest cavity and the other side is in contact with the body film; the gas pressure sensor detects the pressure in the body film airbag, and the pressure in the body film airbag is balanced and stable.

[0051] Furthermore, the air inlet device includes: a gas supply component, an air inlet flowmeter and a pressure reducing valve; the air outlet device includes an air outlet flowmeter and an exhaust valve; the body film airbag and the body film are detachably connected.

[0052] The beneficial effects of this application include but are not limited to:

[0053] Improve efficiency: Through the trained neural network prediction model, it is possible to predict the changes in the tumor position, shape and size of the patient with the implementation of the radiotherapy plan, thus avoiding the problems of needing to re-obtain the patient's positioning images, re-perform image delineation and re-formulate the radiotherapy plan, greatly saving manpower and material resources and improving efficiency; on the other hand, it saves the doctor's time and energy, so as to focus on other important clinical work.

[0054] Improve accuracy and consistency: Through prediction by the neural network prediction model, the dependence on the doctor's technical level, experience, patience, etc. is reduced, the positioning differences caused by doctor differences are reduced, and the accuracy and consistency of positioning are improved.

[0055] Personalized treatment plan: This method allows for the formulation of personalized radiotherapy plans according to the specific conditions of each patient, such as the weight loss situation, etc., and tumor characteristics, which helps to maximize the effectiveness of treatment. Brief Description of the Drawings

[0056] The drawings described herein are provided to further understand the present application and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0057] Figure 1 It is a schematic flow chart of a method for constructing a radiotherapy positioning model based on image analysis provided for one or more embodiments of this specification;

[0058] Figure 2 It is a schematic structural diagram of a radiotherapy positioning model device based on image analysis provided for one or more embodiments of this specification. Detailed Embodiments

[0059] The embodiments of this specification provide a method for constructing a radiotherapy positioning model based on image analysis, a prediction method, and a device.

[0060] In order to enable those skilled in the art to better understand the technical solutions in this specification, the following will clearly and completely describe the technical solutions in the embodiments of this specification with reference to the drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this specification.

[0061] Figure 1 It is a schematic flow chart of a method for constructing a radiotherapy positioning model based on image analysis provided for one or more embodiments of this specification, and this flow can be executed by a radiotherapy positioning model device based on image analysis. Some input parameters or intermediate results in the flow allow manual intervention and adjustment to help improve accuracy.

[0062] The method flow steps of the embodiments of this specification are as follows:

[0063] S101, Basic clinical data collection: Obtain the human parameter data of several patients in the early and late stages of the treatment course, and obtain the medical image data of several patients in the early and late stages of the treatment course for preprocessing to obtain a human parameter data set and a first medical image data set;

[0064] In the embodiments of this specification, data classification can be performed according to the human parameter data; different training sets are formed to improve the accuracy of the prediction model.

[0065] Specifically, in the embodiments of this specification, the human parameter data includes age, gender, BMI, tumor type, tumor location, number of radiotherapy sessions, radiotherapy time, and radiotherapy dose; the medical image data includes CT images, MRI images, PET images, or ultrasonic images.

[0066] Specifically, in the embodiments of this specification, it also includes preprocessing the human parameter data, including but not limited to:

[0067] Dividing the age from 1 to 120 years old into different age groups, with a 5-year interval for each group, for example: 1 - 5, 6 - 10, 11 - 15, 16 - 20, 21 - 25, 26 - 30, 31 - 35, 36 - 40, 41 - 45, 46 - 50, 51 - 55, 56 - 60, 61 - 65, 66 - 70, 71 - 75, 76 - 80, 81 - 85, 86 - 90, 91 - 95, 96 - 100, 101 - 105, 106 - 110, 111 - 115, 115 - 120.

[0068] Setting male gender as A and female gender as B.

[0069] In the embodiments of this specification, the preprocessing includes gray - level inhomogeneity correction, noise removal, contrast enhancement, unifying the format and scale of images, and segmentation to obtain tumor feature information.

[0070] In the embodiments of this application specification, the gray - level inhomogeneity correction includes: First, using the histogram equalization method to stretch the gray - level distribution of the image to the entire gray - level range, thereby increasing the contrast of the image. Second, detecting the background region and foreground region in the image, calculating the gray - level means of the background and foreground respectively. Then, calculating the difference between the two gray - level means and using it as a correction factor to linearly adjust the gray - level values of the background and foreground regions of the image to achieve gray - level inhomogeneity correction.

[0071] The noise removal includes: Using the local information of the medical image for filtering, including but not limited to mean filtering and median filtering, and eliminating the high - frequency noise in the medical image through the filter to make the image clearer. Additionally, wavelet transform can be used for denoising, reducing the impact of noise by removing the high - frequency components in the medical image.

[0072] In the embodiments of this application specification, for the selection of the segmentation algorithm: According to the tumor characteristics, select a suitable segmentation algorithm. Common segmentation algorithms include threshold segmentation, region growing, edge detection, model - based segmentation, etc. Multiple segmentation algorithms can be considered for comparison and verification, and the algorithm with the best effect is selected.

[0073] In the embodiments of the specification of this application, after segmentation, it further includes evaluation of the segmentation result: evaluating the segmentation result, and using quantitative indicators such as accuracy, recall rate, Dice coefficient, etc. to evaluate the accuracy of the segmentation. The performance of the segmentation algorithm can be verified by comparing it with the manually segmented result.

[0074] Specifically, through the above preprocessing operations, the medical image has better image quality and more accurate information, providing more precise data support for subsequent prediction.

[0075] The embodiments of the present invention combine medical image data, human parameter data, and thoracic information data to train the model, obtaining a localization prediction model. The obtained localization prediction model is trained, which can make full use of the complementarity between different data modalities. And based on the prediction result of the thoracic information data, the individualized association prediction model has stronger sample adaptability, can better extract complex non-linear features and make more accurate predictions, and can provide a relatively accurate guiding direction for clinical radiotherapy, which is beneficial to the further application of clinical radiotherapy.

[0076] S102, constructing a primary generalization neural network prediction model: classifying the human parameter data to obtain different human parameter data sets, and processing, learning, and training the human parameter data sets and the first medical image data sets to obtain a tumor size and position prediction model that establishes a time series based on the first medical image data sets and the human parameter data sets;

[0077] In the embodiments of the present invention, processing, learning, and training the human parameter data sets and the first medical image data sets include:

[0078] Dividing the preprocessed first medical image data and human parameter data into data sets, specifically divided into a training set, a validation set, and a test set;

[0079] Loading the training set into the neural network prediction model for training;

[0080] Saving the trained and optimal model weights;

[0081] Loading the model weights and performing actual tests on the test set to obtain the final result.

[0082] In this embodiment, the first medical image data sets and human parameter data sets obtained and processed are divided into data sets. Among them, the training set is used for model training, the validation set is used for model tuning and validation, and the test set is used for final evaluation of the model performance. By reasonably splitting the data sets, the data can be used more reasonably, effectively avoiding the overfitting phenomenon of the model for specific data, improving the performance of the model on unknown data, and further ensuring the generalization ability and robustness of the model.

[0083] In this embodiment, the data can be processed by data processing methods such as data cleaning, data selection, and feature dimensionality reduction, but are not limited thereto.

[0084] In the embodiment of the present invention, the training set is loaded into the neural network prediction model for training, which specifically includes:

[0085] Load the training set into the neural network prediction model and initialize the parameters of the neural network prediction model;

[0086] Use the stochastic gradient descent algorithm to perform forward and backward propagation on the feature vectors learned by the neural network prediction model,

[0087] Continuously update the learning parameters of the neural network prediction model;

[0088] Calculate the cross-entropy loss function between the prediction result after each iterative training and the real data until the loss value reaches the global optimal solution;

[0089] Record the training and validation losses of each iterative training;

[0090] When the training and validation losses are both decreasing, the neural network prediction model continues to be trained;

[0091] When the training loss decreases while the validation loss increases, the neural network prediction model stops training;

[0092] Train the neural network prediction model to the specified number of iterations and then stop training.

[0093] Specifically, the neural network model is a convolutional neural network and a recurrent neural network.

[0094] S103, Thoracic cavity information data collection: Obtain the thoracic cavity information data of the patient during radiotherapy in the above dataset and the corresponding medical image data, that is, the second medical image data, and process the data to obtain the thoracic cavity information dataset and the second medical image dataset;

[0095] Specifically, the thoracic cavity information data includes the gas volume and pressure data of the body membrane airbag; the second medical image data includes CT images, nuclear magnetic resonance images, PET images, or ultrasonic images.

[0096] S104, Intermediate individualized neural network prediction model construction: Learn and train the thoracic cavity information dataset and the corresponding second medical image dataset to obtain a tumor size and position prediction model that establishes a time series based on the thoracic cavity information data, the first medical image dataset, the second medical image dataset, and the human parameter dataset, so as to correct the primary general population neural network prediction model;

[0097] In the embodiments of the present invention, learning and training the thoracic information data set and the corresponding second medical image data set includes:

[0098] Dividing the preprocessed second medical image data and thoracic information data into data sets, specifically into a training set, a validation set, and a test set;

[0099] Loading the training set into the neural network prediction model for training;

[0100] Saving the trained and optimal model weights;

[0101] Loading the model weights and performing actual tests on the test set to obtain the final result.

[0102] In this embodiment, the second medical image data set and the thoracic information data set obtained and processed are divided into data sets. The training set is used for model training, the validation set is used for model parameter tuning and validation, and the test set is used for final evaluation of the model performance. By reasonably splitting the data set, the data can be used more reasonably, effectively avoiding the overfitting phenomenon of the model for specific data, improving the performance of the model on unknown data, and further ensuring the generalization ability and robustness of the model.

[0103] In this embodiment, the data can be processed by data processing methods such as but not limited to data cleaning, data selection, and feature dimensionality reduction.

[0104] In the embodiments of the present invention, loading the training set into the neural network prediction model for training specifically includes:

[0105] Loading the training set into the neural network prediction model and initializing the parameters of the neural network prediction model;

[0106] Using the stochastic gradient descent algorithm to perform forward and backward propagation on the feature vectors learned by the neural network prediction model,

[0107] Continuously updating the learning parameters of the neural network prediction model;

[0108] Calculating the cross-entropy loss function between the prediction results after each iterative training and the real data until the loss value reaches the global optimal solution;

[0109] Recording the training and validation losses of each iterative training;

[0110] When the training and validation losses are both decreasing, the neural network prediction model continues to train;

[0111] When the training loss decreases while the validation loss increases, the neural network prediction model stops training;

[0112] Train the neural network prediction model to a specified number of iterations and then stop the training.

[0113] Specifically, the neural network model is a convolutional neural network and a recurrent neural network.

[0114] In the embodiments of the present application specification, data classification can be performed according to human parameter data; different training sets are formed to improve the accuracy of the prediction model.

[0115] Specifically, the preprocessed human parameter data set and the first medical image data set are loaded into the neural network model, and the neural network model is trained to obtain a tumor size and position prediction model that establishes a time series based on the obtained tumor size and position; that is, the construction of a primary general population neural network prediction model; combined with the thoracic cavity information data set and the second medical image data set, the neural network model is trained to obtain a tumor size and position prediction model that establishes a time series based on the thoracic cavity information data, the first medical image data set, the second medical image data set, and the human parameter data set, so as to correct the primary general population neural network prediction model.

[0116] S105, radiotherapy positioning model based on image analysis: Divide the radiotherapy cycle of the patient into several stages according to the above prediction model, and use the medical image data of the nth stage to verify the prediction result of the (n - 1)th stage and dynamically correct the model parameters of the intermediate individualized neural network prediction model to obtain a radiotherapy positioning model based on image analysis.

[0117] Specifically, in different treatment stages, dynamic correction and adjustment are performed according to the progress of the patient's personal condition, so that the positioning model can better adapt to individuality and has higher accuracy and adaptability.

[0118] One or more embodiments of the present application also provide a prediction method for a radiotherapy positioning model based on image analysis. Input the human parameter data, the first medical image data, the thoracic cavity information data, and the second medical image data of the current case into the trained radiotherapy positioning model based on image analysis to obtain the prediction result of the patient.

[0119] Specifically, for the current case, the model can be verified and corrected according to the latest medical image data of the current case, so that the model can adapt to the characteristics of individualized cases and make the result more accurate and stable.

[0120] Figure 2 A radiotherapy positioning device based on image analysis provided by one or more embodiments of this specification, characterized in that it includes:

[0121] The first data acquisition module 201 is used to acquire human parameter data;

[0122] The second data acquisition module 202 is configured to acquire first medical image data and second medical image data;

[0123] The third data acquisition module 203 is configured to acquire thoracic cavity information data;

[0124] The image processing module 204 is configured to preprocess the first medical image data and the second medical image data to obtain a first medical image data set and a second medical image data set;

[0125] The calculation module 205 is configured with a trained radiotherapy positioning model based on image analysis. The current case human parameter data, the first medical image data, and the thoracic cavity information data during the radiotherapy process are input into the trained neural network prediction model for calculation to obtain the predicted radiotherapy area positioning.

[0126] As a specific implementation manner, the third data acquisition module includes:

[0127] A body film, covering the outer side of the patient's chest and abdomen;

[0128] A body film airbag, disposed between the patient's chest and abdomen and the body film, and both sides of the body film airbag are in contact with the body film and the patient's chest and abdomen respectively;

[0129] An air inlet device, disposed on the body film airbag to discharge the gas in the body film airbag and measure the exhaust gas volume;

[0130] An air outlet device, disposed on the body film airbag to input gas into the body film airbag and measure the intake gas volume;

[0131] A gas pressure sensor to detect the pressure in the body film airbag;

[0132] A controller, which is respectively connected to the air inlet device, the air outlet device, and the gas pressure sensor in terms of signals.

[0133] As a specific implementation manner, the air inlet device includes: a gas supply component, an intake gas flowmeter, and a pressure reducing valve; the air outlet device includes an outlet gas flowmeter and an exhaust valve; the body film airbag and the body film are detachably connected.

[0134] Specifically, the present application does not make specific limitations on the materials of the body film and the body film airbag, and those skilled in the art can make appropriate selections according to the actual situation.

[0135] Specifically, the material of the body film airbag is a polymer synthetic material.

[0136] Specifically, with the implementation of the radiotherapy plan, due to reasons such as a sudden drop in the patient's weight, a space is generated between the patient's chest cavity and the body film. The patient's chest cavity and the body film do not fully contact, which affects the accuracy of positioning. In this application, a body film airbag is provided and arranged in the space between the patient's chest cavity and the body film. The air intake device inputs gas into the body film airbag, so that one side of the body film airbag contacts the patient's chest cavity and the other side contacts the body film; the gas pressure sensor detects the pressure inside the body film airbag, and the pressure inside the body film airbag is balanced and stable.

[0137] Specifically, through the feedback of pressure and gas volume, the position of the patient can also be corrected to reduce the deviation; when the deviation is greater than the set threshold, corresponding warning operations are performed.

[0138] The radiotherapy positioning model based on image analysis obtained by the radiotherapy positioning model construction method provided in this application can predict the changes in the position, shape, and size of the patient's tumor with the implementation of the radiotherapy plan, thus avoiding the problems of needing to re-obtain the patient's positioning image, re-delineate the image, and re-formulate the radiotherapy plan. It greatly saves manpower and material resources, improves efficiency, saves the doctor's time and energy, and facilitates focusing on other important clinical work. On the other hand, through the prediction by the radiotherapy positioning model based on image analysis, the dependence on the doctor's technical level, experience, patience, etc. is reduced, the positioning difference caused by doctor differences is reduced, and the accuracy and consistency of positioning are improved. In addition, this method allows for the formulation of personalized radiotherapy plans according to the specific conditions of each patient, such as the weight loss situation, etc., and tumor characteristics, which helps to maximize the effectiveness of treatment.

[0139] As described above, only the embodiments of this application are given. The protection scope of this application is not limited by these specific embodiments, but is determined by the claims of this application. For those skilled in the art, various changes and modifications can be made to this application. Any modifications, equivalent replacements, improvements, etc. made within the technical idea and principle of this application shall be included in the protection scope of this application.

Claims

1. A method for constructing a radiotherapy positioning model based on image analysis, characterized in that: include: Basic clinical data collection: obtaining human body parameter data of several patients before and after the treatment course, obtaining medical imaging data of several patients before and after the treatment course for preprocessing, and obtaining a human body parameter data set and a first medical imaging data set; Construction of a primary pan-population neural network prediction model: classifying the human body parameter data to obtain different human body parameter data sets, processing, learning and training the human body parameter data sets and the first medical image data set to obtain a tumor size and location prediction model that establishes a time series based on the first medical image data set and the human body parameter data set; Thoracic information data collection: obtaining thoracic information data of the patient during radiotherapy in the above data set and medical image data corresponding to the thoracic information data, i.e., second medical image data, and processing the data to obtain a thoracic information data set and a second medical image data set; Construction of intermediate individualized neural network prediction model: learning and training the chest information data set and the corresponding second medical image data set, and obtaining a tumor size and location prediction model based on the chest information data, the first medical image data set, the second medical image data set, and the human body parameter data set to establish a time series, so as to correct the primary pan-population neural network prediction model; Radiotherapy positioning model based on image analysis: The patient's radiotherapy cycle is divided into several stages according to the above prediction model, and the medical imaging data of the nth stage is used to verify the prediction results of the n-1 stage and dynamically correct the model parameters of the intermediate individualized neural network prediction model to obtain the radiotherapy positioning model based on image analysis.

2. The method for constructing a radiotherapy positioning model based on image analysis according to claim 1, characterized in that: The human body parameter data includes age, gender, BMI, tumor type, tumor location, number of radiotherapy times, radiotherapy time and radiotherapy dose; the first medical image data includes CT images, MRI images, PET images or ultrasound images; the thoracic information data includes body membrane airbag gas volume and pressure data; the second medical image data includes CT images, MRI images, PET images or ultrasound images.

3. The method for constructing a radiotherapy positioning model based on image analysis according to claim 1, characterized in that: The preprocessing includes grayscale unevenness and correction, noise removal, contrast enhancement, unification of image format and scale, and segmentation to obtain tumor feature information.

4. The method for constructing a radiotherapy positioning model based on image analysis according to claim 1, characterized in that: The processing, learning and training of the human body parameter data set and the first medical image data set in the construction of the primary pan-population neural network prediction model, and the learning and training of the thorax information data set and the corresponding second medical image data set in the construction of the intermediate individualized neural network prediction model include: The preprocessed human body parameter data set, the first medical image data set or the thorax information data set, and the second medical image data set are divided into a training set, a validation set, and a test set; Load the training set into the neural network prediction model for training; Save the trained and optimal model weights; Load the model weights and perform actual testing on the test set to get the final results.

5. The method for constructing a radiotherapy positioning model based on image analysis according to claim 4, characterized in that: Load the training set into the neural network prediction model for training, including: Load the training set into the neural network prediction model and initialize the neural network prediction model parameters; The stochastic gradient descent algorithm is used to forward and backward propagate the feature vectors learned by the neural network prediction model. Continuously update the learning parameters of the neural network prediction model; The predicted results after each iterative training are compared with the real data to calculate the cross entropy loss function until the loss value reaches the global optimal solution; Record the training and validation losses for each training iteration; When the training and validation losses are decreasing simultaneously, the neural network prediction model continues to train; When the training loss decreases and the validation loss increases, the neural network prediction model stops training; Train the neural network prediction model to the specified number of iterations and then stop training.

6. A prediction method for a radiotherapy positioning model based on image analysis, wherein the human body parameter data, the first medical image data, the chest information data, and the second medical image data of the current case are input into a trained radiotherapy positioning model based on image analysis to obtain a prediction result for the case.

7. A radiotherapy positioning device based on image analysis, characterized in that: include: A first data acquisition module, used to acquire human body parameter data; A second data acquisition module, used to acquire the first medical image data and the second medical image data; A third data acquisition module is used to acquire thorax information data; An image processing module, used for preprocessing the first medical image data and the second medical image data to obtain a first medical image data set and a second medical image data set; A calculation module is configured with a trained radiotherapy positioning model based on image analysis, and the human body parameter data of the current case, the first medical image data, and the chest information data during the radiotherapy process are input into the trained neural network prediction model for calculation to obtain the predicted radiotherapy area positioning.

8. The radiotherapy positioning device based on image analysis according to claim 7, characterized in that: The third data acquisition module includes: Body membrane, which is placed on the outside of the patient's chest and abdomen; The body membrane airbag is arranged between the patient's chest and abdomen and the body membrane, and the two sides of the body membrane airbag are in contact with the body membrane and the patient's chest and abdomen respectively; An air intake device is provided on the body membrane airbag to discharge the gas in the body membrane airbag and measure the exhaust volume; An air outlet device is provided on the body membrane airbag to input air into the body membrane airbag and measure the amount of air intake; Gas pressure sensor to detect the pressure inside the body membrane airbag; The controller is respectively connected with the air inlet device, the air outlet device and the gas pressure sensor signal.

9. The radiotherapy positioning device based on image analysis according to claim 8, characterized in that: The air intake device comprises: an air supply part, an air intake flow meter and a pressure reducing valve; the air outlet device comprises an air outlet flow meter and an exhaust valve; the body membrane airbag is detachably connected to the body membrane.

Images