Nasopharynx cancer radiotherapy plan generation method, device and equipment

Abstract

The invention discloses a nasopharyngeal carcinoma radiotherapy plan generation method, device and equipment, and relates to the field of intelligent medical treatment, and the method comprises the steps: obtaining multi-modal and multi-omics data of a nasopharyngeal carcinoma patient; the multi-modal multi-omics data comprises radiotherapy plan data, radiomics data, similarity data and Hausdorff distance data; performing data cleaning and feature extraction processing on the multi-modal multi-omics data in sequence to obtain target feature data; according to the target feature data, an artificial intelligence generation method is adopted to generate a nasopharyngeal carcinoma radiotherapy plan; the nasopharyngeal carcinoma radiotherapy plan comprises an equal dose curve graph of dose distribution, a dose volume histogram of a target region and a normal organ, control point data of a linear accelerator and motion data of a multi-leaf collimator. According to the invention, the precision and generation efficiency of the nasopharyngeal carcinoma radiotherapy plan are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent medicine, and particularly to a method, device and equipment for generating a radiotherapy plan for nasopharyngeal carcinoma based on multi-modal and multi-omics. Background Art

[0002] In the context of the big data era, generative artificial intelligence has gradually come to the forefront, making intelligent medicine possible, and significant changes have also taken place in the radiotherapy strategies for tumors. Radiotherapy assisted by artificial intelligence has been widely emphasized. Radiotherapy is the primary and most important treatment method for nasopharyngeal carcinoma. In the precise treatment of nasopharyngeal carcinoma, the mainstream technology is intensity-modulated radiotherapy (IMRT). While maximizing the irradiation dose of the tumor, it also minimizes the irradiated dose and volume of the surrounding normal tissues to the greatest extent, achieving adjustable intensity, dose, and range in the target area. The wide application of the IMRT technology in the radiotherapy of nasopharyngeal carcinoma has greatly improved the local control rate and survival rate of patients.

[0003] Currently, the application of artificial intelligence technology in the medical field is in full swing. Generative artificial intelligence (GAI) is a type of artificial intelligence that can create new content and ideas, including conversations, stories, images, videos, and music. It is a development direction of artificial intelligence, attempting to imitate human intelligence in non-traditional computing tasks such as image recognition, natural language processing (NLP), and translation. GAI uses advanced algorithms such as deep learning and neural networks to simulate the creative thinking process of humans and generate various forms of information, such as text, images, music, and code.

[0004] Although the dose prediction based on artificial intelligence is still in the rapid development stage, many research technologies are still immature. For example, the predicted dose-volume histogram (DVH) cannot reflect the dose distribution characteristics, and the data features of dose omics analysis are too single, etc. The general problem is that using traditional artificial intelligence methods can only predict many parameters in radiotherapy step by step, and there is still a distance from actually implementing the radiotherapy plan on a medical linear accelerator. Summary of the Invention

[0005] The purpose of the present application is to provide a method, device and equipment for generating a radiotherapy plan for nasopharyngeal carcinoma, which can quickly and accurately generate a complex radiotherapy plan for nasopharyngeal carcinoma.

[0006] To achieve the above purpose, the present application provides the following solutions:

[0007] In a first aspect, the present application provides a method for generating a radiotherapy plan for nasopharyngeal carcinoma, including:

[0008] Obtaining multi-modal multi-omics data of a nasopharyngeal carcinoma patient; the multi-modal multi-omics data includes radiotherapy plan data, radiomics data, similarity data, and Hausdorff distance data;

[0009] Successively performing data cleaning and feature extraction processing on the multi-modal multi-omics data to obtain target feature data;

[0010] According to the target feature data, using an artificial intelligence generation method to generate a radiotherapy plan for nasopharyngeal carcinoma; the radiotherapy plan for nasopharyngeal carcinoma includes an isodose curve graph of dose distribution, a dose volume histogram of the target area and normal organs, linear accelerator control point data, and multi-leaf collimator movement data.

[0011] In a second aspect, the present application provides a device for generating a radiotherapy plan for nasopharyngeal carcinoma, including:

[0012] A data acquisition module for obtaining multi-modal multi-omics data of a nasopharyngeal carcinoma patient; the multi-modal multi-omics data includes radiotherapy plan data, radiomics data, similarity data, and Hausdorff distance data;

[0013] A data processing module for successively performing data cleaning and feature extraction processing on the multi-modal multi-omics data to obtain target feature data;

[0014] A plan generation module for generating a radiotherapy plan for nasopharyngeal carcinoma according to the target feature data using an artificial intelligence generation method; the radiotherapy plan for nasopharyngeal carcinoma includes an isodose curve graph of dose distribution, a dose volume histogram of the target area and normal organs, linear accelerator control point data, and multi-leaf collimator movement data.

[0015] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the above-mentioned method for generating a radiotherapy plan for nasopharyngeal carcinoma.

[0016] According to the specific embodiments provided by the present application, the present application has the following technical effects:

[0017] The present application provides a method, device, and equipment for generating a radiotherapy plan for nasopharyngeal carcinoma. Based on radiotherapy plan data, radiomics data, similarity data, and Hausdorff distance data, the generation of a radiotherapy plan for nasopharyngeal carcinoma is carried out. The data types are rich and the data volume is large, which improves the accuracy of the radiotherapy plan for nasopharyngeal carcinoma. And using an artificial intelligence generation method to generate a radiotherapy plan for nasopharyngeal carcinoma can skip intermediate steps such as predicting radiotherapy parameters, improving the generation efficiency of the radiotherapy plan for nasopharyngeal carcinoma. Description of the Drawings

[0018] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0019] Figure 1 It is an application environment diagram of a method for generating a radiotherapy plan for nasopharyngeal carcinoma in an embodiment of the present application;

[0020] Figure 2 It is a schematic flowchart of a method for generating a radiotherapy plan for nasopharyngeal carcinoma provided in an embodiment of the present application;

[0021] Figure 3 It is a schematic diagram of the overall process of generating a radiotherapy plan for nasopharyngeal carcinoma in an embodiment of the present application;

[0022] Figure 4 It is a schematic diagram of the acquisition process of multi-modal multi-omics data in an embodiment of the present application;

[0023] Figure 5 It is a schematic diagram of the processing process of multi-modal multi-omics data in an embodiment of the present application;

[0024] Figure 6 It is a schematic diagram of a language-image pre-training model based on contrastive learning in an embodiment of the present application;

[0025] Figure 7 It is a schematic diagram of the detailed process of generating a radiotherapy plan for nasopharyngeal carcinoma in an embodiment of the present application;

[0026] Figure 8 It is a schematic diagram of the functional modules of a device for generating a radiotherapy plan for nasopharyngeal carcinoma provided in an embodiment of the present application;

[0027] Figure 9 It is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Detailed implementation manners

[0028] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0029] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] The nasopharyngeal carcinoma radiotherapy plan generation method provided by the embodiments of the present application can be applied to an application environment as shown in Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set separately, integrated on the server 104, placed in the cloud or on other servers. The terminal 102 can send the multi-modal multi-omics data of nasopharyngeal carcinoma patients to the server 104. After receiving the multi-modal multi-omics data of nasopharyngeal carcinoma patients, the server 104 sequentially performs data cleaning and feature extraction processing on the multi-modal multi-omics data to obtain target feature data; according to the target feature data, an artificial intelligence generation method is used to generate a nasopharyngeal carcinoma radiotherapy plan. The server 104 can feedback the nasopharyngeal carcinoma radiotherapy plan to the terminal 102. In addition, in some embodiments, the nasopharyngeal carcinoma radiotherapy plan generation method can also be implemented separately by the server 104 or the terminal 102.

[0031] Among them, the terminal 102 can be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0032] In an exemplary embodiment, as shown in Figure 2 and Figure 3 shown, a nasopharyngeal carcinoma radiotherapy plan generation method is provided. This method is executed by a computer device, and can be specifically executed alone by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. The program of the nasopharyngeal carcinoma radiotherapy plan generation method is written in C# and Python languages. The development environment is Visual Studio, Anaconda, and Pycharm. The collaborative software is Pyradiomics and Plastimatch. It includes three parts: multi-modal multi-omics data collection, data analysis and processing, and generative method establishment. In the embodiments of the present application, taking this method applied to the server 104 in Figure 1 as an example for description, it includes the following steps 201 to step 203.

[0033] Step 201, obtain the multi-modal multi-omics data of nasopharyngeal carcinoma patients.

[0034] The multi-modal multi-omics data includes radiotherapy plan data, radiomics data, similarity data, and Hausdorff distance data.

[0035] In a specific application example, step 201 includes the following steps 11 to 14.

[0036] Step 11: Obtain the radiotherapy plan data and CT image sequences of nasopharyngeal carcinoma patients.

[0037] In this application, the data acquisition program is written in C# language, edited, compiled, and generated into an exe file using Microsoft Visual Studio 2017. It interacts with the Aria server based on the Varian Eclipse Scripting API package. As Figure 4 shown, the data acquisition program first reads the image numbers of nasopharyngeal carcinoma patients input by the user, collects the nasopharyngeal carcinoma data set in the database, eliminates unqualified data (including data with too large, too small, or one-sided target areas), and then automatically extracts the radiotherapy plan data to construct a basic case information database.

[0038] The radiotherapy plan data includes: patient name, medical record number, date of birth, date of plan design, target volume, total prescription dose, fractionated dose, treatment equipment, plan name, cancer stage, surgical type, total volume of CT sequence, volume of normal organs (including left lung volume, right lung volume, heart volume, spinal cord volume), volume of primary lesion, volume of clinical lesion, number of radiation fields, radiation field angles, collimator angles, monitor units of each radiation field, radiation field mode, normalization method, normalization factor, optimization algorithm, target area dose (including 95% dose of primary focus, 95% dose of clinical lesion), dose of normal organs (including maximum dose of brainstem, maximum dose of spinal cord, maximum dose of left lens, maximum dose of right lens, maximum dose of left optic nerve, maximum dose of right optic nerve, average dose of anterior lymphatic drainage area, average dose of posterior lymphatic drainage area, average dose of sublingual gland, average dose of submandibular gland, average dose of parotid gland, average dose of thyroid gland), dose conformity, dose homogeneity, global maximum dose percentage.

[0039] Only radiotherapy plan data is far from enough for establishing an artificial intelligence tool. The main problem lies in the low prediction accuracy. Therefore, more feature data needs to be extracted to expand the database in order to improve the prediction accuracy. The expanded data includes radiomics data, similarity data, and Hausdorff distance data.

[0040] Step 12: Extract features from the CT image sequences to obtain radiomics data. Specifically, the radiomics data includes first-order statistical features and 26 shape-based features (16 in three dimensions and 10 in two dimensions).

[0041] Step 13, calculating the similarity between the target area and the normal organ in the CT image sequence to obtain similarity data. Specifically, the DICE similarity coefficient between the target area and the normal organ structure is calculated using the formula: Among them, s is the DICE similarity coefficient, X is the target area, and Y is the normal organ.

[0042] Step 14, calculating the Hausdorff distance between the target area and the normal organ in the CT image sequence to obtain Hausdorff distance data.

[0043] The Hausdorff distance is defined as "the maximum distance from one set to the nearest point in another set". More formally, the Hausdorff distance from set A to set B is a maximum-minimum function, with the formula: Among them, h(A,B) is the Hausdorff distance from set A to set B, and d(a,b) is the distance between a and b.

[0044] The radiomics data, DICE correlation coefficient, and Hausdorff distance are processed by a batch processing subroutine. The subroutine first converts and disassembles the DCM format CT file and structure set file to form an NRRD format image file and a mask file named by each normal organ. Then the Pyradiomics program is called to extract radiomics features, and the Plastimatch program is called to calculate the DICE correlation coefficient and Hausdorff distance. Finally, the above data are integrated into one file for the next stage of processing and analysis.

[0045] In another exemplary embodiment, before step 202, the method for generating a nasopharyngeal carcinoma radiotherapy plan further includes:

[0046] According to the radiotherapy plan data, the distribution of plan design dates, the distribution of treatment equipment, the distribution of cancer stage, the distribution of field angles, the quantile-quantile plot (QQ plot) of plan design dates, the quantile-quantile plot of treatment equipment, the quantile-quantile plot of cancer stage, and the quantile-quantile plot of field angles are determined.

[0047] Based on the distribution of plan design dates, the distribution of treatment equipment, the distribution of cancer stages, the distribution of field angles, the quantile-quantile plot of plan design dates, the quantile-quantile plot of treatment equipment, the quantile-quantile plot of cancer stages and the quantile-quantile plot of field angles, it is determined whether the radiotherapy plan data conforms to a normal distribution.

[0048] If the radiotherapy planning data does not conform to the normal distribution, the data volume of the radiotherapy planning data is increased.

[0049] Step 202: Perform data cleaning and feature extraction on the multi-modal multi-omics data in sequence to obtain target feature data.

[0050] In a specific application example, the target feature data includes the cleaned radiotherapy plan data, the cleaned radiomics data, similarity data, Hausdorff distance data, target volume ratio, normal organ volume ratio, target dose ratio, normal organ volume ratio, isodose curve graph, region of interest features, sensitivity, accuracy, specificity, area under the curve, and subjective and objective evaluation scores of the artificial neural network.

[0051] Step 202 includes the following steps 21 to 28.

[0052] Step 21: Perform data cleaning on the radiotherapy plan data to obtain the cleaned radiotherapy plan data.

[0053] Specifically, as Figure 5 shown, remove the medical record number, replace the planned design date with the age of the patient at the time of radiotherapy, and replace the plan name with the author's name. Among them, the age of the patient at the time of radiotherapy is equal to the planned design date minus the date of birth. The purpose of replacing the plan name with the author's name is to subsequently judge the influence of subjective factors on the plan design.

[0054] Step 22: Perform Gaussian smoothing filtering on the radiomics data to obtain the cleaned radiomics data.

[0055] Since the extracted data includes radiomics data, and CT images are vulnerable to noise interference during the acquisition process, with a low signal-to-noise ratio, making the boundaries of normal organs and tumors blurred. To improve the image quality and suppress noise, Gaussian smoothing filtering is performed on the original CT images.

[0056] Step 23: Determine the target volume ratio and the normal organ volume ratio according to the target volume, the normal organ volume, and the total CT sequence volume.

[0057] Since the extracted data includes the target volume, the target volume and the normal organ volume are respectively divided by the Body volume (the total volume of the CT sequence) to obtain their respective ratio data, a total of 11 columns of data. The target volume ratios include the ratios of the gross target volume for the primary tumor in the nasopharynx (GTVnx), the ratios of the gross target volume for the regional nodes (GTVnd) seen in clinical examinations and / or imaging, the ratios of the clinical positive lymph node CTV, the ratios of the high-risk lymphatic drainage area (CTV1), and the ratios of the low-risk lymphatic drainage area (CTV2). The normal organ volume ratios include the ratios of the brainstem, the lens, the parotid gland, the sublingual gland, the submandibular gland, and the thyroid gland.

[0058] Step 24, determine the target dose ratio and the normal organ dose ratio according to the target dose, the normal organ dose, and the total prescription dose.

[0059] Divide the target dose and the normal organ dose by the total prescription dose to obtain their respective ratio data, a total of 13 columns of data. The target dose ratios include: the GTVnx dose ratio, the GTVnd dose ratio, the CTV dose ratio, the CTV1 dose ratio, and the CTV2 dose ratio. The normal organ dose ratios include the maximum dose ratio of the brainstem, the maximum dose ratio of the lens, the average dose ratio of the parotid gland, the average dose ratio of the anterior lymphatic drainage area, the average dose ratio of the posterior lymphatic drainage area, the average dose ratio of the sublingual gland, the average dose ratio of the submandibular gland, and the average dose ratio of the thyroid gland.

[0060] Further calculate the Z-score of the dose received by the target and the normal organ, and use the Z-score to cover the original true value. The Z-score measures how many standard deviations the original data deviates from the overall mean of the data and is used to evaluate the distance of the sample point to the overall mean. In this application, the main purpose of calculating the Z-score of the volume and the dose is to determine whether there are outliers according to the interquartile range of the data and eliminate them.

[0061] Step 25, convert the target dose, the normal organ dose, and the total prescription dose into isodose curves.

[0062] Since the extracted data includes dose information, in the process of processing the dose image file, first register the RTDose file with the corresponding CT image. Convert the dose information in the RTDose file into relative dose first, then convert it into isodose curves, and finally output the DICOM image information of the isodose curves.

[0063] Step 26, extract the region of interest features according to the cleaned radiomics data and the isodose curves.

[0064] The region of interest features include gray - level features based on local regions, filtering features based on 3D convergence indices, and features based on Sobel operators. The gray - level features based on local regions include mean, variance, skewness, kurtosis, maximum value, and minimum value.

[0065] Step 27: According to the target volume ratio, the normal organ volume ratio, the target dose ratio, and the normal organ dose ratio, use machine learning methods to predict the number of radiation fields, and determine the sensitivity, accuracy, specificity, and area under the curve of the machine learning methods.

[0066] The machine learning methods include K - Nearest Neighbor algorithm, decision tree, Support Vector Machine (SVM), Bayesian, and Convolutional Neural Networks (CNN).

[0067] The machine learning method can generate training data for the generative artificial intelligence method. This application takes the prediction of the number of radiation fields as an example to introduce the main process of the machine learning method. Based on the data columns obtained in Steps 23 and 24, analyze the correlation between multiple columns of data and the number of radiation fields, and sort them in descending order of correlation. Among them, the correlation analysis uses the correlation coefficient method, which is a statistical index reflecting the degree of closeness of the relationship between variables. The value range of the correlation coefficient is between 1 and - 1. 1 indicates that the two variables are completely linearly correlated, - 1 indicates that the two variables are completely negatively correlated, and 0 indicates that the two variables are not correlated. The formula is: where r xy is the correlation coefficient between x and y, s xy is the covariance between x and y, s x is the variance of x, s y is the variance of y.

[0068] Then select the top 5, top 10, top 15, and top 20 most relevant data columns, and establish five types of machine learning methods, namely K - Nearest Neighbor, decision tree, SVM, Bayesian, and CNN, to predict the number of radiation fields. Finally, generate the sensitivity, accuracy, specificity, and area - under - the - curve data of their respective machine learning methods.

[0069] Step 28: According to the cleaned radiotherapy plan data, the cleaned radiomics data, the similarity data, the Hausdorff distance data, the target volume ratio, the normal organ volume ratio, the target dose ratio, the normal organ dose ratio, the region of interest features, the sensitivity, accuracy, specificity, and area under the curve of the machine learning method, use an artificial neural network to predict the isodose curve and the dose - volume histogram, and determine the subjective and objective evaluation scores of the artificial neural network.

[0070] The basic building block of an artificial neural network is a neuron, which is a node that receives and processes information, including an input layer, hidden layers, and an output layer. Neurons are connected by weights, which act on other neurons. All artificial neural networks in this application are set as follows: The feedforward backpropagation algorithm has four layers: an input layer (with 8 neurons), two hidden layers (each with 50 neurons), and an output layer (with 1 neuron); all layers are connected by the Tan-Sigmoid transfer function; the training function uses the Scaled Conjugate Gradient (SCG) algorithm; the learning function uses the Gradient Descent with Momentum (GDM) algorithm; the weights and biases are randomly initialized; the error threshold is 10 -8 According to the dose limits of organs at risk in "The Quantitative Analysis of Normal Tissue Effects in the Clinic" (QUANTEC), the maximum dose of the patient's brainstem, the maximum dose of the lens, the average dose of the anterior lymphatic drainage area, the average dose of the posterior lymphatic drainage area, the average dose of the sublingual gland, the average dose of the submandibular gland, the average dose of the parotid gland, and the average dose of the thyroid gland are selected as the output criteria from the input data of the input layer. The entire artificial neural network is constructed using the feedforward backpropagation algorithm. The artificial neural network is trained 1000 times to generate as many combinations of training sets and validation sets as possible from the training samples, minimizing the error to the greatest extent possible.

[0071] In the part of evaluating the results of the artificial neural network's predicted dose distribution, both subjective and objective evaluation methods are used. Before the evaluation is implemented, 20 newly diagnosed nasopharyngeal carcinoma patients who meet the requirements are collected, and the doses of normal tissues and target areas are evaluated jointly by the artificial neural network and physicians with work experience at the deputy chief physician level or above. The results of the plans designed by physicists are reviewed by another senior physicist and then participate in the final evaluation.

[0072] The implementation method of the subjective evaluation method is as follows: 5 to 10 physicians with deputy senior professional titles or above evaluate the prediction results of the artificial neural network and the results of the plans made by physicians, and score from aspects such as the coverage of the target area, whether the target area dose meets the requirements of clinical guidelines, and the quality of the dose distribution of the lens, brainstem, spinal cord, submandibular gland, parotid gland, sublingual gland, and anterior and posterior rings. Each item has a full score of 10 points. Finally, the highest and lowest scores are removed, and the average value is obtained.

[0073] For the objective evaluation, a program package that can read CT image sequences and RT Structure is used, which is written based on the Python language. The above ten evaluation aspects are also extracted, and each item is evaluated on a scale of 0 to 10 points. The dose distribution of the same organs is compared for objective evaluation.

[0074] Step 203: According to the target feature data, an artificial intelligence generation method is used to generate a radiotherapy plan for nasopharyngeal carcinoma. The radiotherapy plan for nasopharyngeal carcinoma includes an isodose curve graph of dose distribution, a dose volume histogram of the target area and normal organs, linear accelerator control point data, and multi-leaf collimator movement data.

[0075] In a specific application example, the artificial intelligence generation method is a Contrastive Language-Image Pre-Training (CLIP) model based on contrastive learning.

[0076] Specifically, an artificial intelligence generation method for four data types (isodose curve graph of dose distribution, dose volume histogram of the target area and normal organs, linear accelerator control point data, multi-leaf collimator movement data) is constructed. The CLIP model is used to construct the alignment relationship between text and dose map, and between text and mechanical movement control points. A large amount of paired data of the above-mentioned text / digits and dose map / mechanical movement control points is used for pre-training to learn the alignment relationship therein. After establishing the alignment relationship between text and dose map, and between text and mechanical movement control points, the dose map can obtain the DVH curve through statistical methods, and the mechanical movement control points include the movement data of the linear accelerator gantry and collimator.

[0077] As Figure 6 shown, the CLIP model in this application has two modalities, one is the text modality and the other is the image modality, including two main parts:

[0078] 1) Text encoder: used to convert text / digits into a low-dimensional vector representation. The text editor uses the Transformer architecture.

[0079] 2) Image encoder: used to convert the dose map / mechanical movement control points into a similar vector representation. The image encoder uses the deep residual network (ResNet50) as the basic architecture, and on this basis, improvements such as ResNet-D and anti-aliasing rect-2 blur pooling modifications are made.

[0080] Among them, ResNet50 introduces residual blocks to solve the problem of gradient disappearance in deep networks, and can learn complex features, thus achieving high-precision image classification and object detection. Based on ResNet50, ResNet-D first moves the downsampling of the residual branch to the subsequent 3×3 convolution to avoid a large amount of information loss; then decouples it, and hands over the downsampling of the identity part to average pooling to avoid information loss caused by the simultaneous appearance of 1×1 convolution and downsampling.

[0081] The anti-aliasing rect-2 blur pooling modification is part of ResNet-D, specifically decomposing the max pooling operation into two steps: densely computing the maximum value and simple downsampling, and adding a low-pass filter between the two steps as a means of anti-aliasing. Furthermore, the low-pass filter enhances the max pooling rather than replacing it, thus maintaining translational invariance and translational equivariance.

[0082] This application uses the CLIP model to generate four types of data for 100 newly admitted nasopharyngeal carcinoma patients, namely isodose curves, dose volume histograms of the target area and normal organs, linear accelerator control point data, and multi-leaf collimator motion data. After storing them as files conforming to the data format standard, they are combined correspondingly to form a radiotherapy plan for a patient, including RT-CT sequence files, RT-Dose dose files, and RT-Plan plan files.

[0083] Furthermore, as Figure 7 shown, first preprocess the multimodal omics data (including noise reduction, smoothing, sharpening, filtering, and feature extraction of the region of interest), preprocess the RT Structure file (including extraction of target area delineation information, distribution characteristics of normal organ delineation, distribution characteristics of target area delineation, and segmentation of normal organ files), and integrate and register the preprocessed multimodal omics data with the preprocessed RT Structure file. Generate a radiotherapy plan through an artificial intelligence generation method.

[0084] Specifically, according to the radiotherapy plan of each patient, project the irradiated object from the patient's body CT image sequence onto an electronic portal imaging device on the radiotherapy planning system (Treatment Planning System, TPS) software, convert the three-dimensional dose information into two-dimensional planar dose, and make a QA plan file. Finally, verify the execution of the radiotherapy plan file and the QA plan file on a medical accelerator. Collect the execution data of the medical accelerator, judge whether the actually executed dose is consistent with the dose generated by the CLIP model, and quantitatively calculate the passing rate and the accuracy of the multi-leaf collimator (MLC) movement.

[0085] Compared with the prior art, this application has the following advantages:

[0086] (1) Collected multimodal omics data, and further formed the DICE similarity coefficient and Hausdorff distance on this basis. The data types are rich and the data volume is large, improving the compatibility and practicability of the nasopharyngeal carcinoma radiotherapy plan generation method.

[0087] (2) Compared with using traditional artificial intelligence technology for radiotherapy planning design, the nasopharyngeal carcinoma radiotherapy plan generation method can skip intermediate steps such as predicting radiotherapy parameters and directly generate data that can be executed by a medical linear accelerator, greatly improving the generation efficiency of nasopharyngeal carcinoma radiotherapy plans and shortening the training cycle of staff.

[0088] In summary, the nasopharyngeal carcinoma radiotherapy plan generation method provided by this application can improve the work efficiency of physicists and physicians, replace the repetitive labor of physicists, obtain higher-quality radiotherapy plans more easily in a short time, and reduce the waiting time of patients; and provide a reference basis for physicians to evaluate nasopharyngeal carcinoma radiotherapy plans, provide remote assistance for physicians in low-level hospitals, improve the learning and training efficiency, and promote cooperation between affiliated hospitals; it can also provide a prognosis reference basis for physicians and formulate a radiation protection plan in advance.

[0089] Based on the same inventive concept, the embodiment of this application also provides a nasopharyngeal carcinoma radiotherapy plan generation device for implementing the above-mentioned nasopharyngeal carcinoma radiotherapy plan generation method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the nasopharyngeal carcinoma radiotherapy plan generation device provided below can refer to the limitations on the nasopharyngeal carcinoma radiotherapy plan generation method in the above text, and will not be repeated here.

[0090] In an exemplary embodiment, as Figure 8 shown, a nasopharyngeal carcinoma radiotherapy plan generation device is provided, including: a data acquisition module 801, a data processing module 802, and a plan generation module 803.

[0091] The data acquisition module 801 is used to acquire multi-modal multi-omics data of nasopharyngeal carcinoma patients. The multi-modal multi-omics data includes radiotherapy plan data, radiomics data, similarity data, and Hausdorff distance data.

[0092] The data processing module 802 is used to sequentially perform data cleaning and feature extraction processing on the multi-modal multi-omics data to obtain target feature data.

[0093] The plan generation module 803 is used to generate a nasopharyngeal carcinoma radiotherapy plan according to the target feature data by using an artificial intelligence generation method. The nasopharyngeal carcinoma radiotherapy plan includes an isodose curve diagram of dose distribution, a dose-volume histogram of the target area and normal organs, and a linear acceleration

[0094] In an exemplary embodiment, a computer device is provided. This computer device can be a server or a terminal, and its internal structure diagram can be as Figure 9As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store multi-modal multi-omics data of nasopharyngeal carcinoma patients. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for generating a radiotherapy plan for nasopharyngeal carcinoma.

[0095] Those skilled in the art can understand that Figure 9 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different component layout.

[0096] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0097] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0098] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0099] In the present application, all actions of obtaining signals, information, or data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where the location is located and obtaining authorization from the owner of the corresponding device.

[0100] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0101] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0102] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0103] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for generating a radiotherapy plan for nasopharyngeal carcinoma, characterized in that: The method for generating a nasopharyngeal carcinoma radiotherapy plan comprises: Acquire multimodal multi-omics data of nasopharyngeal carcinoma patients; the multimodal multi-omics data includes radiotherapy planning data, imaging omics data, similarity data and Hausdorff distance data; Performing data cleaning and feature extraction processing on the multimodal multi-omics data in sequence to obtain target feature data; According to the target feature data, an artificial intelligence generation method is used to generate a nasopharyngeal carcinoma radiotherapy plan; the nasopharyngeal carcinoma radiotherapy plan includes an isodose curve diagram of dose distribution, a dose volume histogram of the target area and normal organs, linear accelerator control point data and multi-leaf collimator motion data.

2. The method for generating a nasopharyngeal carcinoma radiotherapy plan according to claim 1, characterized in that: Obtain multimodal multi-omics data of NPC patients, including: Obtain radiotherapy planning data and CT image sequences for patients with nasopharyngeal carcinoma; Extracting features from the CT image sequence to obtain radiomics data; Calculating the similarity between the target area and the normal organ in the CT image sequence to obtain similarity data; The Hausdorff distance between the target area and the normal organ in the CT image sequence is calculated to obtain Hausdorff distance data.

3. The method for generating a nasopharyngeal carcinoma radiotherapy plan according to claim 1, characterized in that: The radiotherapy plan data include the plan design date, treatment equipment, cancer stage, radiation field angle, target volume, normal organ volume, total CT sequence volume, target dose, normal organ dose and total prescribed dose.

4. The method for generating a nasopharyngeal carcinoma radiotherapy plan according to claim 3, characterized in that: The method for generating a nasopharyngeal carcinoma radiotherapy plan further comprises: Determine, based on the radiotherapy plan data, the distribution of plan design dates, the distribution of treatment equipment, the distribution of cancer stage, the distribution of radiation field angles, a quantile-quantile map of plan design dates, a quantile-quantile map of treatment equipment, a quantile-quantile map of cancer stage, and a quantile-quantile map of radiation field angles; Determine whether the radiotherapy plan data conforms to a normal distribution according to the distribution of plan design dates, the distribution of treatment equipment, the distribution of cancer stage, the distribution of field angles, the quantile-quantile graph of plan design dates, the quantile-quantile graph of treatment equipment, the quantile-quantile graph of cancer stage, and the quantile-quantile graph of field angles; If the radiotherapy planning data does not conform to the normal distribution, the data volume of the radiotherapy planning data is increased.

5. The method for generating a nasopharyngeal carcinoma radiotherapy plan according to claim 3, characterized in that: The target feature data include cleaned radiotherapy plan data, cleaned imaging omics data, similarity data, Hausdorff distance data, target volume ratio, normal organ volume ratio, target dose ratio, normal organ volume ratio, isodose curve diagram, region of interest features, sensitivity, accuracy, specificity, area under the curve and subjective and objective evaluation scores of artificial neural network; The multimodal multi-omics data is sequentially cleaned and feature extracted to obtain target feature data, specifically including: Performing data cleaning on the radiotherapy plan data to obtain cleaned radiotherapy plan data; Performing Gaussian smoothing filtering on the imaging omics data to obtain cleaned imaging omics data; Determining a target volume ratio and a normal organ volume ratio according to the target volume, the normal organ volume and the total volume of the CT sequence; Determining a target dose ratio and a normal organ dose ratio according to the target dose, the normal organ dose and the total prescribed dose; Converting the target area dose, the normal organ dose and the total prescription dose into isodose curves; Extracting features of the region of interest based on the cleaned radiomics data and the isodose curve diagram; According to the target volume ratio, the normal organ volume ratio, the target dose ratio and the normal organ dose ratio, a machine learning method is used to predict the number of radiation fields, and the sensitivity, accuracy, specificity and area under the curve of the machine learning method are determined; According to the cleaned radiotherapy plan data, cleaned radiomics data, similarity data, Hausdorff distance data, target volume ratio, normal organ volume ratio, target dose ratio, normal organ dose ratio, region of interest characteristics, sensitivity, accuracy, specificity and area under the curve of the machine learning method, an artificial neural network was used to predict the isodose curve and dose-volume histogram, and the subjective and objective evaluation scores of the artificial neural network were determined.

6. The method for generating a nasopharyngeal carcinoma radiotherapy plan according to claim 5, characterized in that: The region of interest features include grayscale features based on local regions, filtering features based on 3D convergence indexes, and features based on Sobel operators.

7. The method for generating a nasopharyngeal carcinoma radiotherapy plan according to claim 5, characterized in that: The machine learning methods include K nearest neighbor algorithm, decision tree, support vector machine, Bayesian and convolutional neural network.

8. The method for generating a nasopharyngeal carcinoma radiotherapy plan according to claim 1, characterized in that: The artificial intelligence generation method is a language-image pre-training model based on contrastive learning.

9. A nasopharyngeal carcinoma radiotherapy plan generation device, applied to the nasopharyngeal carcinoma radiotherapy plan generation method according to any one of claims 1 to 8, characterized in that: The nasopharyngeal carcinoma radiotherapy plan generating device comprises: A data acquisition module, used to acquire multimodal multi-omics data of nasopharyngeal carcinoma patients; the multimodal multi-omics data includes radiotherapy plan data, imaging omics data, similarity data and Hausdorff distance data; A data processing module, used to perform data cleaning and feature extraction processing on the multimodal multi-omics data in sequence to obtain target feature data; A plan generation module is used to generate a nasopharyngeal carcinoma radiotherapy plan based on the target feature data using an artificial intelligence generation method; the nasopharyngeal carcinoma radiotherapy plan includes an isodose curve diagram of dose distribution, a dose volume histogram of the target area and normal organs, linear accelerator control point data and multi-leaf collimator motion data.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for generating a nasopharyngeal carcinoma radiotherapy plan according to any one of claims 1 to 8.

Images