A method for establishing a joint regression prediction model for radiation pneumonitis

Abstract

The present invention discloses a method for establishing a joint regression prediction model for radiation pneumonia, establishes a multi-feature fusion regression prediction model for joint prediction of machine learning and deep learning, introduces radiomics features, expert models, transfer learning and other methods on the basis of deep learning to realize the classification of radiation pneumonia, establishes a prediction model through machine learning algorithm, fuses different features, and jointly predicts the prognosis of radiation pneumonia. The present invention adopts a method combining theoretical analysis, multivariate regression, neural network modeling and experimental verification, closely links the prediction research of radiation pneumonia with actual diagnosis, establishes rule constraints on diagnostic experience and mines implicit laws of historical data, so that the prediction results and actual diagnosis can be mutually verified and promoted.

Description

Technical Field

[0001] The patent of this invention belongs to the technical field of radiation pneumonia prediction, and specifically relates to a method for establishing a joint regression prediction model for radiation pneumonia. Background Art

[0002] At present, there are two main research directions for the prediction of radiation pneumonia: extracting imaging features from lung radiographs and using machine learning algorithms to establish classification models for prognosis prediction; and using deep learning algorithms to train imaging data and perform classification and prediction based on the training results.

[0003] (1) Research on prediction of radiation pneumonitis based on machine learning

[0004] Imaging plays a key role in the diagnosis of lung cancer, breast cancer and other malignant tumors of the chest, including early diagnosis, efficacy monitoring and prognosis assessment, which are inseparable from medical imaging. With the development of computer science, many computer-assisted analysis technologies have gradually emerged in recent years. The maturity of related technologies has also led to the gradual integration of computer-assisted analysis and medical diagnosis. Among them, radiomics is a hot topic in current research. Radiomics refers to the high-throughput extraction of quantitative image features and the conversion of medical images into data resources that can be further mined. Through specific methods and procedures, it explores the correlation between imaging and diseases. The rapid development of artificial intelligence technology has further promoted the application of radiomics in actual clinical diagnosis. The main way to combine artificial intelligence and radiomics is to deeply explore the intrinsic relationship between image features extracted by radiomics and diseases through machine learning methods, screen effective and reliable features from the extracted high-dimensional features, eliminate redundant features, and build machine learning models to detect different disease features. In order to establish this model, many scholars have made many attempts in this direction. Zhang Zhen et al. screened the radiomic features related to the occurrence of radiation pneumonia based on the chest positioning CT images of lung cancer patients, constructed a machine learning model, and explored the value of radiomics in predicting the occurrence of radiation pneumonia; Guo Meiying et al. proposed the method of intelligent imaging omics, which combines imaging omics with artificial intelligence methods, and uses artificial intelligence methods to screen and construct imaging markers to solve different medical problems while deeply mining image features; Chen Wentao et al. explained the mechanism and clinical imaging manifestations of radiation pneumonia, grading and related factor analysis, and then introduced the principles of several conventional machine learning algorithms, and constructed a model for predicting radiation pneumonia through imaging omics and machine learning methods. In addition, the hypothesis of imaging omics believes that changes in microscopic genes or protein patterns are expressed in macroscopic images, so artificial intelligence methods can further explore the potential molecular biological mechanisms reflected by imaging markers.

[0005] (2) Research on prediction of radiation pneumonitis based on deep learning

[0006] Deep learning is an important branch of machine learning. It combines low-level features to form more abstract high-level representation attribute categories or features, learns the internal laws and representation levels of sample data, and enables computers to have human-like thinking and analysis capabilities, identify things, and even predict the development of things. The most important foundation in the field of deep learning is the neural network, which is a computer model abstracted from the operation mode of the human brain nervous system. The neural network is a network structure model formed by connecting neurons to each other. Neurons have a single output corresponding to a weight coefficient and are connected to other similar neurons as their input. Neural networks obtain information through storage and processing. After training, neural networks generate corresponding model structures. Neural networks can mine hidden relationships between data, and their comprehensive analysis and processing capabilities for data are superior to traditional statistical methods. They have great application value in medicine. Neural networks can be used to assist doctors in diagnosing diseases and predicting prognosis. At present, the research in the direction of deep learning is still in its infancy. Chen et al. used neural networks to build prediction models for radiation pneumonia and judge the identification ability of the models, and screen the advantageous features of the models; Wu Yunfeng et al. proposed a lung CT classification model based on the improved Inception-ResNet and an image semantic segmentation model based on the improved U-Net full convolution network to classify and detect radiation pneumonia; Wang B et al. established a coronavirus pneumonia model based on CT image data of 723 positive pneumonia and 413 negative pneumonia. The above studies all have problems such as small data volume and shallow exploration and research in the direction of deep learning. Therefore, it is necessary to further study the problem of radiation pneumonia prediction in combination with deep learning algorithms.

[0007] The native features and derived features extracted from medical CT images are the key data for reference in the prediction of radiation pneumonia. How to extract native CT radiomic features and mine various derived image features is the key to accurately predict radiation pneumonia. The native CT radiomic features alone can fit the radiomic data curve well, but cannot be combined with the disease progression in the patient area, and it is often difficult to fit the real objective laws when there are fewer data sets. Deep learning algorithms can accurately detect targets through multi-scale feature channels, but highly abstract features will lose local detail information, focus on the category attribution of the target, strengthen the classification ability, and have weak regression prediction ability. Therefore, how to establish a fusion strategy on the radiomic features of CT images and the abstract features of deep learning, and establish a joint regression prediction model through machine learning methods, is the core issue for achieving accurate prediction of radiation pneumonia.

[0008] In order to solve the above problems, this paper proposes a method to establish a joint regression prediction model for radiation pneumonitis. Summary of the invention

[0009] In order to solve the above technical problems, the present invention designs a method for establishing a joint regression prediction model for radiation pneumonia. The present invention establishes a multi-feature fusion regression prediction model for joint prediction of machine learning and deep learning. On the basis of deep learning to realize the classification of radiation pneumonia, imaging genomics features, expert models, transfer learning and other methods are introduced. A prediction model is established through a machine learning algorithm, and different features are fused to jointly predict the prognosis of radiation pneumonia.

[0010] In order to achieve the above technical effects, the present invention is implemented by the following technical solutions: A method for establishing a joint regression prediction model for radiation pneumonitis, comprising the following steps:

[0011] S1. The radiomics mapping theory of radiation pneumonia is used to guide the radiomics feature extraction, clean and preprocess the original CT image data, and extract high-correlation features;

[0012] S2. Establish a classification algorithm based on a deep learning neural network model. Modify the network model according to the mapping properties of the 3D image of radiation pneumonia CT in the 2D image, strengthen the network's retention of the overall characteristics, and improve the network's ability to detect small targets. At the same time, establish a transfer learning training process, use pneumonia and obstructive pneumonia and other diseases with similar characteristics on CT images to train the network, and realize a network model with high precision, high stability and strong generalization ability;

[0013] S3. Establish a prediction model for radiation pneumonia based on machine learning: Build a decision tree expert model based on the experience of professional physicians, select rule experience based on the principle of "information entropy", build an elastic network regression algorithm based on the expert model's score, imaging genomics characteristics, and deep learning output category confidence, and continuously optimize the regression algorithm based on historical data to ultimately achieve the prediction of radiation pneumonia.

[0014] Furthermore, the S2 specifically includes: first, preprocessing the medical CT images of radiation pneumonia and the medical CT images of similar lung diseases, reducing the dimension of 3D medical CT image data, and unifying them into a neural network file for feature extraction; secondly, modifying the network model of the YOLOX algorithm, based on the characteristic expression properties of radiation pneumonia, modifying the convolution layer, normalization layer, up and down sampling layer, and classification head, and continuously debugging and verifying the network performance; secondly, performing transfer learning on the modified neural network, using the modified network model to pre-train a similar lung disease data set, and then migrating the pre-trained model weights to the model to adjust the radiation pneumonia data set, and continuously adjusting the loss function solution strategy to achieve an ideal fitting state; finally, performing classification detection on the untrained CT images, verifying the network model performance, and saving the classification results and confidence as output.

[0015] Furthermore, the S3 specifically includes: first, conducting in-depth exchanges with professional physicians engaged in the diagnosis of radiation pneumonia, recording reliable experience in the diagnosis of radiation pneumonia, and establishing an expert model using a decision tree algorithm; second, using the information of radiomics and deep learning classification detection as the input of the elastic regression network to train the network; finally, using the K-fold validation method to test the prediction accuracy of the model for radiation pneumonia, and giving the order of correlation between different features and radiation pneumonia.

[0016] The beneficial effects of the present invention are as follows: the present invention establishes a multi-feature fusion regression prediction model for joint prediction of machine learning and deep learning, introduces methods such as radiomics features, expert models, and transfer learning on the basis of deep learning to achieve classification of radiation pneumonia, establishes a prediction model through a machine learning algorithm, fuses different features, and jointly predicts the prognosis of radiation pneumonia; adopts a method combining theoretical analysis, multivariate regression, neural network modeling, and experimental verification to closely link the prediction research of radiation pneumonia with actual diagnosis, and establishes rule constraints on diagnostic experience and mines implicit laws in historical data to make the prediction results and actual diagnosis mutually confirm and promote each other. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 is a flow chart of the present invention; DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example 1

[0020] like Figure 1 As shown, a method for establishing a joint regression prediction model for radiation pneumonitis includes the following steps:

[0021] S1. The radiomics mapping theory of radiation pneumonia is used to guide the radiomics feature extraction, clean and preprocess the original CT image data, and extract high-correlation features;

[0022] S2. Establish a classification algorithm based on a deep learning neural network model. Modify the network model according to the mapping properties of the 3D image of radiation pneumonia CT in the 2D image, strengthen the network's retention of the overall characteristics, and improve the network's ability to detect small targets. At the same time, establish a transfer learning training process, use pneumonia and obstructive pneumonia and other diseases with similar characteristics on CT images to train the network, and realize a network model with high precision, high stability and strong generalization ability;

[0023] S3. Establish a prediction model for radiation pneumonia based on machine learning: Build a decision tree expert model based on the experience of professional physicians, select rule experience based on the principle of "information entropy", build an elastic network regression algorithm based on the expert model's score, imaging genomics characteristics, and deep learning output category confidence, and continuously optimize the regression algorithm based on historical data to ultimately achieve the prediction of radiation pneumonia. Example 2

[0024] like Figure 1 As shown, the specific content of S2 is: firstly, the medical CT images of radiation pneumonia and the medical CT images of similar lung diseases are preprocessed, the 3D medical CT image data are reduced in dimension, and unified into a neural network file for feature extraction;

[0025] Secondly, the network model of the YOLOX algorithm was modified. Based on the characteristic expression properties of radiation pneumonia, the convolution layer, normalization layer, up and down sampling layer, and classification head were modified to continuously debug and verify the network performance.

[0026] Secondly, transfer learning is performed on the modified neural network. The modified network model is used to pre-train a similar lung disease dataset. The pre-trained model weights are then transferred to the model to adjust the radiation pneumonia dataset. The loss function solution strategy is continuously adjusted to achieve an ideal fitting state.

[0027] Finally, the untrained CT images are classified and tested to verify the performance of the network model, and the classification results and confidence levels are saved as output. Example 3

[0028] like Figure 1 As shown, the specific content of S3 is: first, conduct in-depth exchanges with professional physicians engaged in the diagnosis of radiation pneumonia, record reliable experience in the diagnosis of radiation pneumonia, and use decision tree algorithms to establish an expert model;

[0029] Secondly, the information of radiomics and deep learning classification detection is used as the input of elastic regression network to train the network;

[0030] Finally, the K-fold validation method was used to test the prediction accuracy of the model for radiation pneumonitis, and the order of correlation between different features and radiation pneumonitis was given.

[0031] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0032] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for establishing a joint regression prediction model for radiation pneumonitis, characterized in that: The following steps are involved: S1. The radiomics mapping theory of radiation pneumonia is used to guide the radiomics feature extraction, clean and preprocess the original CT image data, and extract high-correlation features; S2. Establish a classification algorithm based on a deep learning neural network model. Modify the network model according to the mapping properties of the 3D image of radiation pneumonia CT in the 2D image, strengthen the network's retention of the overall characteristics, and improve the network's ability to detect small targets. At the same time, establish a transfer learning training process, use pneumonia and obstructive pneumonia and other diseases with similar characteristics on CT images to train the network, and realize a network model with high precision, high stability and strong generalization ability; S3. Establish a prediction model for radiation pneumonia based on machine learning: Build a decision tree expert model based on the experience of professional physicians, select rule experience based on the principle of "information entropy", build an elastic network regression algorithm based on the expert model's score, radiomics features, and deep learning output category confidence, and continuously optimize the regression algorithm based on historical data to ultimately achieve the prediction of radiation pneumonia; The S2 specifically includes: first, preprocessing the medical CT images of radiation pneumonia and the medical CT images of similar lung diseases, reducing the dimension of 3D medical CT image data, and unifying them into a neural network file for feature extraction; secondly, modifying the network model of the YOLOX algorithm, based on the characteristic expression properties of radiation pneumonia, modifying the convolution layer, normalization layer, up and down sampling layer, and classification head, and continuously debugging and verifying the network performance; secondly, performing transfer learning on the modified neural network, using the modified network model to pre-train a similar lung disease data set, and then migrating the pre-trained model weights to the model to adjust the radiation pneumonia data set, and continuously adjusting the loss function solution strategy to achieve an ideal fitting state; finally, performing classification detection on untrained CT images, verifying the network model performance, and saving the classification results and confidence as output; The S3 specifically includes: first, conducting in-depth exchanges with professional physicians engaged in the diagnosis of radiation pneumonia, recording reliable experience in the diagnosis of radiation pneumonia, and establishing an expert model using a decision tree algorithm; second, using the information of radiomics and deep learning classification detection as the input of the elastic regression network to train the network; finally, using the K-fold validation method to test the prediction accuracy of the model for radiation pneumonia, and giving the order of correlation between different features and radiation pneumonia.

Images