A method and system for classifying and processing medical imaging data of lung malignant tumors

Abstract

The present invention relates to the field of data processing technology, and specifically to a method and system for classifying and processing medical imaging data of lung malignant tumors, wherein the method comprises obtaining collected lung medical imaging data for preprocessing; establishing and training a lung malignant tumor data classification model based on the preprocessed data; obtaining medical imaging data to be classified and processed, and obtaining a data uncertainty estimate based on the lung malignant tumor data classification model; obtaining the data uncertainty estimate based on a three-branch decision theory comparison, and outputting a classification processing result through a terminal. By preprocessing, modeling, processing, and comparing the medical imaging data, a classification result is obtained, and the data classification processing accuracy is high, which is conducive to doctors making subsequent decisions quickly and accurately.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for classifying and processing medical imaging data of lung malignant tumors. Background Art

[0002] Currently, medical imaging has become the primary method for lung tumor analysis. Computer-assisted analysis is a very promising technology that can significantly improve medical care and the accuracy of data analysis. However, existing models fail to account for the data uncertainty inherent in medical imaging. Specifically, medical imaging data includes various types, such as X-rays, CT scans, and MRIs, all of which are subject to uncertainty. Data uncertainty in X-rays arises from factors such as exposure time and angle, which affect image quality. Data uncertainty in CT scans arises from scanning conditions and parameters, such as the scanner model, scanning speed, and scanning position, all of which affect image quality. Data uncertainty in MRIs arises from scanning parameters, such as the scanner model, scanning speed, and scanning position, all of which affect image quality. Furthermore, different interpretations of the same image by different physicians also contribute to uncertainty.

[0003] Due to these uncertainty issues, the classification and processing accuracy of lung tumor medical imaging data is low, leading to decision-making errors caused by data uncertainty. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for classifying and processing medical imaging data of lung malignant tumors, aiming to solve the problem of low accuracy in classification and processing of medical imaging data of lung tumors, which leads to decision-making errors caused by data uncertainty.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for classifying and processing medical imaging data of lung malignant tumors, which may include the following steps:

[0006] Obtain the collected lung medical imaging data for preprocessing;

[0007] Establish and train a lung malignancy data classification model based on preprocessed data;

[0008] Obtaining medical image data to be classified and processed, and obtaining a data uncertainty estimate based on the lung malignant tumor data classification model;

[0009] Obtaining the data uncertainty estimate is based on a three-way decision theory comparison, and outputting the classification processing result through the terminal.

[0010] In one embodiment, the acquired lung medical image data is preprocessed, wherein the preprocessing process includes:

[0011] Grayscale processing, noise suppression processing, image enhancement processing, and nodule detection processing.

[0012] In one embodiment, a lung cancer data classification model is established and trained based on the preprocessed data, wherein the lung cancer data classification model uses ResNet-18 as a backbone network.

[0013] In one embodiment, the use of ResNet-18 as a backbone network specifically includes:

[0014] Based on ResNet-18, a backbone network was established, excluding the last fully connected layer of ResNet-18. The backbone network included a 7×7 convolutional layer, a batch normalization layer, a 3×3 maximum pooling layer, 4 residual blocks, and a global average pooling layer.

[0015] In one embodiment, a lung cancer data classification model is established and trained based on preprocessed data. The specific steps include:

[0016] For each data sample x i Output two 512-dimensional vectors as the mean u in the feature space i and standard deviation σ i During the training process, the reparameterization technique is used to simulate random sampling in the feature space to obtain the feature vector s i =μ i +εσ i ,ε~N(0,Ι), where I is the unit matrix, that is, ε obeys the standard normal distribution, and the loss function L=L softmax +λL kl Optimize; in practical applications, use the obtained mean u i Characterization feature vector s i ;

[0017] Among them, each data sample x i Representation z in feature space i is a Gaussian distribution, is the classification loss, N is the number of training set samples, y i represents the true label of the i-th training sample, s i Refers to the vector representation of the model for the i-th sample in the feature space, C represents the total number of categories in the classification problem, and W c represents the weight vector associated with the cth class, W yi Indicates that the yth i The weight vector associated with each class, is the KL regularization term, λ is the regularization coefficient, μ is the mean, σ 2 The goal of this loss function is to minimize the KL divergence between the true distribution and the model distribution.

[0018] In one embodiment, obtaining the data uncertainty estimate is based on a three-way decision theory comparison, and outputting the classification processing result through the terminal specifically includes:

[0019] Based on the comparison between the uncertainty estimate and the empirical threshold, a classification result of the medical image being in the positive domain, the negative domain, or the boundary domain is obtained; wherein, when the uncertainty estimate is less than the empirical threshold, a classification result of the medical image being in the positive domain or the medical image being in the negative domain is obtained; when the uncertainty estimate is greater than or equal to the empirical threshold, a classification result of the medical image being in the boundary domain is obtained.

[0020] In a second aspect, the present invention provides a system for classifying and processing medical imaging data of lung malignant tumors, which is applied to the method for classifying and processing medical imaging data of lung malignant tumors as described in the first aspect, and includes a data acquisition module, a data preprocessing module, a model establishment and training module, a data processing and output module, and a classification processing result query module; wherein,

[0021] The data acquisition module is used to collect basic information of the patient and medical imaging data of the patient's lungs captured by medical equipment;

[0022] The data preprocessing module is used to perform grayscale processing, noise suppression processing, image enhancement processing, and nodule detection processing on the collected lung medical image data;

[0023] The model establishment training module establishes and trains a lung malignancy data classification model based on preprocessed data;

[0024] The data processing output module is used to obtain medical imaging data to be classified and processed, and obtain a data uncertainty estimate based on the lung malignant tumor data classification model; the data uncertainty estimate is obtained based on a three-branch decision theory comparison, and the classification processing result is output through the terminal.

[0025] The present invention provides a method and system for classifying and processing medical imaging data for lung malignancies. The method and system preprocess the collected lung medical imaging data; establish and train a lung malignancy data classification model based on the preprocessed data; obtain the medical imaging data to be classified and, based on the lung malignancy data classification model, obtain a data uncertainty estimate; obtain the data uncertainty estimate based on a three-branch decision theory comparison, and output the classification processing result through a terminal. By preprocessing, modeling, processing, and comparing the medical imaging data, a classification result is obtained. The data classification processing has a high accuracy rate, which facilitates doctors to make quick and accurate decisions later. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 This is a flow chart of a method for classifying and processing medical imaging data of lung malignant tumors provided by the present invention.

[0028] Figure 2 This is a schematic diagram of a network structure provided by the present invention;

[0029] Figure 3 This is a schematic diagram of a specific flow chart of step S103 provided by the present invention;

[0030] Figure 4 This is a timing diagram of a lung malignant tumor medical image data classification and processing system provided by the present invention. DETAILED DESCRIPTION

[0031] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0032] The embodiments of the present invention disclose a method and system for classifying and processing medical image data of lung malignant tumors, which implements preprocessing and analysis, model building, processing, comparison, and data classification results for lung medical image data. When the classification result indicates that the medical image is in the positive domain or the medical image is in the negative domain, the result is directly obtained, and the doctor directly makes the corresponding decision. When the classification result indicates that the medical image is in the boundary domain, the corresponding medical image data is transmitted and displayed to the doctor, helping the doctor to make decisions quickly, accurately, and comprehensively, improving the accuracy of the classification processing and shortening the time it takes for the doctor to make subsequent decisions. The following are detailed descriptions of each.

[0033] First, see Figure 1 , Figure 1 The flowchart of the method for classifying and processing medical imaging data of lung malignant tumors provided by the present invention is as follows. The method for classifying and processing medical imaging data of malignant tumors according to the embodiment of the present invention may include the following steps:

[0034] S101, obtaining collected lung medical image data and performing preprocessing;

[0035] In this embodiment, the lung medical imaging data includes X-ray data, CT scan data, and MRI data; the preprocessing process includes grayscale processing, noise suppression processing, image enhancement processing, nodule detection processing, and cropping the nodule area into 64×64 nodule image segments to improve the accuracy of subsequent processing.

[0036] S102. Establishing and training a lung malignancy data classification model based on the preprocessed data;

[0037] In this embodiment, a lung cancer detection model is established and trained. The model uses ResNet-18 as the backbone network. The backbone network includes a 7×7 convolutional layer, a batch normalization layer, a 3×3 maximum pooling layer, 4 residual blocks, and a global average pooling layer. Two fully connected layers are used after the backbone network to perform a multi-layer classification on each data sample x. i Output two 512-dimensional vectors as the mean u in the feature space i (used for modeling feature representation) and standard deviation σ i (used to model data uncertainty), during training, assume that each sample x i Representation z in feature space i For Gaussian distribution, the reparameterization technique is used to simulate random sampling in the feature space to obtain the feature vector s i =μ i +εσ i,ε~N(0,Ι), where I is the unit matrix, that is, ε obeys the standard normal distribution, and the softmax classifier is used to classify and output the data uncertainty value. The model is trained by integrating the model output, data uncertainty and KL regularization loss function, that is, using the loss function L=L softmax +λL kl Optimize.

[0038] in, is the classification loss;

[0039] N is the number of training set samples, y i represents the true label of the i-th training sample, s i Refers to the vector representation of the model for the i-th sample in the feature space, C represents the total number of categories in the classification problem, and W c represents the weight vector associated with the cth class, Indicates that the yth i The weight vector associated with each class;

[0040] is the KL regularization term;

[0041] λ is the regularization coefficient, which is a hyperparameter, μ is the mean, σ 2 The goal of this loss function is to minimize the KL divergence between the true distribution and the model distribution, and the network is trained using stochastic gradient descent.

[0042] S103, obtaining medical imaging data to be classified and processed, and obtaining a data uncertainty estimate based on the lung malignant tumor data classification model; obtaining the data uncertainty estimate based on a three-branch decision theory comparison, and outputting a classification processing result through a terminal.

[0043] In this embodiment, the uncertainty of lung medical imaging data is described by using data uncertainty to calculate the classification loss L cls (composed of softmax and KL divergence regularization), the medical images are divided into positive domain, negative domain and boundary domain according to the loss using three-branch decision theory.

[0044] Specifically, during the training phase, when lung tumor data is used for image processing, a comprehensive judgment is made by combining the neural network decision fusion results and the doctor expert knowledge data, and the neural network is fed back and corrected based on the comprehensive judgment results.

[0045] In the actual application stage, please refer to Figure 2 , Figure 2This is a schematic diagram of a network structure provided by the present invention. The trained lung malignancy classification model outputs the classification result and the data uncertainty estimate. By comparing the uncertainty estimate σ with the empirical threshold θ, the empirical threshold θ can be used to verify the mean of all uncertainty estimates σ output by the network in the data. The comparison index can be Euclidean distance or cosine distance. A decision reliability verification mechanism is established based on the doctor expert knowledge data. Figure 3 , Figure 3 This is a schematic diagram of the specific process of step S103 provided by the present invention. Using the three-branch decision theory, when the uncertainty estimate is less than the empirical threshold (σ<θ), the model result is considered credible, and the model classification result {malignant, benign} is used as the positive and negative domains of the three-branch decision theory. When the uncertainty estimate is greater than or equal to the empirical threshold (σ≥θ), the model result is considered unreliable and is determined to be the boundary domain of the three-branch decision. Subsequently, medical experts will make a re-decision and hand over the prediction result to the medical experts for their decision. This, on the one hand, minimizes misdiagnosis, and on the other hand, uses the expert analysis results as a basis for further feedback correction of the neural network. That is, after the classification processing result is output by the terminal, the method further includes obtaining the final analysis result based on the classification result of the medical image as the boundary domain and inputting it into the lung malignant tumor data classification model for correction, thereby improving the accuracy of the data classification processing. This feedback correction mechanism is a method for dynamically adjusting network parameters or network topology. By combining information from both the model output and expert knowledge data, it further improves the accuracy and reliability of the results, ensuring that the proposed lung malignant tumor medical image data classification processing method has the dual characteristics of high efficiency and high precision.

[0046] In a second aspect, the present invention provides a lung malignant tumor medical image data classification and processing system comprising a data acquisition module, a data preprocessing module, a model building and training module, and a data processing and output module;

[0047] In this implementation, see Figure 4 , Figure 4This is a timing diagram of a lung malignant tumor medical image data classification and processing system provided by the present invention. The data acquisition module is used to collect basic patient information and lung medical image data of patients captured by medical equipment; basic patient information includes name, age, medical history, and contact information; medical images include X-ray data, CT scan data, and MRI data. The user logs in to the system, enters basic information, and chooses whether to automatically upload medical images; when choosing to automatically upload medical images or manually upload medical images, the data preprocessing module performs grayscale processing, noise suppression processing, image enhancement processing, and nodule detection processing on the collected lung medical image data; the model establishment and training module establishes and trains a lung malignant tumor data classification model based on the preprocessed data, that is, obtains the preprocessed data for model establishment and training; the data processing output module is used to obtain medical image data to be classified and obtain a data uncertainty estimate based on the lung malignant tumor data classification model; the data uncertainty estimate is obtained based on a three-branch decision theory comparison, and the classification processing result is output through the terminal. When the classification result is that the medical image is in the positive domain or the medical image is in the negative domain, the result is directly obtained, and the doctor directly makes the corresponding decision. The user can log in to the system to download and print the report. When the classification result is that the medical image is in the boundary domain, the corresponding medical image data will be transmitted and displayed to the doctor. The doctor analyzes the result and the user can log in to the system to download and print the report, helping the doctor to make decisions quickly, accurately and comprehensively, improving the accuracy of the classification process and shortening the time for subsequent doctors to make decisions. For a description of the specific implementation method, please refer to the description of the specific implementation method of S101 to S103 of the first aspect, which will not be repeated here.

[0048] The above disclosure is only a preferred embodiment of the present invention, and certainly cannot be used to limit the scope of the rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for classifying and processing medical imaging data of lung malignant tumors, characterized in that: The following steps are involved: Obtain the collected lung medical imaging data for preprocessing; Establish and train a lung malignancy data classification model based on preprocessed data; The specific steps include: For each data sample x i Output two 512-dimensional vectors as the mean u in the feature space i and standard deviation σ i During the training process, the reparameterization technique is used to simulate random sampling in the feature space to obtain the feature vector s i =μ i +εσ i ,ε~N(0,Ι), where I is the unit matrix, that is, ε obeys the standard normal distribution, and the loss function L=L softmax +λL kl Optimize; in practical applications, use the obtained mean u i Characterization feature vector s i ; Among them, each data sample x i Representation z in feature space i is a Gaussian distribution; is the classification loss; N is the number of training set samples, y i represents the true label of the i-th training sample, s i Refers to the vector representation of the model for the i-th sample in the feature space, C represents the total number of categories in the classification problem, and W c represents the weight vector associated with the cth class, Indicates that the yth i The weight vector associated with each class; is the KL regularization term; λ is the regularization coefficient, μ is the mean, σ 2 The goal of this loss function is to minimize the KL divergence between the true distribution and the model distribution; Obtain medical image data to be classified and processed, and obtain a data uncertainty estimate based on the lung malignant tumor data classification model; obtain the data uncertainty estimate based on a three-branch decision theory comparison; and output the classification processing result through a terminal; specifically including: Based on the comparison between the uncertainty estimate and the empirical threshold, a classification result of the medical image being a positive domain, a negative domain, or a boundary domain is obtained; wherein, when the uncertainty estimate is less than the empirical threshold, a classification result of the medical image being a positive domain or a negative domain is obtained; when the uncertainty estimate is greater than or equal to the empirical threshold, a classification result of the medical image being a boundary domain is obtained; The final analysis result obtained by obtaining the classification result based on the medical image as the boundary domain is input into the lung malignant tumor data classification model for correction.

2. The method for classifying and processing lung malignant tumor medical image data according to claim 1, wherein: The acquired lung medical image data is preprocessed, wherein the preprocessing process includes: Grayscale processing, noise suppression processing, image enhancement processing, and nodule detection processing.

3. The method for classifying and processing medical imaging data of lung malignant tumors according to claim 2, wherein: A lung malignant tumor data classification model is established and trained based on the preprocessed data, wherein the lung malignant tumor data classification model uses ResNet-18 as the backbone network.

4. The method for classifying and processing lung malignant tumor medical image data according to claim 3, wherein: The use of ResNet-18 as the backbone network specifically includes: Based on ResNet-18, a backbone network was established, excluding the last fully connected layer of ResNet-18. The backbone network included a 7×7 convolutional layer, a batch normalization layer, a 3×3 maximum pooling layer, 4 residual blocks, and a global average pooling layer.

5. A system for classifying and processing medical image data of malignant lung tumors, applied to the method for classifying and processing medical image data of malignant lung tumors according to any one of claims 1 to 4, characterized in that: It includes data acquisition module, data preprocessing module, model building and training module, data processing output module and classification processing result query module; among them, The data acquisition module is used to collect basic information of the patient and medical imaging data of the patient's lungs captured by medical equipment; The data preprocessing module is used to perform grayscale processing, noise suppression processing, image enhancement processing, and nodule detection processing on the collected lung medical image data; The model establishment training module establishes and trains a lung malignancy data classification model based on preprocessed data; The data processing output module is used to obtain medical imaging data to be classified and processed, and obtain a data uncertainty estimate based on the lung malignant tumor data classification model; the data uncertainty estimate is obtained based on a three-branch decision theory comparison, and the classification processing result is output through the terminal.

Images