CN109065165B - A prediction method for chronic obstructive pulmonary disease based on reconstructed airway tree images - Google Patents

Abstract

Translated fromChinese

本发明提供一种基于重建气道树图像的慢性阻塞性肺病预测方法，涉及医学图像处理技术领域。该方法首先采集来自同一家医院的COPD患者和健康人的多层CT图像文件，生成健康人和COPD患者的重建气道树图像，并转换不同的视角对生成的健康人和COPD患者的重建气道树图像进行截图，然后去除所截取气道树图像的多余背景信息，并基于卷积神经网络进行训练并分类，预测测试集中是否有人患有COPD；最后对重建气道树截取的图像集进行十字交叉验证，保证预测结果的准确性。本发明提供的基于重建气道树图像的慢性阻塞性肺病预测方法，将作为一种初步筛选该疾病的技术手段，准确且高效，有效避免了误诊和漏诊；同时也免去了肺功能检查的复杂过程，大大减少了医院工作量。

The invention provides a method for predicting chronic obstructive pulmonary disease based on reconstructed airway tree images, and relates to the technical field of medical image processing. The method first collects multi-slice CT image files of COPD patients and healthy people from the same hospital, generates reconstructed airway tree images of healthy people and COPD patients, and converts different perspectives to the generated reconstructed airway images of healthy and COPD patients. Take a screenshot of the airway tree image, then remove the redundant background information of the intercepted airway tree image, and train and classify based on the convolutional neural network to predict whether someone has COPD in the test set; Cross-validation to ensure the accuracy of the prediction results. The method for predicting chronic obstructive pulmonary disease based on the reconstructed airway tree image provided by the present invention will be used as a technical means for preliminary screening of the disease, which is accurate and efficient, and effectively avoids misdiagnosis and missed diagnosis; The complex process greatly reduces the workload of the hospital.

Description

Chronic obstructive pulmonary disease prediction method based on reconstructed airway tree image

Technical Field

The invention relates to the technical field of medical image processing, in particular to a chronic obstructive pulmonary disease prediction method based on a reconstructed airway tree image.

Background

Chronic Obstructive Pulmonary Disease (COPD) is a common respiratory Disease, seriously harms the physical and mental health of patients, and has become an important problem which must be faced by global public health. Thus, effective diagnostic means are also of great importance.

In traditional clinical Tests, lung respiratory Function Tests (PFTs) are frequently used, the ratio of forced expiratory volume (FEV1) in the first second after bronchodilation to Forced Vital Capacity (FVC) is less than 70% as a standard for accurate diagnosis, and PFTs measure macroscopic functional parameters of the whole lung and cannot provide structural information of an airway tree of COPD lungs. In addition, the ratio of FEV1 to FVC of healthy population decreases with age, so the method is also prone to misdiagnosis and missed diagnosis.

Meanwhile, the pulmonary function examination process in hospitals is very complicated and long in steps, and generally comprises ventilation function, respiratory regulation and pulmonary circulation function. The detection criteria are often different, and clinical diagnosis of COPD requires comprehensive assessment in combination with symptoms, health status and the like of patients, and some patients with severe disease cannot tolerate the disease and are difficult to distinguish from airflow limitation caused by other diseases. Meanwhile, medical staff and workers have limited knowledge about relevant knowledge, and clinical application is greatly limited.

In summary, the current diagnostic methods for COPD have many defects, which increase the workload of doctors and the pain of patients, and the diagnostic results are not satisfactory.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a chronic obstructive pulmonary disease prediction method based on a reconstructed airway tree image, which realizes the prediction of chronic obstructive pulmonary disease.

A chronic obstructive pulmonary disease prediction method based on reconstructed airway tree images comprises the following steps:

step 1, based on a medical image segmentation and three-dimensional visualization method, respectively extracting airway trees of COPD patients and healthy people, and storing the airway tree images, wherein the specific method comprises the following steps:

step 1.1, reading collected multilayer CT image files of COPD patients and healthy people from the same hospital by adopting software, and generating reconstructed airway tree images of the healthy people and the COPD patients;

step 1.2, setting the background of the generated reconstructed airway tree images of healthy people and COPD patients to be white;

step 1.3, converting different visual angles to respectively capture the generated reconstructed airway tree images of healthy people and COPD patients, simultaneously controlling the size and resolution of the captured images to be consistent, and storing the captured airway tree images;

step 2, removing redundant background information of the intercepted airway tree image with the same size and resolution;

step 3, training and classifying by using the intercepted airway tree color images with different visual angles based on a 3-channel convolutional neural network model, and predicting whether people in the test set have COPD or not, wherein the specific method comprises the following steps:

3.1, setting a training set, a verification set and a test set of the convolutional neural network model according to the ratio of 8: 1 based on the image intercepted by the reconstructed airway tree;

step 3.2, constructing a convolutional neural network structure comprising an input layer, a convolutional layer, a pooling layer and a full-connection layer; the first layer of the convolutional neural network is an input layer, the second layer is a convolutional layer C1, the convolutional layer C1 is next to a Batch Normalization layer and a ReLU activation function layer, the third layer is a convolutional layer C2, the next to the Batch Normalization layer and the ReLU activation function layer, the fourth layer is a maximum pooling layer P1, the fifth layer is a convolutional layer C3, the sixth layer is a convolutional layer C4, the seventh layer is a maximum pooling layer P2, the eighth layer is an average pooling layer P3, the next to the dropout layer, and the ninth layer is a full-connection layer F2;

3.3, training the convolutional neural network model, and determining the optimal parameters of the convolutional neural network model according to the training time and the accuracy of the verification set;

step 3.4, inputting the test set into the trained convolutional neural network model, and further predicting whether people in the test set have COPD;

and 4, performing cross validation on the image set intercepted by the reconstructed airway tree to ensure the accuracy of the prediction result.

According to the technical scheme, the invention has the beneficial effects that: the method for predicting the chronic obstructive pulmonary disease based on the reconstructed airway tree image is accurate and efficient, and misdiagnosis and missed diagnosis are effectively avoided; meanwhile, the method can be used as a technical means for primarily screening the chronic obstructive pulmonary disease before the examiner confirms the diagnosis, and the complex process of fussy process and long steps of the past hospital pulmonary function examination is avoided, so that the pain of the examiner and the workload of the hospital are greatly reduced; in addition, the invention also provides a new way different from the past lung function examination, and aims to predict the chronic obstructive pulmonary disease from the perspective of reconstructing an airway tree image by multi-layer CT directly, thereby overcoming the defect that most of primary hospitals have no lung function instruments.

Drawings

Fig. 1 is a flowchart of a method for predicting chronic obstructive pulmonary disease based on a reconstructed airway tree image according to an embodiment of the present invention;

fig. 2 is a reconstructed airway tree image of a COPD patient and an airway tree image of a healthy person in a multi-slice CT image provided by an embodiment of the present invention, wherein (a) is the reconstructed airway tree image of the healthy person, and (b) is the reconstructed airway tree image of the COPD patient;

fig. 3 is an airway tree image of a healthy person captured from 3 different viewing angles based on CT image reconstruction according to an embodiment of the present invention; wherein (a) is a front view of an airway tree reconstructed by a healthy person based on a CT image; (b) a back view of an airway tree reconstructed for a healthy person based on a CT image; (c) a top oblique view of an airway tree reconstructed for a healthy person based on a CT image;

fig. 4 is an airway tree image of a COPD patient at 3 different viewing angles, which is taken after being reconstructed based on a CT image, provided by an embodiment of the present invention, wherein (a) is a front view of an airway tree which is reconstructed by the COPD patient based on the CT image; (b) a back view of an airway tree reconstructed based on CT images for COPD patients; (c) a top-down oblique view of an airway tree reconstructed based on CT images for COPD patients;

FIG. 5 is a diagram of removing redundant background images of an airway tree image reconstructed based on a CT image according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of a simplified convolutional neural network model according to an embodiment of the present invention.

Detailed Description

The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.

A method for predicting chronic obstructive pulmonary disease based on reconstructed airway tree images, as shown in fig. 1, includes the following steps:

step 1, based on a medical image segmentation and three-dimensional visualization method, respectively extracting airway trees of a COPD patient and a healthy person, wherein the specific method comprises the following steps:

step 1.1, reading collected multilayer CT image files of COPD patients and healthy persons from the same hospital by adopting software, and generating reconstructed airway tree images of the healthy persons and the COPD patients as shown in figure 2;

the data used in this example are all from the same hospital, and data acquisition of 190 patients with COPD and data acquisition of 90 healthy people are performed, and the training set, the validation set and the test set are set according to the ratio of 8: 1.

Step 1.2, in order to facilitate processing of the reconstructed airway tree image, namely, saving a large amount of time cost required for processing a large amount of image backgrounds, setting the generated backgrounds of the reconstructed airway tree images of healthy people and COPD patients to be white;

step 1.3, converting different visual angles to respectively perform screenshot on reconstructed airway tree images generated by healthy people and COPD patients, simultaneously controlling the size and resolution of the intercepted images to be consistent, and storing the intercepted airway tree images;

in this embodiment, images captured by healthy people based on 3 different viewing angles of the airway tree reconstructed by the CT image are shown in fig. 3; an image of a COPD patient taken from 3 different perspectives of an airway tree reconstructed from a CT image is shown in figure 4.

In this embodiment, in order to ensure that the converted view angles of the reconstructed airway tree images of each COPD patient and healthy person are the same, and therefore the view angles cannot be manually converted, a direction button provided by the method needs to be selected for operation in a three-dimensional visualization method, for example, when a Front image is selected to be captured, a Select 3D view button needs to be clicked on a page of the reconstructed image, and Front needs to be selected. Similarly, for the Back view and the oblique view (rotated 45 ° downward based on the front view), Back and Isometric will be selected in the Select 3D view button, respectively.

In order to ensure that the size and resolution of the reconstructed airway tree image are the same for each COPD patient and healthy person, a default value should be selected for three-dimensional visualization when saving the reconstructed airway tree image, and the reconstructed 3D airway tree image should not be enlarged or reduced, and likewise, should not be manually truncated.

Step 2, removing redundant background information of the intercepted airway tree image with the same size and resolution;

in this embodiment, the airway tree image with the redundant background information removed is shown in fig. 5. Since the smaller the image size when training the convolutional neural network, the lower the time cost consumed for training. Therefore, here, removing the extra white background does not affect the loss of image information.

The method for removing the redundant background is specifically as follows: detecting images from top to bottom and from bottom to top according to lines respectively, stopping when useful information of the images is detected, and recording the number of lines; and detecting the images from left to right and from right to left according to the columns respectively, stopping when useful information of the images is detected, and recording the number of the columns respectively. And (3) performing difference on the numerical values recorded in the two rows and the numerical values recorded in the two columns, comparing the difference values, selecting a larger numerical value to perform square interception on the airway tree image, and storing the acquired square image to be used as a data set for later training.

Step 3, training and classifying by using the intercepted airway tree color images with different visual angles based on a convolutional neural network model with 3 channels, and predicting whether people in the test set have COPD or not, wherein the specific method comprises the following steps:

3.1, setting a training set, a verification set and a test set of the convolutional neural network model according to the ratio of 8: 1 based on the image intercepted by the reconstructed airway tree;

step 3.2, constructing a convolutional neural network structure comprising an input layer, a convolutional layer, a pooling layer and a full-link layer as shown in FIG. 6; the first layer of the convolutional neural network is an input layer, the second layer is a convolutional layer C1, the convolutional layer C1 is next to a Batch Normalization layer and a ReLU activation function layer, the third layer is a convolutional layer C2, the next to the Batch Normalization layer and the ReLU activation function layer, the fourth layer is a maximum pooling layer P1, the fifth layer is a convolutional layer C3, the sixth layer is a convolutional layer C4, the seventh layer is a maximum pooling layer P2, the eighth layer is an average pooling layer P3, the next to the dropout layer, and the ninth layer is a fully-connected layer F2;

where the input layer size is 224 x 224, the filter size of the convolutional layer is 3 x 3, the second layer has 64 convolution kernels, and the fifth and sixth layers have 128 convolution kernels.

3.3, training the convolutional neural network model, and determining the optimal parameters of the convolutional neural network model according to the training time and the accuracy of the verification set;

step 3.4, inputting the test set into the trained convolutional neural network model, and further predicting whether people in the test set have COPD;

and 4, performing cross validation on the image set intercepted by the reconstructed airway tree to ensure the accuracy of the prediction result, namely, recycling the training set, the validation set and the test set, and averaging the accuracy rates of all the prediction results.

In this embodiment, the average value of the accuracy of the prediction results is 90%.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit of the corresponding technical solutions and scope of the present invention as defined in the appended claims.

Claims (1)

Translated fromChinese

1.一种基于重建气道树图像的慢性阻塞性肺病预测方法，其特征在于：包括以下步骤：1. a method for predicting chronic obstructive pulmonary disease based on reconstructed airway tree image, is characterized in that: comprise the following steps:步骤1、基于医学影像分割和三维可视化方法，分别提取COPD患者和健康人的气道树，并保存气道树图像，具体方法为：Step 1. Based on medical image segmentation and three-dimensional visualization methods, extract the airway trees of COPD patients and healthy people respectively, and save the airway tree images. The specific methods are as follows:步骤1.1、采用软件读取采集的来自同一家医院的COPD患者和健康人的多层CT图像文件，并生成健康人和COPD患者的重建气道树图像；Step 1.1. Use software to read the collected multi-slice CT image files of COPD patients and healthy people from the same hospital, and generate reconstructed airway tree images of healthy people and COPD patients;步骤1.2、将生成的健康人和COPD患者的重建气道树图像的背景设置为白色；Step 1.2. Set the background of the generated reconstructed airway tree images of healthy people and COPD patients to white;步骤1.3、转换不同的视角分别对生成的健康人和COPD患者的重建气道树图像进行截图，同时控制所截取的图像尺寸大小和分辨率一致，并保存截取的气道树图像；Step 1.3. Convert different perspectives to take screenshots of the generated reconstructed airway tree images of healthy people and COPD patients, while controlling the size and resolution of the captured images to be consistent, and save the captured airway tree images;步骤2、去除所截取的相同尺寸大小和分辨率的气道树图像的多余背景信息；Step 2. Remove redundant background information of the captured airway tree image of the same size and resolution;步骤3、利用截取的不同视角的气道树彩色图像，基于3通道的卷积神经网络模型进行训练并分类，预测测试集中是否有人患有COPD；Step 3. Use the intercepted color images of the airway tree from different perspectives, train and classify based on the 3-channel convolutional neural network model, and predict whether anyone in the test set suffers from COPD;步骤4、对重建气道树截取的图像集进行十字交叉验证，保证预测结果的准确性；Step 4. Perform cross-validation on the image set intercepted by the reconstructed airway tree to ensure the accuracy of the prediction results;所述步骤3的具体方法为：The specific method of the step 3 is:步骤3.1、基于重建气道树截取的图像，将卷积神经网络模型的训练集、验证集和测试集按照8：1：1的比例进行设置；Step 3.1. Based on the image intercepted by the reconstructed airway tree, set the training set, validation set and test set of the convolutional neural network model according to the ratio of 8:1:1;步骤3.2、构建包括输入层、卷积层、池化层和全连接层的卷积神经网络结构；所述卷积神经网络的第一层为输入层，第二层为卷积层C1，卷积层C1紧接着Batch Normalization层和ReLU激活函数层，第三层为卷积层C2，后面紧跟着Batch Normalization层和ReLU激活函数层，第四层是最大池化层P1，第五层为卷积层C3，第六层为卷积层C4，第七层为最大池化层P2，第八层为均值池化P3，后面紧接着dropout层，第九层为全连接层F2；Step 3.2, construct a convolutional neural network structure including an input layer, a convolutional layer, a pooling layer and a fully connected layer; the first layer of the convolutional neural network is the input layer, the second layer is the convolutional layer C1, the volume The layer C1 is followed by the Batch Normalization layer and the ReLU activation function layer, the third layer is the convolutional layer C2, followed by the Batch Normalization layer and the ReLU activation function layer, the fourth layer is the maximum pooling layer P1, and the fifth layer is The convolutional layer C3, the sixth layer is the convolutional layer C4, the seventh layer is the maximum pooling layer P2, the eighth layer is the mean pooling layer P3, followed by the dropout layer, and the ninth layer is the fully connected layer F2;步骤3.3、训练卷积神经网络模型，通过训练时间和验证集的准确率来确定卷积神经网络模型的最优参数；Step 3.3, train the convolutional neural network model, and determine the optimal parameters of the convolutional neural network model through the training time and the accuracy of the validation set;步骤3.4、将测试集输入训练好的卷积神经网络模型中，进而预测测试集中是否有人患有COPD。Step 3.4. Input the test set into the trained convolutional neural network model, and then predict whether anyone in the test set has COPD.

Images