KR20100010973A - Method for automatic classifier of lung diseases - Google Patents

Abstract

PURPOSE: A method for automatic classifier of lung diseases is provided to increase the reliability for diagnosing diseases by automatically classifying the disease in the lung area. CONSTITUTION: The closed region is classified from 3D breast volume data. The 3D breast volume data obtains through the CT(Computerized Tomography) technique. The picture quality preprocessing operates for texture and morphological analysis of the closed region(S110). The texture and form of the closed region are compared with texture/configuration data of the pre-stored specific disease lungs(S120). According to the comparison, each part of the closed region is classified according to the specific disease(S130).

Description

Method for automatic classifier of lung diseases

The present invention relates to a method for automatically classifying lung diseases, and more particularly, to a method for automatically classifying lung diseases using texture and shape analysis of CT images of patients with lung diseases.

Diffuse interstitial lung disease (DILD) is a disease that affects various lung parenchyma, causing respiratory distress. These diffuse lung diseases cover about 200 various lung parenchymal abnormalities, so it is difficult to know the size or progression of lung parenchymal abnormalities.

Recently, high-resolution computed tomography (CT) has been used as a standard for imaging tests for diagnosing diffuse lung diseases. The imaging characteristics of the main High Resolution Computed tomography (HRCT), which is used for the diagnosis of diffuse lung disease, are used to identify the presence and characteristics of various lung parenchymal abnormalities due to destruction or overexpansion of the lung parenchyma.

However, the accuracy of the differential diagnosis and its evaluation using characteristics such as the distribution and shape of various abnormalities with respect to the lung parenchyma is known to be limited in the above method. This limitation is due to the problem of changing the parenchyma according to the patient's breathing level, the problem of changing the lung parenchyma during follow-up, and the qualitative evaluation of the observer.

As a more objective method, a method of extracting texture or shape characteristics of a CT image using a computer program has been introduced, and Chabat et al. Have attempted to differentially diagnose obstructive pulmonary disease. However, this method is applied to obstructive pulmonary disease, and also has a problem in that it does not directly use the shape of the low shade region, which is an important criterion for judging existing diseases, using only the texture of the lung.

It is an object of the present invention to provide a method for automatically classifying lung diseases by automatically classifying lung diseases using texture and morphology analysis of lungs, which are known as clinically important criteria.

The present invention includes the steps of: (a) separating the lung region from the three-dimensional chest volume data obtained through the CT technique, and performing image quality preprocessing for texture and shape analysis of the lung region; (b) comparing and analyzing the texture and shape of the lung region with the texture and shape data of the lung part for each specific disease previously stored; And (c) classifying each part of the lung region by the specific disease according to the comparative analysis.

Here, the method includes performing image quality preprocessing on lung parts representing a specific disease from specific 3D chest volume data, and machine learning the texture and shape data of the lung parts by specific diseases; And (e) generating an automatic classifier that automatically classifies the texture and shape data of the lung by the specific disease through the machine learning. At this time, the steps (b) to (c), by using the automatic classifier, by analyzing the texture and shape of the lung area, it is possible to automatically classify each part of the lung area by the specific disease.

On the other hand, the present invention provides a computer-readable recording medium recording a program for executing the automatic lung disease classification method on a computer.

Pulmonary disease automatic classification method according to the present invention, it is possible to determine the disease severity of the patient through the three-dimensional CT imaging of the lung disease patients. In addition, when the automatic classifier is generated by performing machine learning on the texture and shape of the lung part of each disease, the disease of each part of the entire lung area may be diagnosed and evaluated more reliably.

That is, the automatic classification of lung diseases, by using the local texture and morphological specificity of the lungs, each part of the lung area is automatically classified by disease, quantifying how much part of the lungs are infected with a specific disease It can be helpful to provide the data. These data can be used more importantly than other known indicators that are insensitive to the patient's breathing level.

1 is a flow chart of the automatic lung disease classification method according to an embodiment of the present invention. 2 is another flow chart of FIG. 1, and FIG. 3 is a system diagram for the method of FIG.

First, prior to the description of the automatic lung disease classification method, it will be described with respect to the automatic lung disease classification system 100 for the method as follows. The automatic lung disease classification system 100 includes a photographing unit 110, a display unit 130, an input unit 120, and a control analysis unit 140.

The photographing unit 110 captures 3D chest volume data by CT and acquires photographing information. The control analyzer 140 automatically classifies the lung disease through the texture and shape analysis of the lung area after performing the pretreatment on the lung area from the acquired photographing information. Of course, in addition to the control analysis unit 140 may control the respective components, such as the photographing unit 110, the display unit 130, the input unit 120.

Meanwhile, the input unit 120 may receive various manipulation signals from a user (expert or doctor) and transmit them to the control analysis unit 140. In addition to the photographing information, the display unit 130 may visualize and automatically classify lung disease information that is automatically displayed on the screen. In addition, the control analysis unit 140 processes the signal input to the input unit 120 and transmits a corresponding operation to the display unit 130, so that the corresponding information is displayed on the real-time screen.

Hereinafter, a method for automatically classifying lung diseases according to an embodiment of the present invention will be described in more detail with reference to FIGS. 1 to 3.

First, the control analyzer 140 receives the 3D chest volume data obtained through the CT technique from the photographing unit 110, classifies the lung region from the 3D chest volume data, and the texture of the lung region. And image quality preprocessing for shape analysis (S110).

Here, the control analyzer 140 may recognize the lung region as an object from the 3D chest volume data transmitted from the photographing unit 110, and perform a task of distinguishing the left lung from the right lung. The pre-processing of the image quality includes a process of removing noise, filtering, and increasing a signal-to-noise ratio (SNR) for analyzing each texture and shape of the closed region. According to the above preprocessing process, it may contribute to the improvement of the analysis accuracy of the control analysis unit 140.

The obtained volume data may be volume data photographed from the outside and received through the input unit 130 in addition to the data photographed by the photographing unit 110. On the other hand, the texture and shape analysis will be described later in detail.

After this pretreatment step (S110), the control analysis unit 140 compares and analyzes the texture and shape of the lung region with the texture and shape data of the lung part for each specific disease previously stored (S120). That is, the previously stored texture and shape data of the specific lung part is used as reference data of lung disease analysis for each part of the lung area currently photographed.

Next, the control analysis unit 140 may classify each part of the lung region by the specific disease according to the comparative analysis (S130). That is, the automatic lung disease classification method as described above may automatically classify the lung disease corresponding to each part of the lung region through the texture and shape analysis of the inputted lung image.

Hereinafter, a method of acquiring texture and shape data of a lung part for each specific disease previously stored in the control analyzer 140 and automatic classifier generation using the same will be described in detail with reference to FIG. 1 or 2. do. Of course, the method of acquiring texture and shape data of the lung part for each specific disease may be equally applicable to the image quality pretreatment process for analyzing texture and shape of the corresponding lung area in step S110.

First, the control analysis unit 140 performs image quality preprocessing on a lung part representing a specific disease from one or a plurality of specific 3D chest volume data, thereby machine learning the texture and shape data of the lung part for each specific disease. (S140). Of course, the specific three-dimensional chest volume data may be data photographed by the photographing unit 110 or data separately photographed from the outside and received by the input unit 130.

In addition, the control analysis unit 140 generates an automatic classifier for automatically classifying the texture and shape data of the lung by the specific disease through the machine learning (S150). Accordingly, the control analysis unit 140 after the step S110, by using the generated automatic classifier, by analyzing the texture and shape of the lung area (S120), each part of the lung area to the specific disease Automatic classification by star (S130). The automatic classifier generation process according to steps S140 to S150 may be performed before the step S110, but may be performed at any point before the step S120.

Hereinafter, steps S140 to S150 will be described in more detail. First, the specific 3D chest volume data photographed by the photographing unit 110 or input through the input unit 130 is input to the closed partition module in the control analyzer 140 in the form of a DICOM file. The lung dividing module analyzes a closed image from the input volume data and divides only the closed region (S141).

That is, the lung splitting module may generate a lung mask that is an image having a value of '1' in the lung area and a value of '0' in the non-lung area. Such a closed mask serves to determine a required calculation area in the texture and shape analysis process. Of course, the process such as the lung splitting, the generation of the lung mask to be performed in step S140 corresponds to the pre-processing process of the image, it is also applicable to the step S110.

In addition, the input unit 120 receives a region of interest (ROI), that is, a region of interest, from a specialist, such as a chest radiologist, representing a specific lung disease among the lung regions (S142). The input specific 3D chest volume data, the lung mask, and the ROI are used as input data of machine learning performed by the control analyzer 140. 4 to 6 show examples of an image in which an expert is selected to select a region of interest (ROI) on a variety of specific lung shapes (normal, hepatic vitreous shading, mesh shading, cellulitis, emphysema, and hardening).

Here, the control analyzer 140 performs preprocessing such as noise removal, filtering, and signal-to-noise ratio adjustment on the image of the selected ROI portion (S143). The various texture and shape analysis factors of the corresponding part are calculated with respect to the preprocessed image (S144).

Here, when the shape analysis in step S144, the shape analysis of the low shading region for the lung region is performed. The low shaded area is an important criterion for judging disease. That is, the reliability of lung disease classification may be improved through morphological analysis of the low shaded area. Of course, the morphological analysis of the low shading site should be equally applied to the morphological analysis performed in step S110.

The texture and shape analysis elements will be described with reference to Tables 1 and 2 as follows.

Table 1: Texture Analysis Factors

Table 1 shows an example of 13 representative texture elements as texture analysis elements. Based on this, the control analyzer 140 may calculate the 13 texture elements by performing histogram analysis, gradient image analysis, run length analysis, and coherence matrix analysis on the computed tomography lung images. .

Table 2: Form Analysis Factor

Table 2 shows the quantification of the shape of the low-shaded region through the shape analysis technique. For example, a thresholding technique (-950 HU, threshold) may be applied to a low shade region, which is a major indicator of lung disease, to generate a cluster. In addition, the shape analysis includes a method using a top-hat transform, which is a shape analysis technique, and a method using a number, average area, standard deviation, circularity, etc. of clusters in each ROI. Can be divided into two categories.

 The formula for analyzing the circularity in the cluster analysis is as follows. Within one ROI, each mean and standard deviation of the circularity and the aspect ratio can be calculated for each recognized cluster. For the calculation of the circularity and the aspect ratio, see Equations 1 and 2 below.

[Equation 1]

[Equation 2]

On the other hand, the control analysis unit 140 extracts the texture and shape elements as described above, and then performs the optimization of the texture analysis and shape analysis elements (S145). Thereafter, the control analysis unit 140 machine learns the texture and shape data of the lung part for each specific disease by using the optimized factor (S146).

In addition, the control analysis unit 140 generates an automatic classifier for automatically classifying the texture and shape data of the lung by the specific disease through the machine learning (S150). 7 to 9 show examples of lung images that are automatically classified by the classifier.

Meanwhile, during the optimization process, a prospective feature selection technique may be used to distinguish factors helpful for pulmonary disease classification from a plurality of texture analysis and morphological analysis elements. It learns from one element, then evaluates its accuracy, selects the element with the highest accuracy, and then adds the remaining elements that are not selected on the basis of the selected element, sequentially sequencing the factors that give the best accuracy. This is how you choose. In conjunction with this method, a 5-folding technique is used to avoid selection errors in the training data. When calculating the accuracy of the learning element, it is divided into five random groups, four of which are used in the learning process (S146), and the other one is used in the performance test process (S147), and then automatically generated The accuracy of the classifier S150 is calculated. The final classification result may be generated using the results of the test process S147 and the classifier S150.

Summarizing the above automatic lung disease classification method as follows. First, the lung region is automatically recognized through a randomly input lung image, and preprocessing for texture and shape analysis is performed (S110). Thereafter, using the automatic classifier generated in the step S150, after analyzing the lung disease for all parts of the lung area (S120), the respective parts are classified by a specific disease (S130).

Meanwhile, after the lung disease classification step (S130), the display unit 130 may visualize and express a result of classifying each part of the lung area by the specific disease on a screen (S140). For example, by overlaying the classification result on the existing CT image obtained through the CT technique, it is possible to facilitate the evaluation and diagnosis of the disease.

In addition, the control analysis unit 140 may calculate and quantify the area and location of the specific disease for each part of the lung area. 10 is an exemplary view of an image in which quantitative data is displayed on any lungs in which lung diseases are automatically classified. That is, referring to FIG. 10, the display unit 130 may visualize and display the area and location of the numerical value of the specific disease on the screen.

As described above, the automatic lung disease classification method, it is possible to determine the disease severity of the patient through the three-dimensional CT imaging in the lung disease patients. That is, machine learning is performed on the lung data obtained by 3D computed tomography through texture and shape analysis for specific diseases, and the entire lung can be classified by disease using an automatic classifier obtained by this learning. .

In addition, the method, using the local texture and morphological specificity of the lungs, automatically classifies each part of the lung area by disease to provide data quantifying which part of the lung how infected with a particular disease. It can be helpful. These data may be more important than other known indicators that are insensitive to the patient's breathing level. In addition, the method for automatically classifying lung diseases may be applied to various types of lung diseases, but more particularly, may be applied more effectively during automatic classification of diffuse interstitial lung disease (DILD).

On the other hand, the automatic lung disease classification method as described above, it is possible to implement as a computer-readable code on a computer-readable recording medium. Computer-readable recording media include all types of recording devices on which data is stored as media that can be read and executed by a computer system.

Examples of computer-readable recording media include ROM, RAM, CO-ROM, magnetic tape, floppy disks, optical data storage devices, and the like, which are also implemented in the form of carrier waves (for example, transmission over the Internet). Include. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

As described above, the automatic classification of lung diseases can visually and quantitatively diagnose the severity and the pattern of lung diseases, and can provide a diagnostic aid immediately applicable to actual clinical practice. Furthermore, if the medical form is changed by automating and verifying the qualitative reading method which is currently dependent on the naked eye, it is determined that more quantitative and sophisticated reading is possible.

In addition, if the automatic classification of lung diseases is applied to the existing 3D computed tomography apparatus or picture archive and communication system, it may not only bring a significant amount of software sales and export effect, Furthermore, it will be able to strengthen the competitiveness of domestic medical systems in the global market.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

1 is a flow chart of the automatic lung disease classification method according to an embodiment of the present invention,

2 is another flow chart of FIG. 1;

3 is a system diagram for the method of FIG. 1;

4 to 6 are views illustrating an image in which an expert selects a region of interest (ROI) on various specific lung shapes;

7 to 9 are exemplary diagrams of lung images automatically classified by the classifier;

10 is an exemplary view of an image in which quantitative data is displayed on any lungs in which lung diseases are automatically classified.

<Brief description of symbols for the main parts of the drawings>

100: automatic lung disease classification system 110: filming unit

120: display unit 130: input unit

140: control analysis unit

Claims (6)

(a) classifying lung areas from three-dimensional chest volume data obtained through a CT technique and performing image quality preprocessing to analyze texture and shape of the lung areas; (b) comparing and analyzing the texture and shape of the lung region with the texture and shape data of the lung part for each specific disease previously stored; And and (c) classifying each part of the lung region by the specific disease according to the comparative analysis. The method according to claim 1, (d) performing image quality preprocessing on the lung parts representing the specific disease from the specific three-dimensional chest volume data, and machine learning the texture and shape data of the lung parts for each particular disease; And (e) generating an automatic classifier that automatically classifies the texture and shape data of the lung by the specific disease through the machine learning; Step (b) to step (c), Using the automatic classifier, by analyzing the texture and shape of the lung area, automatically classifying lung diseases for each part of the lung area by the specific disease. The method according to claim 1 or 2, after the step (c), and (f) overlaying and visualizing a result of classifying each part of the lung region by the specific disease on a CT image obtained through the CT technique. The method according to claim 1 or 2, after the step (c), (g) calculating and quantifying the area and location of the specific disease for each part of the lung region; And (H) the lung disease automatic classification method further comprising the step of visualizing the area and location for each particular disease. The method according to claim 1, wherein step (a), Automatic lung disease classification method for performing the morphological analysis of the low shaded area for the lung area. A computer-readable recording medium having recorded thereon a program for executing the method of any one of claims 1 to 5 on a computer.

Images