CN106600584A - Tsallis entropy selection-based suspected pulmonary nodule detection method - Google Patents

Abstract

The present invention is a suspected pulmonary nodule detection method based on Tsallis entropy selection, which screens out suspected nodules from segmented candidate nodules by calculating the Tsallis entropy value, and then constructs a multi-scale feature extraction mask. Then feature extraction, and put the extracted features into the classifier, and finally get the result. The present invention introduces Tsallis entropy, and proposes a new method for screening candidate nodules by comparing the Tsallis entropy value of candidate nodules with a given empirical threshold T, which can better reduce the false positive rate of nodule detection in the entire system, The new method can assist doctors in the diagnosis of pulmonary nodules more conveniently and quickly.

Description

A Suspected Pulmonary Nodule Detection Method Based on Tsallis Entropy Selection

technical field

The invention belongs to the field of computer-aided diagnosis, and more specifically relates to a method for detecting suspected pulmonary nodules based on Tsallis entropy selection.

Background technique

There are 3.12 million new cases of cancer in China every year, and more than 2 million deaths due to cancer every year, among which the most deadly cancer is lung cancer. The cure rate of lung cancer is closely related to the clinical stage at the time of diagnosis. The 5-year survival rate of patients with early stage lung cancer is over 90%, the survival rate of patients with stage I lung cancer is reduced to 60%, and the annual survival rate of patients with stage II to IV lung cancer is 90%. rate from 40% to 5%. Therefore, "early detection, early diagnosis, and early treatment" is the key to improving the survival rate of lung cancer patients. But it is not optimistic that only 15% of lung cancers are detected in the early stage. However, when there are long-term unhealed symptoms such as cough and hemoptysis, and then go to the hospital for an examination, it is often already in the advanced stage of lung cancer. Therefore, how to achieve "early detection" and "early diagnosis" is an important topic that needs to be studied.

A CADe system for detecting pulmonary nodules usually consists of five subsystems: acquisition, preprocessing, segmentation, nodule detection, and false positive reduction. Acquisition is mainly to collect medical images. The main preprocessing techniques are: median filter, enhancement filter, contrast-limited adaptive histogram equalization, automatic enhancement, Wiener filter, fast Fourier transform, wavelet transform, anti-geometric diffusion, erosion filter, smoothing filter and noise correction etc.

In recent decades, although there are various types of related algorithms in the field of image segmentation, they still cannot fully meet people's actual needs. The reasons are quite complex, including: the mathematical model cannot be used to simply describe the practical problems faced by people; the structural properties of segmentation objects vary widely; image degradation and people's expectations for segmentation results are different, etc. These reasons determine that it is impossible to achieve a universal and general segmentation method. Only reasonable choices can be made for specific problems and specific needs, and balance or focus on key indicators such as accuracy, speed, and robustness. Two main methods for segmenting lung images are: threshold-based segmentation and deformable model segmentation. In threshold-based segmentation, a brightness threshold performs the separation operation. The main types of deformable models used to segment lung images are: active contours and level set based deformable models. Other techniques for segmenting pulmonary nodules include: cylindrical and spherical filters, morphological operations, thresholding, multiple gray-level thresholding, and connected element analysis. The main extracted features are: grayscale features, morphological features, texture features, context features, and external features. The main classifiers are: linear discriminant analysis, rule-based, clustering, Markov random field, artificial neural network (ANN), support vector machine (SVM), double threshold cut.

Contents of the invention

The purpose of the present invention is to design a method for screening suspected pulmonary nodules using Tsallis entropy calculations, pre-select candidate nodules before classification, and then put representative features into the classifier to improve computing speed and reduce false positive rate.

To achieve the above object, the present invention provides a method for detecting suspected pulmonary nodules based on Tsallis entropy selection. It mainly includes the following content: Introduce the concept of Tsallis entropy in the nodule mask of feature extraction, and distinguish pulmonary nodules from blood vessels and bronchi perpendicular to the slice through the Tsallis entropy value, effectively reducing the number of ROIs, and improving lung nodules The detection of suspected pulmonary nodules before the detection can effectively reduce the false positive of the experiment and improve the calculation speed. The technical principle is shown in Figure 1, and the specific technical process is as follows:

Step 1: Acquisition of pictures, it is planned to use the public database LIDC, and use median filtering to preprocess the pictures;

Step 2: First, use the threshold value to perform preliminary segmentation on the original image, and then use the elements of the disc structure to perform two-dimensional morphological opening operation on the roughly segmented lungs, and then use the three-dimensional region growing algorithm to segment the lung area in the image, The segmented part is the lung area, then the hole filling algorithm is used to repair the hole, and finally the 3D morphological closing operation is used to refine the edge of the lung;

Step 3: First, use a series of thresholds to threshold the original image, then perform a logical AND operation on the segmentation results and the lung mask, and then use a series of disc structural elements to perform two-dimensional morphology on the segmented nodules Learn to open the operation, and finally merge all the intermediate nodule masks to form the final nodule mask;

Step 4: Perform on suspected nodules. Select the threshold T, and calculate the Tsallis entropy value of the nodule. When the Tsallis entropy value is greater than T, it is an obvious false positive, and it will be eliminated, and the remaining nodules will be used as candidate nodules.

Step five, feature extraction, intends to use some features in the traditional sense and extract based on three-dimensional blocks of different scales. There are three main types of features: geometric, brightness, and gradient features, all of which include two-dimensional and three-dimensional features. Finally, use SFS to optimize a series of features;

Step 6: Classify the nodules with the SVM classifier.

Description of drawings

FIG. 1 is a schematic block diagram of a method for detecting suspected pulmonary nodules based on Tsallis entropy selection in the present invention.

Fig. 2 is a technical scheme diagram of a method for detecting suspected pulmonary nodules based on Tsallis entropy selection in the present invention.

Specific implementation

Specific embodiments of the present invention will be described below in conjunction with the accompanying drawings, so that those skilled in the art can better understand the present invention. It should be noted that in the following description, detailed descriptions of known functions and designs that may dilute the main content of the present invention will be omitted.

In this embodiment, a suspected pulmonary nodule detection method based on Tsallis entropy selection of the present invention mainly includes the following links: 1. Image acquisition, 2. Lung segmentation, 3. Nodule segmentation, 4. Tsallis entropy selection, 5. Multi-scale feature extraction, 6. Classify suspected nodules.

The acquisition of the pictures is planned to use the public database LIDC, and use the median filter to preprocess the images so that they can be directly processed by matlab.

For the segmentation of the lungs, the threshold method and morphologically related content are mainly used for segmentation. Firstly, two thresholds are used to initially segment the original image, and secondly, the elements of the disk structure are used to perform two-dimensional morphological opening operation on the roughly segmented lungs, and then the three-dimensional region growing algorithm is used to segment the lung area. Part of it is the lung area. A hole-filling algorithm is then used to patch these holes, and finally a 3D morphological closing operation is used to refine the edges of the lungs.

The segmentation of pulmonary nodules uses a morphological method for segmentation. First, a series of thresholds are used to perform threshold segmentation on the original image, and then the segmentation results and the lung mask are logically ANDed, and then a series of discs are used to The elements of the structure perform the opening operation on the segmented nodules, and finally merge all the intermediate nodule masks to reconstruct the nodule mask.

Select the threshold T to calculate the Tsallis entropy value of the candidate nodules. When the Tsallis entropy value is greater than T, it is an obvious false positive, and it is eliminated. The nodules with entropy values greater than T are retained as suspected nodules.

Feature extraction intends to use some features in the traditional sense. There are three main types of features: geometric, brightness, and gradient features. First, a bounding box is constructed for the segmented nodules, and then 3 pixels are expanded on the basis of the bounding box, so that most of the nodules that are not accurately segmented can be included. Since the precise segmentation of nodules has certain limitations, feature extraction is not performed on the segmented nodule mask, but feature extraction is performed on the expanded 3D block, so that most of the nodules can be included. All these features include two-dimensional and three-dimensional, in order to be able to fully represent nodules. Finally, feature selection, that is, a dimensionality reduction algorithm, is used to optimize a series of features, and the forward algorithm SFS is proposed to be used to optimize features.

Classifier classification, put the features extracted in the above steps into the SVM classifier to perform binary classification on all nodules, so as to obtain the final result.

A kind of suspected pulmonary nodule feature extraction method based on Tsallis entropy selection of the present invention has the following characteristics:

The present invention proposes a new method for screening candidate nodules, which can better reduce the false positive rate of nodule detection in the entire system, and the new method can assist doctors in diagnosing pulmonary nodules more conveniently and quickly.

Although the illustrative specific embodiments of the present invention have been described above, so that those skilled in the art can understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, As long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are included in the protection list.

Claims (2)

1. A method for detecting suspected pulmonary nodules based on Tsallis entropy selection of the present invention mainly includes the following contents: Introduce the concept of Tsallis entropy in the nodule mask of feature extraction, and distinguish pulmonary nodules and vertical slices by Tsallis entropy value Blood vessels, bronchi and other tissues can effectively reduce the number of ROIs, and the detection of suspected pulmonary nodules before the detection of pulmonary nodules can effectively reduce the false positives of the experiment and improve the calculation speed. The technical solution is as follows: Step 1: Acquisition of pictures, it is planned to use the public database LIDC, and use median filtering to preprocess the pictures; Step 2: First, use the threshold value to perform preliminary segmentation on the original image, and then use the elements of the disc structure to perform two-dimensional morphological opening operation on the roughly segmented lungs, and then use the three-dimensional region growing algorithm to segment the lung area in the image, The segmented part is the lung area, then the hole filling algorithm is used to repair the hole, and finally the 3D morphological closing operation is used to refine the edge of the lung; Step 3: First, use a series of thresholds to threshold the original image, then perform a logical AND operation on the segmentation results and the lung mask, and then use a series of disc structural elements to perform two-dimensional morphology on the segmented nodules Learn to open the operation, and finally merge all the intermediate nodule masks to form the final nodule mask; Step 4: Select the threshold T, calculate the Tsallis entropy value of the candidate nodules, when the Tsallis entropy value is greater than T, it is an obvious false positive, remove it, and take the remaining nodules as suspected nodules. Step five, feature extraction, intends to use some features in the traditional sense and extract based on three-dimensional blocks of different scales. There are three main types of features: geometric, grayscale, and gradient features, all of which include two-dimensional and three-dimensional features. Finally, use SFS to optimize a series of features; Step 6: Classify the nodules with the SVM classifier. 2. A kind of suspected pulmonary nodule detection method based on Tsallis entropy selection according to claim 1, it is characterized in that the concept of Tsallis is introduced before the nodule feature extraction, distinguishes pulmonary nodules and vertical slices by Tsallis entropy value Blood vessels, bronchi and other tissues to screen candidate nodules. The main feature of the present invention is that the concept of Tsallis entropy is introduced during feature extraction. Since pulmonary nodules and blood vessels are very similar in size and shape, there is no big difference between the two in gray mean value, but in Tsallis entropy, there is There is a relatively obvious difference, so that the suspected nodules are first selected before the final classification. It mainly includes: (1) For the segmented nodules, the Tsallis entropy value is used to distinguish blood vessels and pulmonary nodules, so as to screen out candidate nodules, and then construct a multi-scale feature extraction mask. (2) Finally, use the svm classifier to classify the suspected nodules to obtain the experimental results. First, calculate the Tsallis entropy value for the segmented nodules. Tsallis entropy is a generalized form of Shannon entropy, Tsallis entropy is a mechanism based on non-extensive statistics produced by Boltzmann-Gibbs theory. In the image threshold selection, many images have non-additive information, so using the additive Shannon entropy as the evaluation criterion cannot obtain ideal segmentation results. Its formula is as follows: $\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; )</span>& <span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\end{array}$ In the formula: P＝{p ₁ ,p ₂ ,…，p _N } is any probability distribution and satisfies p _i ≥0, q is the inextensibility coefficient, which is used to describe the inextensibility of Tsallis entropy, that is, for a system that can be decomposed into two subsystems A and B with independent statistics, its Tsallis entropy can be expressed as: S _q (A+B)＝S _q (A)+S _q (B)+(1-q)S _q (A)S _q (B) Tsallis entropy further considers the relationship between the two subsystems. In image threshold selection, many images contain non-additive information. Therefore, using additive Shannon entropy as the evaluation criterion cannot obtain ideal screening results. . The Tsallis entropy of the obtained candidate nodules is compared with the given empirical threshold T, and only the nodules whose Tsallis entropy is smaller than the threshold T are retained as suspected nodules, and then feature extraction is performed on the screened suspected nodules. Put the features extracted in the above steps into the SVM classifier to perform binary classification on all nodules, and finally get the experimental results.