Objective

2

Computer-aided diagnosis of lung cancer is the automatic recognition of lung lesions in computed tomography (CT) images through a computer image processing technology. Given that lung lesions are located in the lung areas

the contours of lung areas should be automatically marked on lung images first. Thus

lung segmentation is the first step in the computer-aided diagnosis of lung cancer. Current segmentation methods for pulmonary parenchyma

including normal and pathological

can be divided into three categories according to visual appearance

location relationship of shape

and whether or not the method is a hybrid. Methods based on visual appearance mostly consider the lower gray values of lung areas instead of surrounding tissues. Therefore

a pathological lung (especially including interstitial lung disease) is segmented by a classifier on the basis of gray and texture features. These methods mainly introduce the anatomical relationship of the lungs with the heart

liver

spleen

and rib. Hybrid methods establishes shape models and evolves lung contours according to surface features. They can achieve good segmentation results without noise samples in training sets but does not consider the problem of noise samples in the training sets during the construction of training shape models. The active shape model (ASM) based on principal component analysis (PCA) is an outstanding method

which can only deal with Gaussian noise. ASM is a statistical deformation model and uses PCA to obtain the average shape and the allowable range of shape of a training sample set and then establishes a shape model. ASM establishes a local feature appearance model according to the training sample set. Finally

ASM evolves the contour of the test image on the basis of the local feature appearance and shape models. Robust PCA (RPCA) in low rank theory is robust to any kind of noise. In addition

the shapes of a training sample set have a low rank attribute because the anatomical shapes of the human lungs are roughly the same. Therefore

this study combined RPCA with ASM to solve the problem that the correct shape model cannot be obtained because of noise samples in training sets.

Method

2

First

a traditional method was used to segment the pulmonary region of training samples and generate lung boundary contours. The segmentation steps include extra-pulmonary region exclusion by iteration threshold and region growing segmentation methods

trachea elimination

internal cavity filling

liver elimination

and left and right lung separation. Second

we extracted 96 marker points on the left or right lung boundary contour. We assigned these marker points as

$$x $$

-coordinates and

$$y $$

-coordinates and placed them into a single matrix. Then

we aligned the lung boundary contours in the training sample set through the Procrustes analysis method. Third

we decomposed the matrix through the low rank theory into the low rank component and noise component matrices. We used the accelerated approximation gradient method to solve the decomposition and employed PCA to the low rank component matrix and obtained the eigenvalues and eigenvectors. They were used to construct the shape model

which is expressed as the average shape and deformation range. Then

we established a local feature appearance model by using gradient features and obtained a gradient-level appearance model that describes the typical image structure around each landmark from pixel profiles sampled (using linear interpolation) around each landmark perpendicular to the contour. Finally

ASM evolved the contour of the test image based on the local feature appearance and shape models. In the training of the local gradient model in ASM

the noise samples of the training set must be eliminated. In this study

the similarity of the curvature histograms of the marked points on the contours of the training samples was evaluated

and noise samples were removed.

Result

2

First

a training dataset was established. It consisted of 35 multi-slice spiral CT lung scan images collected from the First Affiliated Hospital of Guangzhou Medical University. Two scans included juxta-pleural tumor on the left side. The test dataset comprised 22 scans from EMPIRE 10 and is the gold standard for lung contours.The five scans included juxta-pleural tumor on the left side. Using the proposed method to segment the 22 lungs of EMPIRE10 database

we obtained the overlap rate with a gold standard of 94.5%. Meanwhile

the accuracy of ASM based on PCA was only 69.5%. The shape models generated by RPCA-based ASM and traditional ASM (i.e.

PCA-based ASM) were compared when the training set contained noise samples. Notably

the shape model generated by ASM based on RPCA was consistent with the shape model generated by the traditional ASM when the training set was noiseless. Remarkably

RPCA eliminated the influence of the noise samples on the shape model. Adding PRCA to other lung parenchyma segmentation algorithms generated good results for training sets containing noise samples. The gray threshold method and ASM segmentation method based on RPCA were used to segment the same lung image with juxta-pleural tumor.Notably

the ASM segmentation method based on RPCA generated correct segmentation results in contrast to the gray threshold method.

Conclusion

2

The experimental results showed that a good segmentation result was obtained by the ASM based on RPCA when the noise sample was contained in the training sample set. When the pulmonary parenchyma was absent

some markers were updated to the transition region of the normal lung tissues and tumor rather than to the real lung border during searching.Thus

identifying abnormal markers first is necessary. A new abnormal marker identification method should be designed to provide different searching functions for normal and abnormal markers.