Research Of Algorithms For Lung Nodules Segmentation Based On Low Dose CT Images

Lung cancer is one of the most deadly diseases in the world, and the incidence rate of lung cancer is very high due to the increase of the number of smokers and the deterioration of the environment. Lung cancer usually occurs in the form of pulmonary nodules in the early stage, so early detection and diagnosis of lung nodules can play an effective role in the prevention of lung cancer. The pulmonary nodules detection is mainly realized by CT imaging equipment, but conventional dose CT has heavy radioactive and is harm to the human body, so in today’s society for clinical application, the low dose CT detection whose radiation is relatively smaller becomes more and more common. In the process of detection, the low dose results in more noise in CT images, low-contrast of the object region and background region, a great influence on the accuracy of pulmonary nodule segmentation. Therefore, the research of lung nodule segmentation algorithm based on low dose CT images has great clinical application value.In order to reducing the interference of the noise and uneven distribution of gray level in the segmentation process based on low-dose chest CT images, this paper has performed some work in the following aspects:1) In the pre-processing, first of all, the low-dose CT image was denoised by using wavelet transform, so that the majority of noise in the image could be remo ved while the details of the image can be preserved, and avoiding the loss of effective information. Then for the denoised image, binary segmentation and morphological method were used to obtain the rough lung parenchyma, and the rolling ball method was applied to fix the boundary, finally completed the lung parenchyma segmentation. The algorithm is simple, the computation is small and the effect is great.2) By researching and analyzing the active contour model algorithms which have been common applied in the field of medical image segmentation, and combining with the characteristics of low dose CT image, LBF model based on local image pixel information was chose to perform lung nodule segmentation.3) To overcome the defects of LBF model, this paper referenced the fuzzy clustering theory that puts forward the fuzzy membership degrees of pixels, in different regions of the image, with respect to the pulmonary nodules image are different, and introduced a fuzzy speed function, which may be excluded from the interference of noise factors such as segmentation, as weight restriction factor into LBF model. However, in the actual clinical practice, the pulmonary nodules are mostly juxta-vascular type, the gray value of blood vessels is close to the gray level of pulmonary nodules in the lung parenchyma. Therefore, the fuzzy membership of the pixels in these two regions is close to the pulmonary nodules’ fuzzy membership, which may lead to incorrect segmentation that the pulmonary nodules and blood vessels are divided into one class when we segment the juxta-vascular nodules. To solve this problem, in the basis of the characteristic value of blood vessels’ region, we improved the fuzzy membership, introduced the vessel characteristic coefficient and reduced the influence of blood vessel regions, thus segmented the blood vessels and pulmonary nodules effectively. By the experiment, compared with the traditional model, the algorithm proposed in this paper can improve the accuracy of the segmentation of pulmonary nodules effectively and works better for juxta-vascular nodules. Besides, it can also reduce the false segmentation rate of pulmonary nodules effectively.