Segmentation Of Pulmonary Nodules Based On Fuzzy C-Means Cluster And Dictionary Learning Algorithm

The process of industrialization is speeding up increasingly in the present world, which has brought great changes to our material life. Meanwhile, it also causes the degradation of ecosystem and thus extends the increasing pollution problem of environment. Haze, PM2.5, water eutrophication, these problems appears in our life frequently, which harms human’s healthy constantly and leads the lung cancer becoming the culprit cause of cancer death. We need to seek new treatment for lung cancer urgently. The main manifestation of I lung cancer is lung nodules, and the best treatment is early detection and treatment. At present, the development of computer aided diagnosis of lung cancer more and more quickly, with the medical imaging, machine learning, and digital image processing technology constantly break technology barriers. While the Accurate segmentation result is the basis for accurate diagnosis of CAD system. The main means of lung image acquisition is CT(Computed tomography, CT). The noise and large amount of image data cannot be avoided due to the imaging principle of CT image, and without enough researches on the research of ground glass opacity(GGO) pulmonary nodules and juxta-vascular(JV) pulmonary nodules. So noise image, GGO and JV pulmonary nodules are research objects in this paper. Recently, the fuzzy C-means algorithm(FCM) has been extensively applied to image segmentation. But it still very sensitive to noise and easy to produce the over segmentation problem when applied to the pulmonary nodule segmentation.In order to segment more accurate pulmonary nodules from the image, this paper makes a deep research on the FCM algorithm, and proposes two improved algorithms to adapt to pulmonary nodule segmentation problem. The main contents and innovation are as follows:1. Research on resist noise segmentation method based on FCM algorithm, which mainly solve the lung CT images that contain noise, uneven lighting and so on. This article analyzes from the neighborhood of the target pixel information on the pixel classification impact point of view, and holds the view that the segmentation effect from the existing reference neighborhood pixel information is not good because the unsatisfactory effect of the segmentation of the CT image which contains noise. The information of the neighborhood pixels does not necessarily have a positive correlation to the central pixel. Base on this reason, this article proposes an efficient reference neighborhood pixels windows mechanism, redefines the gray level of the pixels in the neighborhood windows, different fuzzy factors are selected according to the reference results. This method can solve the traditional segmentation method of image noise sensitive and image segmentation problem of FLICM algorithm. It can improve the accuracy of pulmonary nodule segmentation.2. Based on FCM algorithm and dictionary learning algorithm for the segmentation of lung nodules, According to the improved FCM algorithm segmentation effect is unsatisfactory, there still have room to improve. This paper makes a deeper analysis on the internal structure information of the pixels, and points out the local information in the image still has a great influence on the classification of the target pixel. Based on the improved FCM algorithm, this paper puts forward a further improvement on the objective function based on the fusion of dictionary learning. By using the process of dictionary learning to merge the local information of the image, obtaining the intrinsic correlation between pixels, at the same time, the optimization method for solving the four unknown variables under the new objective function is presented. The experimental results show that the new algorithm can improve the accuracy of pulmonary nodules segmentation.