Research Of Medical Image Processing Based On SHAPELETS And EM Algorithm

Lung nodules and lung parenchyma are both important structure in lung. They have an important role in lung structure and lung disease diagnosis. Therefore, fast and accurate lung nodules detection and the lung parenchyma extraction have important clinical value. After summarizing and analyzing the present medical image segmentation methods, we have a deep research in detection based on Shapelets and segmentation based on EM algorithm proposed the corresponding improved algorithm for the problems encounterd in detection. Our work are as follows:(1) Low-dose lung CT images, have a wide dynamic range, rich detail, poor contrast characteristics , are very different from the normal image. This paper researched the image characteristics of lung CT image and Shapelets one-dimensional and two-dimensional basic functions, and we propose a LDCT image segmentation algorithm based on Shapelets. This paper made an analysis on Shapelets algorithm in Matlab 6.0 platform, and we determine the characteristic scale by selecting the appropriate parameters, and we calculate the two-dimensional Shapelets coefficient, we detected lung nodules with this method.(2) This paper presents a lung parenchyma extraction method based on the EM algorithm in LDCT. In our paper,a Gaussian model is founded. Assume that the pixs in the image meet the Gaussian mixture distribution. With the help of EM algorithm, the parameters(E step)are initialized,and the parameters maximization(M step) are getted to achieve the clustering of image pixels. The image pixels are divided into several categories,thus lung parenchyma is extracted with EM. I try my experiments in the Matlab platform, and the results show that the algorithm has a good effect.