Research On And Implementation Of Lung CT Image Segmentation Method Based On Level Set

Medical image segmentation is a pre-processing technology for other medical image processing and pattern recognition problems,such as feature quantification,feature alignment and 3D reconstruction,which provides a strong support for the clinical diagnosis and aided therapy.Lung CT image segmentation is the most core step for lung diseases when conducting computer-aided diagnosis as well as a key issue affecting the diagnostic process automation and accuracy of the diagnostic results.Based on level set method,image segmentation technology is a hot research topic in the field of image segmentation.Compared with other segmentation methods,it has the advantages of adaptation to topology changes,easy implementation of numerical value and easy expansion to high dimension.The core idea of level set is to regard the active contour curve in the segmented images as the zero level set of level set function in the high dimension,so the curve is also evolved to a high dimension.However,it still has some deficiencies,for example,to get more stable and accurate results,the level set function needs to regularly reinitialize the level set function in the evolution process,so as to minimize the miscalculation caused by degradation.Re-initialization may not only falsely cause zero level set to be away from the expected position,but also waste unnecessary time.By analyzing the level set segmentation method based on the distance regularization,this thesis modifies the border effect issues existing in the distance regularization of this method.Since the classic Chan-Vese model is unsuitable for medical image segmentation,this thesis introduces the method of offset field correction and image segmentation,solves the wrong segmentation and difficult segmentation caused by intensity homogeneity in medical images.This thesis improves by summarizing the above two points,applies them to specific lung CT image segmentation,and it is proved that the proposed method can better extract the lung boundary and have significantly improved relative to Chan-Vese model.This thesis first introduces Chan-Vese modeling,solution and implementation based on level set method,elaborates the improvements of non-applicability to CT images and necessity of re-initialization,conducts many simulation experiments over the proposed improved algorithms,and verifies its accuracy and high efficiency in processing the lung CT images.On this basis,the thesis puts forward the continuous segmentation algorithm based on above improved segmentation algorithm,applies segmentation results to 3D reconstruction of the lungs,and thus provides strong supports for lung spatial information as well as its clinical diagnosis and adjuvant therapy of doctors.