Multidimensional image segmentation and pulmonary lymph-node analysis.

摘要

Lung cancer remains the leading cause of cancer death in the United States and worldwide. Central to the lung cancer diagnosis and staging process is the assessment of the central chest lymph nodes. This assessment typically requires two major stages: (1) location of the lymph nodes in a three-dimensional (3D) high-resolution volumetric multi-detector computed-tomography (MDCT) chest image (2) subsequent nodal sampling using trans-bronchial needle aspiration (TBNA).Lymph-node segmentation is a preliminary but vital step for lymph-node-related process. However, segmentation of regions of interest (ROIs), such as lymph nodes and suspect cancer nodules, is often difficult because of the complexity of the phenomena that give rise to them. Manual slice tracing has been used widely for years for such problems, because it is easy to implement and guaranteed to work. But manual slice tracing is extremely time consuming, subject to operator biases, and does not enable reproducible results. Numerous automated 3D image-segmentation methods have also been developed. But automatic segmentation is generally strongly application dependent, and even the most robust methods have difficulty in defining complex anatomical ROIs. To address these issues, a semi-automatic interactive paradigm, referred to as "live wire," has been proposed by researchers. In live-wire segmentation, the human operator interactively defines an ROI's boundary, guided by an active automated method. 2D and 3D live-wire methods, which improve upon previously proposed techniques, are discussed in this dissertation. The 2D method incorporates an improved cost function to increase robustness and a search region to improve computational efficiency. For the new 3D method, the operator need only consider a few 2D image sections to begin with and then an automated procedure defines the remainder of a 3D ROI's boundary. Furthermore, a computer-based tool incorporating the methods has been built for 3D MDCT-based planning and follow-on live guidance of bronchoscopy. The experimental results and the clinical applications clearly show the robustness and efficiency of the proposed live-wire methods.To facilitate further lymph-node-related MDCT image analysis, surgery planning, and lymph-node sampling, two paradigms have been established: (1) the Mountain system gives the nominal anatomical locations of pulmonary lymph nodes and (2)Wang's bronchoscopy-based map of possible biopsy sites. Both the Mountains and Wang systems were well established internationally for use in mediastinal lymph-node staging and play critical roles in the clinical studies of pulmonary disease. However, little work has been done on CT- based lymph-node analysis and in relating to these two systems. In this dissertation, a computer-based system is presented for automatic definition of lymph-node stations, nodal station visualization and interaction, and lymph-node detection, classification, and segmentation. This system connects anatomical definitions of the lymph-node stations, based on the Mountain and Wang systems, with MDCT chest data. The defined nodal stations can then be used to guide the user into the 3D locations where lymph nodes are expected to be found. Supplemented with the robust live-wire-based semi-automatic segmentation tools and other utilities, this computer-based system would conceivably speed up lymph-node detection, segmentation, and classification by avoiding unnecessary 2D-slice navigation and enabling the user to concentrate on specific stations. In addition, prior to this dissertation a link between the Mountain and Wang systems has never been formulated. Such a link can strengthen the utility of both systems. The anatomy-based Mountain system is better for CT-only study, while the airway-based Wang system is better for bronchoscopy. By linking these two systems, one can exploit their strengths better and also more fully use the available CT and bronchoscopic video data during live TBNA. Results derived from a set of human 3D MDCT chest images illustrate the usage and efficacy of the proposed system, and show its potential to decrease the examination time of a patient's MDCT scan and facilitate treatment planning.