Text

ETD_1207

http://hdl.rutgers.edu/1782.2/rucore10001600001.ETD.000050484

theses

Diagnosis of lung nodules and cancers is a critical and urgent problem in clinical diagnosis. This thesis is to design and build a computer aided lung ground glass opacity (GGO) nodules and large lung cancers diagnosis system which aims to quantify the volumetric change of the lung GGO nodules and large lung cancers between the pre-treatment and post-treatment. In order to quantify the volumetric change of the lung nodules and cancers over time, we need segmentation and registration methods to determine the same lung nodule or cancer between the pre-treatment and posttreatment, as well as lung nodules and cancers detection and segmentation methods. For the registration method, segmented pulmonary tubular objects will act as landmarks. We extract the centerlines of 3D tubular objects using improved ridge-based methods for tubular objects segmentation with fully automatic detection of bifurcation points. The detection of bifurcation points ensures the continuity of the centerlines of the tubular objects. Since medical images contain anatomical structures of various shapes, we first perform a pre-selection method to identify the region containing the tubular objects and extract the centerlines of tubular objects by applying intensity ridge tracing method. These steps are based on the eigenanalysis of the Hessian matrix, which provides an estimation of the elongated direction of tubular objects as well as cross-sectional planes orthogonal to tubular objects. While tracing tubular objects, bifurcation points are automatically detected from the cross-sectional planes by applying
scan-conversion method or Adaboost algorithm with specially designed steerable filters.
For the registration method, we develop a 3D-3D model based rigid registration method based on bifurcation points. We first perform the 3D tubular objects segmentation method to extract the centerlines of tubular organs and radius estimation in both planning and respiration-correlated CT (RCCT) images. This segmentation method automatically detects the bifurcation points by applying Adaboost algorithm with specially designed filters. We then apply a rigid registration method which minimizes the least square error of the corresponding bifurcation points between the planning CT images and the respiration-correlated CT images.
For the lung GGO nodules and large lung cancers detection and segmentation, we propose a novel method to automatically detect and segment lung GGO nodules and large lung cancers from chest CT images. For lung GGO nodules detection, we develop a classifier by boosting k-Nearest Neighbor, whose distance measure is the Euclidean distance between the nonparametric density estimates of two regions. We then apply a clustering method to detect the regions of the lung GGO nodules. The detected regions of lung GGO nodules are then automatically segmented by analyzing the 3D texture likelihood map of the region. We also present the statistical validation of the proposed classifier for automatic lung GGO nodules (10 datasets contains 10 GGO nodules) detection as well as the very promising results of automatic lung GGO nodules segmentation. The methods for the detection and segmentation of large lung cancers are similar to the method above. The improvement is that we propose a robust active shape model method for automatic segmentation of lung areas which can be distorted by large lung cancers. We present the statistical validation of the proposed classifier for large lung cancers (10 datasets contains 16 large lung cancers) detection as well as the very promising results of automatic large lung cancers segmentation. The proposed method provides a new powerful tool for automatic detection as well as accurate and reproducible segmentation of lung GGO nodules and large lung cancers.

Ph.D.

Includes bibliographical references (p. 85-99)

by Jinghao Zhou

doi:10.7282/T3ZP46D2

ETD doctoral

The author owns the copyright to this work.

ETD

application/pdf

application/x-tar

2334720

0535e1973e12c15831d5374fb654cb3261cd558a