3D IMAGE ANALYSIS FOR DEFINITION OF THE CHEST VASCULATURE

Open Access
Author:
Taeprasartsit, Pinyo
Graduate Program:
Computer Science and Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
November 18, 2010
Committee Members:
  • William Evan Higgins, Dissertation Advisor
  • Henry Joseph Sommer Iii, Committee Member
  • Robert T Collins, Committee Member
  • Jesse Louis Barlow, Committee Member
  • William Evan Higgins, Committee Chair
Keywords:
  • deformable model
  • pulmonary artery
  • aorta
  • vasculature
  • blood vessel
  • segmentation
  • image processing
  • medical imaging
Abstract:
Robust and accurate segmentation of major chest blood vessels from multi-detector computed-tomography (MDCT) chest scans is vital for many pulmonary-imaging applications. Segmented vessels are particularly important for safe, effective image-based bronchoscopic planning and guidance during which a biopsy needle must sample a suspect cancer region while avoiding nearby blood vessels. Therefore, information about blood vessels is critical, as it can be utilized to find optimal biopsy-needle poses. Moreover, segmented vessels can be visualized with other anatomical structures, such as the airway tree and lymph nodes, so that the physician gains better insight into a patient's case. Segmented vessels also provide anatomical landmarks important for identification of central-chest lymph nodes. Unfortunately, existing segmentation methods are inadequate for comprehensive segmentation of major central-chest vessels. This thesis proposes a set of methods for segmentation of major vessels in the central chest and lungs. These vessels are the aorta, pulmonary artery, pulmonary vein, superior vena cava, left/right innominate veins, azygos vein, brachiocephalic trunk, left common carotid artery, left subclavian artery, and peripheral lung vessels. The method set consists of three parts: (1) automatic segmentation of the aorta and pulmonary artery, (2) semi-automatic segmentation of other central chest vessels, and (3) automatic segmentation of peripheral lung vessels. The methods in (1) locate vessel landmarks using deformable models and a model-selection technique. The methods employ these landmarks to extract a medial axis of a vessel and recover a vessel region. The semi-automatic method of (2) requires the user to provide landmarks along the medial axis of a target vessel. The segmentation of peripheral lung vessels (3) searches the entire lung volume for likely vessel cross-sections. The method then forms vessel branches and lung vascular trees that have root vessels in the central chest. To be useful in practice, this thesis also presents a working software system that allows the user to conveniently execute the methods and obtain results that are ready to use by other software tools used in the clinical workflow. In addition, this thesis gives a preliminary study of using PET-CT data to enhance a lung-cancer diagnosis and bronchoscopy procedure planning. Also, we validate the efficacy of the vessel-segmentation methods using ground-truth studies. These studies demonstrate the strong potential of our methods for enhancing the ability to plan and guide a safe bronchoscopic procedure.