AUTOMATIC ANALYSIS OF 3-D AND 4-D RADIOLOGICAL IMAGES

点击翻译按钮获取中文摘要

DESCRIPTION: (Adapted from Applicant's Abstract) Many radiological imaging modalities now exist that generate three-dimensional (3-D) and four dimensional (4-D) images of the anatomy. Unfortunately, techniques for analyzing and making measurements on such data sets remain cumbersome. Simplying the analysis of 3-D/4-D images is critical to being able to assess the long-term utility of 3-D/4-D imaging modalities. This proposal addresses this general issue. It is believed that an automated analysis system, assisted by operator-supplied cues, can more thoroughly, more accurately, and more efficiently extract anatomical information from a 3-D/4-D radiological image than purely manual techniques. The applicants propose to devise an interactive automated 3-D/4-D radiological image analysis system by which the operator can define many types of iconic and symbolic problem cues through a flexible display interface with the cues going into an object-oriented model, the interactive dynamic scene (IDS) model. Drawing on the problem information stored in the IDS model, an automatic scene-analysis (ASA) engine extracts the desired anatomical regions. The ASA engine uses a multi-stage analysis procedure that exploits organ morphological, geometrical (shape-based), and functional characteristics; it also uses global topological (organ-to-organ) constraints. The early emphasis of this project is on devising the algorithms for the ASA engine and IDS model. As methods mature, the applicants propose to construct a working prototype of the system. Validation would be done throughout and would focus on four applications: (1) extraction of the endocardial and epicardial borders from 3-D/4-D X-ray computed tomography (CT) images; (2) analysis of true 3-D angiograms; (3) extraction and analysis of the upper airway in 3-D Magnetic Resonance Imaging (MRI) images; and (4) analysis of congenital heart defects in 3-D MRI images. While algorithms are being devised, validation tests, involving a small subset of data, would be done to establish the algorithm functionality. Later, as algorithms mature, detailed tests, using larger sequences of data, would be done to quantitatively compare automatically generated results to manually generated results.