Beta 1


Title Detection of Connective Tissue Disorders from 4D Aortic MR Images Using Independent Component Analysis
Author Hansen, Michael Sass (Image Analysis and Computer Graphics, Informatics and Mathematical Modelling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Supervisor Ersbøll, Bjarne Kjær (Informatics and Mathematical Modelling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Sonka, Milan
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Master's thesis
Year 2006
Abstract The current report concerns methods of early detection of connective tissue disorders leading to aortic aneurysms and dissections. Automated and accurate segmentation of the aorta in 4D (3D + time) MR image data is reviewed, and a computer-aided diagnosis (CAD) method using independent component analysis is reported. This admits the objective identification of subjects with connective tissue disorders from 4D aortic MR images. The majority of the presented work is concentrated on independent component analysis(ICA), estimating sources to be used for the diagnosis task. Prior knowledge of the source distribution is utilized using an ordering of the components. Two new ordering measures are introduced in current work. A novel approach to constrained dimensionality reduction in ICA is developed. A new idea of time-invariant independent components is introduced, and assists in the disease detection in the presence of sparse data. 4D MR image data sets acquired from 21 normal and 10 diseased subjects are used to evaluate the efficiency of the method. The automated 4D segmentation result produces accurate aortic surfaces. The ICA results are validated by a leave-one-out classification test, and are further substantiated by visual inspection of the components. Using a single phase of the cardiac cycle, 8 out of 10 diseased subjects are identified and the specificity is 100 \$\backslash\$%, classifying all 21 healthy subjects correctly. These results are obtained using components showing correspondence to clinical observations. With 4D information included, the CAD method classifies 9 out of 10 diseased correctly, and still the specificity is 100 \$\backslash\$%.
Imprint Informatics and Mathematical Modelling, Technical University of Denmark, DTU : DK-2800 Kgs. Lyngby, Denmark
Pages 107
Fulltext
Original PDF imm4820.pdf (6.01 MB)
Admin Creation date: 2006-10-06    Update date: 2012-12-19    Source: dtu    ID: 191632    Original MXD