Generative Interpretation of Medical Images

Mikkel B. Stegmann

AbstractThis thesis describes, proposes and evaluates methods for automated analysis and quantification of medical images. A common theme is the usage of generative methods, which draw inference from unknown images by synthesising new images having shape, pose and appearance similar to the analysed images. The theoretical framework for fulfilling these goals is based on the class of Active Appearance Models, which has been explored and extended in case studies involving cardiac and brain magnetic resonance images (MRI), and chest radiographs.

Topics treated include model truncation, model compression using wavelets, handling of non-Gaussian variation by means of cluster analysis, correction of respiratory noise in cardiac MRI, and the extensions to multi-slice two-dimensional time-series and bi-temporal three-dimensional models.

The medical applications include automated estimation of: left ventricular ejection fraction from 4D cardiac cine MRI, myocardial perfusion in bolus passage cardiac perfusion MRI, corpus callosum shape and area in mid-sagittal brain MRI, and finally, lung, heart, clavicle location and cardiothoracic ratio in anterior-posterior chest radiographs.
TypePh.D. thesis [Academic thesis]
Year2004    pp. 248
PublisherInformatics and Mathematical Modelling, Technical University of Denmark, DTU
AddressRichard Petersens Plads, Building 321, DK-2800 Kgs. Lyngby
IMM no.IMM-PHD-2004-127
NoteAwarded the Nordic Award for the Best Ph.D. Thesis in Image Analysis and Pattern Recognition in the years 2003-2004 at SCIA'05.
Electronic version(s)[pdf]
BibTeX data [bibtex]
IMM Group(s)Image Analysis & Computer Graphics