@MISC\{IMM2011-06031,
    author       = "J. Jensen and C. B. Anker",
    title        = "Segmentation of Abdominal Adipose Tissue in {MRI} in a Clinical Study of Growth and Diet",
    year         = "2011",
    publisher    = "Technical University of Denmark, {DTU} Informatics, {E-}mail reception@imm.dtu.dk",
    address      = "Richard Petersens Plads, {DK-}2800 Kgs. Lyngby, Denmark",
    note         = "The thesis was supervised by Professor Rasmus Larsen and Professor Knut Conradsen, {DTU} Informatics",
    url          = "http://www.imm.dtu.dk/English.aspx",
    abstract     = "This thesis describes Graph Cut as a method applied for automatic segmentation of boundaries using information in three dimensions from T1-weighted {3-}dimensional abdominal Magnetic Resonance Images, {MRI}. The data origins from a clinical research study where the subjects are scanned prior to and after 12 weeks of intervention.

The abdomen boundary, interior {SAT} boundary and Scarpa's Fascia divide the abdomen into three compartments containing different adipose tissue classes. The classes are called visceral adipose tissue, {VAT,} deep subcutaneous adipose tissue, dSAT, and supercial subcutaneous adipose tissue, sSAT. Research shows different clinical relevance of the three classes.

Before the segmentation the {MRI} data are preprocessed in several steps. The spatial image intensity inhomogeneities, called the Bias Field, are removed. Two methods, a Thin Plate Spline and N3, for correcting the bias field effect are investigated before choosing which to apply.

The interior {SAT} boundary and Scarpa's Fascia are located using Graph Cuts. A weighted directed graph is constructed from image characteristics. A Maximum Flow / Minimum Cut algorithm cuts the graph by finding the maximum flow. For labeling the adipose and nonadipose tissue two methods are compared, Fuzzy {C-}Means Clustering and Graph Cut.

A statistical analysis is performed on weight losses according to the intervention groups. Furthermore, the segmentation method described in this thesis is statistically compared to a previous method."
}