Meeting Banner
Abstract #0952

Fully Automatic Segmentation of the Amygdala on High Resolution T1 Images Using a Shape Model

Frank Thiele1, 2, Lukas Scheef2, Fabian Wenzel3, Carsten Meyer3, Henning Boecker2, Michael Wagner4, Hans H. Schild2, Frank Jessen4

1Philips Research, Aachen, Germany; 2Radiology, University of Bonn, Bonn, Germany; 3Philips Research, Hamburg, Germany; 4Psychiatry, University of Bonn, Bonn, Germany

Automatic segmentation of amygdala volumes would present an important tool for neuroscience studies in cognition and psychiatry, and a potential diagnostic marker. In this work, a shape model is applied to T1-weighted MRI for fully automatic segmentation of the amygdala in 70 elderly normals. Segmentation is compared to manual tracing as well as a state-of-the-art atlas-based approach. The shape model is found to be a promising approach for reproducible and observer-independent analysis of the amygdala.

Abstract PDF

Keywords

adapted affected anatomical atlas audience automatic benefit best binary brain calculating channel coefficients coil comparable concordance consecutively consistent constrained contained coronal corr corrected correlation correlations covered dataset definition deformable delineated demanding described detail determined dice disease disorders elderly emotion enables evaluate exact example expected failed female field final frank fully generally good greater green hamburg head henning important independent input intra introduced involved known larger leaving like likely manual manually mask masks memory mismatch model models moderate motion observer outside overlap particular plays practical processing protocols psychiatric psychiatry radiology rater reasonable recently regional reliability reported reproducible required requiring resolution role segmentation segmentations segmented sense shape slice springer standardization structure structures studied subject subjects surface table target tracing tracings trained turbo volume volumes yellow yielded