Meeting Banner
Abstract #1104

Robust Segmentation of Clinical Optic Nerve MRI

Swetasudha Panda1, Andrew J. Asman1, Bennett A. Landman1, 2, Seth A. Smith2, Louise A. Mawn3

1Electrical Engineering, Vanderbilt University, Nashville, TN, United States; 2Radiology and Radiological Sciences, Vanderbilt University, Nashville, TN, United States; 3Ophthalmology and Neurological Surgery, Vanderbilt University, Nashville, TN, United States

We aim to develop tools to automatically quantify the location and volumetrics of the optic nerve for integration of multimodal imaging data. We hypothesize that this will increase sensitivity and specificity of pathology assessments relative to coarse, manual region of interest approaches. Manual segmentation struggles significantly in the optic nerve when pathology is present or in the later stages of optic nerve damage. While multi-atlas segmentation promises a robust and model-free approach to segment medical images from exemplar images for brain structures, extrapolation to smaller structures of the human anatomy have largely been unexplored. Our purpose is to extend and evaluate multi-atlas labeling for the segmentation of the optic nerve based on high-resolution T2-weighted MRI of the optic nerve.

Abstract PDF Poster PDF

Keywords

accomplish accounting approaches assessment assessments atlas atlases atrophy audience automated automatic automatically bias body brain central characterize clinical clinicians coarse coil comprehensive conditions context control controls coordinate coronal correspondence datasets develop developing disease diseases disparity distance distances electrical engineering error estimation evaluate excitation explain extent fidelity findings form framework full fully fusion globe human inconsistencies inflammation initial integration involvement isotropic labels largely local localize localized manual manually many maximal medical minor model morel navigation nearest nerve nerves nervous neurological noble nominal note novel occasionally offered ophthalmology optic pairs pairwise patches pathological pathology patient patients physicists plots promises radiological radiology random randomly readout receive reconcile reduced registration regularization remaining resolution respective respectively rigid robust sciences sclerosis sections segmentation segmentations segmented selected sense significantly similarity slice slices software space specificity spin stages staple statistical structure structures struggled struggles subjects surface system target targets theory tools traditional tubular turbo undergo underlying versus volumes volumetric