9 May 2002 Robust semi-automatic segmentation of single- and multichannel MRI volumes through adaptable class-specific representation
Author Affiliations +
Abstract
Segmentation of MRI volumes is complicated by noise, inhomogeneity and partial volume artefacts. Fully or semi-automatic methods often require time consuming or unintuitive initialization. Adaptable Class-Specific Representation (ACSR) is a semi-automatic segmentation framework implemented by the Path Growing Algorithm (PGA), which reduces artefacts near segment boundaries. The user visually defines the desired segment classes through the selection of class templates and the following segmentation process is fully automatic. Good results have previously been achieved with color cryo section segmentation and ACSR has been developed further for the MRI modality. In this paper we present two optimizations for robust ACSR segmentation of MRI volumes. Automatic template creation based on an initial segmentation step using Learning Vector Quantization is applied for higher robustness to noise. Inhomogeneity correction is added as a pre-processing step, comparing the EQ and N3 algorithms. Results based on simulated T1-weighed and multispectral (T1 and T2) MRI data from the BrainWeb database and real data from the Internet Brain Segmentation Repository are presented. We show that ACSR segmentation compares favorably to previously published results on the same volumes and discuss the pros and cons of using quantitative ground truth evaluation compared to qualitative visual assessment.
© (2002) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Casper F. Nielsen, Casper F. Nielsen, Peter J. Passmore, Peter J. Passmore, } "Robust semi-automatic segmentation of single- and multichannel MRI volumes through adaptable class-specific representation", Proc. SPIE 4684, Medical Imaging 2002: Image Processing, (9 May 2002); doi: 10.1117/12.467132; https://doi.org/10.1117/12.467132
PROCEEDINGS
11 PAGES


SHARE
Back to Top