Feature extraction for MRI segmentation.

摘要: Magnetic resonance images (MRIs) of the brain are segmented to measure efficacy treatment strategies for tumors. To date, no reproducible technique measuring tumor size is available clinician, which hampers progress search good protocols. Many segmentation techniques have been proposed, but representation (features) MRI data has received little attention. A genetic algorithm (GA) was used discover a feature set from multi-spectral data. Segmentations were performed using fuzzy c-means (FCM) clustering technique. Seventeen sets five patients evaluated. The GA produces more accurate segmentation. fitness function that achieves best results Wilks's lambda statistic when applied FCM clusters. Compared linear discriminant analysis, requires class labels, same or better accuracy obtained by features constructed without allowing fully operator independent approach therefore provides starting point measurement response treatment.