Segmentation of multiple sclerosis lesions in MRI by fuzzy neural networks: FLVQ and FOSART

In this paper two fuzzy neural networks have been applied for the segmentation of magnetic resonance (MR) images. The objective of the work is to state the effectiveness of a fuzzy-lesions approach for the detection of the small lesions present in thick MR slices of multiple sclerosis patients. The data set included the proton density (PD), T2, T1 weighted spin-echo (SE) bands and a new T1-weighted three dimensional sequence, i.e. the magnetization-prepared rapid gradient echo (MP-RAGE) of a volunteer. The fuzzy learning vector quantization (FLVQ) and the fully self organizing map (FOSART) models have been used for the semi-automatic tissue segmentation of the multi-spectral data set. Both models were trained with the pixels extracted from some labelled areas, interactively selected by a neuro-radiologist on the input raw images. A quantitative comparison between the two neural networks' performance has been provided on the base of the labelled areas.