The accurate knowledge of tumor in brain MRIs and its robust segmentation plays a vital
role in the diagnosis and treatment of various clinical problems. The challenging problem for segmenting
medical images is a selection of initial seed points. It is difficult to locate the seed points,
due to the complex anatomical structure of the brain, which is highly heterogeneous in appearance,
and the ambiguity of the location of a brain tumor. In this paper, we propose an approach for the automatic
segmentation of brain MRIs by selecting the seed points and employing fuzzy graph cut
technique. Fuzzy clustering technique provides the clusters encompassing the potential intensity values
which act as the kernels required for the initialization of the graph cut algorithm. We present a
quantitative evaluation of the publically available data set from the MICCAI multimodal brain tumor
segmentation challenge (BraTS). The experimental results obtained, demonstrated that the proposed
method is automated and more efficient as compared with the results obtained by previous algorithms
using same data set. The mean values of DSC and PPV obtained are: 0.89 and 0.87, respectively
for the LGG images and 0.92, 0.90. respectively for HGG images.
Keywords: MRI, graph cut, fuzzy, seed selection, dice sensitivity coefficient, positive predicted value.
Rights & PermissionsPrintExport