Brain anatomical structure segmentation by adaptive bandwidth density estimation

Determination of region in a space of multimodal features of brain MR images requires kernel estimation tecniques with bandwidths that are adapted locally. The bandwidth selection is a critical aspect at the filtering stage of image segmentation. This work presents two methods for determinate the adaptive bandwidth in the application of density estimation, in the segmentation of regions at the feature space of an MRI. Two adaptive methods: sample point and k-nearest neighbors, where applied for real and synthetic data, achieved similarity indexes of 0.68 and 0.71 for gray matter and white matter respectively.