An atlas-free newborn brain image segmentation and classification scheme based on SOM-DCNN with sparse auto encoder

Abstract

ABSTRACTAutomatic segmentation and classification of infant brain MRI is a fundamentally difficult task because of the low contrast and the development process of the brain tissues. The strategies adapted for adult brain MRI segmentation are not reasonable to the neonatal brain, because of the immense differences in structure and tissue properties amongst infant and adult brains. The current infant brain MRI segmentation techniques depend on manual interaction and they used atlases or formats which could not fragment most extreme number of tissues. In this paper, we proposed an atlas-free infant brain image segmentation and classification scheme in light of self-organised map (SOM) – deep convolutional neural network (DCNN) with sparse auto encoder. The hybridisation of SOM-DCNN scheme gives accurate segmentation of infant brain tissues and the classification is performed through SAE (sparse auto encoder). Our proposed scheme is approved through dice metrics, which demonstrates that our proposed strategy gives accurate result, compare to existing segmentation schemes.