Deep integration model: A robust autonomous segmentation technique for hippocampus in MRI images of human head

Abstract

Segmentation of hippocampus in Magnetic Resonance Images (MRI) is a vital process in neuro imaging studies towards analyzing and extracting the structure of hippocampus. In existing, many automatic segmentation algorithms such as region based segmentation; boundary based Segmentation has been applied to segment the Hippocampus (HC) as it produces the higher accuracy and less mean squared error. Despite several advantages, the model is time consuming and it generates less accuracy on the volumetric changes of the diverse dataset. However, due to the existence of imaging artifacts, changes in anatomy, contrast variability, and poor registration, most of these mechanisms do not produce suitable results. In order to resolve those issues, a novel framework named as Deep Integration Model has been proposed using Sparse Principle Component Analysis, Affinity Propagation and Linear Convolution Neural Network to segment the MRI Images. The model segments the Hippocampus and its subfields in the MRI images. Affinity Propagation identifies the modalities and pathologically altered tissues and represented it as features set. Sparse Principle Component Analysis extracts the sparse information and its interrelationship on the feature set which is represented as max pooled data.