Deep Learning-Based Segmentation in Classification of Alzheimer's Disease

The classification of Alzheimer's disease (AD) using ADNI dataset requires suitable feature segmenting techniques to detect the existing and relevant finer smaller brain region features, together with effective classification model, to eliminate a massive, labor-intensive and time-consuming voxel-based morphometry technique. Here, in this paper, a deep learning-based segmenting method using SegNet to detect AD pertinent brain parts features from structural magnetic resonance imaging (sMRI) and subsequently classifying accurately AD and dementia condition using ResNet-101 is presented. A deep learning-based image segmenting approach is experimented in detecting the delicate features of brain morphological changes due to AD that benefits classification performance for cognitive normal, mild cognitive impairment and AD, and thus provides an easy automatic diagnosis of Alzheimer's diseases. For classification, ResNet-101 is trained applying features extracted from SegNet with ADNI dataset. This paper demonstrated particularly to attain top-level automated classification. The seven morphological features like grey matter, white matter, cortex surface, gyri and sulci contour, cortex thickness, hippocampus and cerebrospinal fluid space extracted from 240 sMRI with SegNet are used to train ResNet for classification, and this classifier achieved a sensitivity of 96% and an accuracy of 95% over 240 ADNI sMRI other than used for training.