Deep Learning Models for Detection and Diagnosis of Alzheimer's Disease

Introduction

Deep learning (LeCun et al., 2015) is a subpart of a machine learning family and has very high computational power. The sudden and extreme growth of deep learning is because of its very high computational power and the availability of huge datasets. The dramatic advancement in the field of deep learning is the manipulation capability of machines especially speech (Hinton et al., 2012), images (Russakovsky et al., 2015), and languages (Hirschberg & Manning, 2015). Deep learning models can manipulate large datasets with the requirement of high computing hardware and will improve gradually with the increasing size of data and thus enhancing its capability to do better than many traditional machine learning approaches. The striking feature of deep learning is to accept many data types as input which specifies its aspect of specific pertinence for different health care data. The figure below depicts a simple, multilayer deep neural network that takes input data from two different classes of data, with two different colors, and separates them in a linear fashion linearly by recursively changing the data as it flows from one layer to next layer. The classification is done by last output layer and generates the possible output from any one of the class. This illustration explains the straightforward perception implemented by huge scale networks.

Figure 1.

Figure 2.

Feature learning from variety of data types by deep learning