Article Preview
Top1. Introduction
Dementia is a neuropsychological disorder that causes loss of memory leading to disability and thus dependency on others for survival. Dementia is commonly found in adults above the age of 60 years. Alzheimer’s disease (AD) is a type of dementia, being most prominent in almost 75% of the cases. Different sorts of dementia incorporate Frontotemporal Dementia, Vascular Dementia, Parkinson’s disease, Dementia with Lewy Bodies, etc. As indicated by the World Health Organization (WHO), almost 50 million people are experiencing Dementia across the world and about 10 million new cases are expected to emerge every year (Dementia, 2020). The treatment of dementia at an early stage is very important for the social and economic impact of the disease.
Progression of dementia can be determined with a timely diagnosis and neuro-imaging analysis can be substantially helpful in its detection (Altaf et al., 2017). Various machine learning techniques implemented for the detection of the disease have been reviewed by various researchers (Bansal et al. 2018; Mirzaei et al., 2016; Ahmed et al., 2019). The multi-class classification of a subject into demented, very mild demented, and normal is the foremost for the diagnosis. But before classification, irrelevant and redundant features are removed through efficient feature selection techniques for achieving a better classification accuracy (Bansal et al., 2018; Bansal et al., 2019; Dallora et al., 2017).
The performance of machine learning approaches is relatively lower with a large amount of data. It can be a challenge for the diagnosis of brain disease. Deep learning approaches can overcome the pitfalls of machine learning approaches. It can also recognize the new features using self-learning of features for the quantitative analysis of MRI. Deep Learning has gained a lot of attention in various medical fields (lin et al., 2016), for example, histopathological disease (Litjens et al., 2016), pulmonary modules (Cheng et al., 2016), and breast lesions (Kooi et al., 2017).
This study would help the researchers to get answers to various research questions listed below pertaining to the implementation of Deep Learning for detecting dementia using MRI images.
RQ1: What are the available pre-trained architectures for Deep learning?
RQ2: What are the various software platforms which can be applied in this area?
RQ3: What kind of preprocessing techniques are necessary for MRI Images?
RQ4: What are the different types of deep learning models used for detecting Dementia?
A comparative analysis of different optimizers and hyperparameters is also presented in this study using transfer-learning. AlexNet architecture is used for tweaking the hyperparameters. Multiclass classification is performed for the detection of dementia using MRI images obtained from the OASIS dataset.
The remaining part of the paper is sorted out as follows: A comprehensive literature survey is given in Section 2. Section 3 outlines the experimental results and analysis, followed by a conclusion of the complete work.
Top2. Literature Survey
Deep Learning is contributing a lot in the early detection and restriction of the progression of Dementia (LeCun et al., 2015). A great number of architectures (Ravi et al., 2017) stand out in vogue, among different methodological renditions of deep learning. The Convolutional Neural Network (CNN) (Lecun et al., 1998) is one of the most accepted algorithms utilized for deep learning in medical imaging. The idea of CNN roused from the neurobiological model of the visual cortex (Hubel and Wiesel, 1962). Other credible architectures for Deep Learning incorporate Deep Belief Networks (DBNs) (Hinton et al., 2006), Restricted Boltzmann Machines (RBMs) (Hinton and Sejknowski, 1986), Deep Boltzmann Machines (R. and Hinton, 2009), Deep Autoencoders (Hinton, 2006), and Recurrent Neural Networks (RNNs) (Williams and Zipser, 1989).