Artificial neural networks in Alzheimer's diagnosis: a perspective

Abstract

A clear tendency of an aging population (2.5 billion elders are estimated on a global scale by the year 2050) has brought about an increase of its associated diseases, one of which is the higher prevalence dementia focusing in Alzheimer's Disease (AD). Today, it is estimated that there are 18 million people suffering from AD worldwide, and the disease affects 5% -10% of 65-year old and more than 30% of 85-year old. This situation has important repercussions in the scope of the patient but also in the familiar, social and sanitary spheres. Therefore, early diagnosis of AD is a major public healthcare concern, and the Differential Diagnosis of Dementia (DDD) is also one of the crucial points to which clinical medicine faces at every level of attention. The definite diagnosis of AD is only post-mortem. Furthermore, there are not yet a specific set of diagnostic criteria for the confirmation of the diagnosis. In this context it's necessary to develop new alternative methods and instruments of diagnosis, especially on early and differential diagnosis, and introducing its use in all healthcare areas. This chapter will be dedicated to explore the ability of a complementary approach to face these problems, the Artificial Neural Networks (ANNs). The ANNs are highly non-linear systems. Its more appealing property is its learning capability. Its behaviour emerges from structural changes driven by local learning rules, having the capability of generalisation. In addition to this approach, especially computer-intensive algorithms based on "ensemble learning"-methods that generate many classifiers and aggregate their results are being developed in regard of Mild Cognitive Impairment (MCI), AD and DDD classification. We will present a study of ANNs, where it will be analysed a new neural architecture, HUMANN-S, which has shown to be a very suitable ANN for Alzheimer's diagnosis scope. The neural network ensemble approach is introduced. Finally we will discuss the ability of ANN and neural network ensembles, to address this issue, describing the outcomes of implementations of such approaches for AD, DDD and MCI diagnosis using for the inputs several types of data: Electroencephalogram (EEG) type signals, neuroimages, like Single Photon Emission Computerized Tomography (SPECT), and/or scores of different neuropsychological tests, among others