Detection of Alzheimer’s Disease Versus Mild Cognitive Impairment Using a New Modular Hybrid Neural Network

Abstract

Nowadays, there is a population ageing which leads to an increasing of geriatric and non-communicable diseases. One of the major socio-sanitary challenges our society is facing is dementia, with Alzheimer’s disease (AD) as the most prevalent one. AD is a progressive neurodegenerative disorder over years, with several stages. One of them is the prodromal one, also called Mild Cognitive Impairment (MCI). Despite the recent advances in diagnostic criteria for AD, its definitive diagnosis is just possible post-mortem because there is nonspecific AD biomarker. Therefore, an early and differential diagnosis of AD is still an issue of high concern. Extensive research looking for appropriate methods of diagnosis has been done. In this paper, we will present an innovative smart computing solution based on a hybrid and ontogenetic neural architecture, to deal with these challenges. It is an intelligent clinical decision-making system which has a non-neural pre-processing module and a neural processing one. This latter is a Modular Hybrid Growing Neural Gas (MyGNG), developed in this work. MyGNG consists of an input layer a Growing Neural Gas and a labelling layer based on the Perceptron algorithm. These modules are hierarchically organized and have different neurodynamic, connection topologies and learning laws. Using just neuropsychological tests of 495 patients (150 AD, 345 MCI) from ADNI repository, our proposal has provided very promising results in the early detection of AD versus MCI, reaching values of AUC of 0.95; Sensitivity of 0.89 and Accuracy of 0.81. It is an appropriate diagnosis system for any clinical setting.