Implementation of an Artificial Neural Network on the Test Barcelona Workstation As a Predictive Model for the Classification of Normal, Mild Cognitive Impairment and Alzheimer’s Disease Subjects Using the Neuronorma Battery

Abstract

Objective: To develop and implement an online Artificial Neural Network (ANN) that provides the probability of a subject having mild cognitive impairment (MCI) or Alzheimer’s disease (AD). Method: Different ANNs were trained using a sample of 350 controls (CONT), 75 MCI and 93 AD subjects. The ANN structure chosen was the following: (1) an input layer of 33 cognitive variables from the Neuronorma battery plus two sociodemographic variables, age and education. This layer was reduced to a 15 features input vector using Multiple Discriminant Analysis method, (2) one hidden layer with 8 neurons, and (3) three output neurons corresponding to the 3 expected cognitive states. This ANN was defined in a previous study [28]. The ANN was implemented on the web site www.test-barcelona.com (Test Barcelona Workstation) [9]. Results: When comparing CONT, MCI and AD participants, the best ANN correctly classifies up to 94,87% of the study participants. Conclusions: The online implemented ANN, delivers the probabilities (%) of belonging to the CONT, MCI and AD groups of a subject assessed using the 35 characteristics (variables) of the Neuronorma profile. This tool is a good complement for the interpretation of cognitive profiles. This technology improves clinical decision making.