Analysis of postural kinetics data using Artificial Neural Networks in Alzheimer's Disease

Abstract

Inertial measurement Units (IMU) (accelerometers and gyroscopes), placed in strategic parts of the human body, are a growing field on kinetic posture and imbalance study in Alzheimer's Disease (AD). On the other hand, Artificial Neural Network (ANN) are a powerful statistical tool used on pattern recognition on big data such as IMU kinetic records. Still, on ANN research, issues like the best number of hidden layers and the best number of neurons in each hidden layer remain open. In our study we developed a software tool of Multilayer Perceptrons ANN analysis (Back Propagation and Scale Gradient Conjugate training algorithms) that automatically tests different configurations for the ANNs on the diagnosis of Alzheimer's disease. Analysis was performed primarily on all 159 variables, biometrics and IMU records of 21 AD patients and 21 healthy subjects exposed to seven different tasks with increasing postural stress, and posteriorly on selected data derived from Mann-Whitney analysis. Multilayer Perceptron ANN reached a performance of 95% on the diagnosis of AD, proving to be a potential useful clinical tool.