A new scheme for the automatic assessment of Alzheimer's disease on a fine motor task with Transfer Learning.

We present a new scheme for Alzheimer's Disease (AD) automatic assessment, based on Archimedes spiral, drawn on a digitizing tablet. We propose to enrich spiral images generated from the raw sequence of pen coordinates with dynamic information (pressure, altitude, velocity) represented with a semi-global encoding in RGB images. By exploiting Transfer Learning, such hybrid images are given as input to a deep network for an automatic high-level feature extraction. Experiments on 30 AD patients and 45 Healthy Controls (HC) showed that the hybrid representations allow a considerable improvement of classification performance, compared to those obtained on raw spiral images. We reach, with SVM classifiers, an accuracy of 79% with pressure, 76% with velocity, and 70.5% with altitude. The analysis with PCA of internal features of the deep network, showed that dynamic information included in images explain a much higher amount of variance compared to raw images. Moreover, our study demonstrates the need for a semi-global description of dynamic parameters, for a better discrimination of AD and HC classes. This description allows uncovering specific trends on the dynamics for both classes. Finally, combining the decisions of the three SVMs leads to 81.5% of accuracy.Clinical Relevance- This work proposes a decision-aid tool for detecting AD at an early stage, based on a non-invasive simple graphic task, executed on a Wacom digitizer. This task can be considered in the battery of usual clinical tests.

Epoch-based Entropy for Early Screening of Alzheimer's Disease.

In this paper, we introduce a novel entropy measure, termed epoch-based entropy. This measure quantifies disorder of EEG signals both at the time level and spatial level, using local density estimation by a Hidden Markov Model on inter-channel stationary epochs. The investigation is led on a multi-centric EEG database recorded from patients at an early stage of Alzheimer's disease (AD) and age-matched healthy subjects. We investigate the classification performances of this method, its robustness to noise, and its sensitivity to sampling frequency and to variations of hyperparameters. The measure is compared to two alternative complexity measures, Shannon's entropy and correlation dimension. The classification accuracies for the discrimination of AD patients from healthy subjects were estimated using a linear classifier designed on a development dataset, and subsequently tested on an independent test set. Epoch-based entropy reached a classification accuracy of 83% on the test dataset (specificity = 83.3%, sensitivity = 82.3%), outperforming the two other complexity measures. Furthermore, it was shown to be more stable to hyperparameter variations, and less sensitive to noise and sampling frequency disturbances than the other two complexity measures.