Identifying HRV patterns in ECG signals as early markers of dementia

Abstract

The appearance of Artificial Intelligence (IA) has improved our ability to process large amount of data. These tools are particularly interesting in medical contexts, in order to evaluate the variables from patients’ screening analysis and disentangle the information that they contain. We propose in this work a novel method for evaluating the role of electrocardiogram (ECG) signals in the human cognitive decline. This framework offers a complete solution for all the steps in the classification pipeline, from the preprocessing of the raw signals to the final classification stage. Numerous metrics are computed from the original data in terms of different domains (time, frequency, etc.), and dimensionality is reduced through a Principal Component Analysis (PCA). The resulting characteristics are used as inputs of different classifiers (linear/non-linear Support Vector Machines, Random Forest, etc.) to determine the amount of information that they contain. Our system yielded an area under the Receiver Operating Characteristic (ROC) curve of 0.80 identifying Mild Cognitive Impairment (MCI) patients, showing that ECG contain crucial information for predicting the appearance of this pathology. These results are specially relevant given the fact that ECG acquisition is much more affordable and less invasive than brain imaging used in most of these intelligent systems, allowing our method to be used in environments of any socioeconomic range.