Graph and Handwriting Signals-Based Machine Learning Models Development in Parkinson’s Screening and Telemonitoring

In this study, a series of machine learning models, based on the analysis of graph and handwriting signals, were developed for the screening and telemonitoring of Parkinson’s disease. The latter one is a multisystem neurodegenerative disease, which slowly affects the brain, and it is characterized by motor and non-motor symptoms that tend to worsen with the passage of time. The most common motor alterations are bradykinesia (motor slowing and reduced amplitude of motion), tremor, plastic hypertonia (muscle rigidity), balance disorders and postural instability. These motor disorders can also cause dysgraphia, i.e., changes in writing, as early as the onset. In particular, “micrography” is defined as the anomalous decrease in the size of the graphological tract. A careful analysis of the handwriting makes possible an early diagnosis and study of the evolution of the disease. For this purpose, data were collected from 16 parkinsonian and 42 healthy subjects. The patients carried out a simple and fast protocol that involves the execution of specific drawing and writing tasks. It was possible to do everything remotely; in fact, the objective of this work, which is part of a research project already started, is to intervene early at home and then direct the subject to instrumental examinations in the hospital. The data collection was carried out through a graphic tablet that, thanks to a specific acquisition software, allowed to know in real time information such as position, inclination and pressure of the digital pen. Therefore, different characteristics were extracted, then selected by statistical testing and with these different models were developed, which showed interesting accuracy values above 90%, in line and in some cases more performing than the current literature. Finally, an update was made to the application for smartphones and tablets in iOS and Android systems, which allowed the collection and saving of both graphic and voice signals in real time.