Classification of Patients with Parkinson’s Disease Using Free Handwriting Features Collected through a Smart Ink Pen

Abstract

Systems for monitoring of Parkinson’s disease (PD) patients, able to complement clinical assessment, are needed. These solutions should be objective, based on technology that captures physical characteristics of the pathology, and capable of providing frequent measures conducted both on-site and remotely. Since one of the most typical clinical hallmarks of PD is handwriting deterioration, we devised an innovative smart ink pen for quantitative and reliable handwriting monitoring, without altering the natural writing conditions. 30 PD patients and 30 age-matched controls performed two unconstrained writing tasks (free text and grocery list) with the smart ink pen. A series of 47 writing and tremor indicators were computed and used to classify patients from age-matched controls. Catboost and Logistic Regression classifiers were used, and the SHAP model explanation technique was applied to explore the contribution of the features in the classification. Very good performances were obtained through the Catboost classifier when combining features extracted from both tasks (Accuracy: 93%, Precision: 96%, Recall: 90%; F1: 93%; AUC: 98.9%). We achieved a classification performance in line with previous work, with two main advantages: writing data acquisition through an ink pen used on common paper, and proposition of an unconstrained protocol mimicking daily-life writing.