Conference paper Open Access

Early Parkinson's Disease Detection via Touchscreen Typing Analysis using Convolutional Neural Networks

Dimitrios Iakovakis; Stelios Hadjidimitriou; Vasileios Charisis; Sevasti Bostanjopoulou; Zoe Katsarou; Lisa Klingelhoefer; Simone Mayer; Heinz Reichmann; Sofia B. Dias; José A. Diniz; Dhaval Trivedi; Ray K. Chaudhuri; Leontios J. Hadjileontiadis

Parkinson’s Disease (PD) is the second most common neurodegenerative disorder worldwide, causing both motor and non-motor  symptoms. In the early stages, symptoms are mild and patients may ignore their existence. As a result, they do not undergo any related clinical examination; hence delaying their PD diagnosis. In an effort to remedy such delay, analysis of data passively  captured from user’s interaction with consumer technologies has been recently explored towards remote screening of early PD motor signs. In the current study, a smartphone-based method analyzing subjects’ finger interaction with the smartphone screen is developed for the quantification of fine-motor skills decline in early PD using Convolutional Neural Networks. Experimental results from the analysis of keystroke typing in-the-clinic data from 18 early PD patients and 15 healthy controls have shown a classification
performance of 0.89 Area Under the Curve (AUC) with 0.79/0.79 sensitivity/specificity, respectively. Evaluation of the generalization ability of the proposed approach was made by its application on typing data arising from a separate self-reported cohort of 27 PD patients’ and 84 healthy controls’ daily usage with their personal smartphones (data in-the-wild), achieving 0.79 AUC with 0.74/0.78 sensitivity/specificity, respectively. The results show the potentiality of the proposed approach to process keystroke dynamics arising from users’ natural typing activity to detect PD, which contributes to the development of digital tools for remote pathological symptom screening.

Files (5.0 MB)
Name Size
5.0 MB Download
All versions This version
Views 6666
Downloads 77
Data volume 34.9 MB34.9 MB
Unique views 5959
Unique downloads 77


Cite as