Convolutional neural networks-based early Parkinson's disease classification using cycling data from a steerable indoor bicycle.

Parkinson's disease (PD) is a neurodegenerative disorder that leads to a decline in functions that are directly related to quality of life, and it is vital to detect PD in the early stages. Currently, the diagnosis of PD relies on comprehensive assessment of medical history and clinical examination. However, it is difficult to capture short-term variations in the patient's disability level during medical visitations, especially in its early stage. In this study we explore the feasibility of PD detection using a CNN model and a bicycle mounted on a specially designed bicycle platform. The bicycle platform, Ultiracer, allows for some lateral movement via steering in order to simulate outdoor cycling experiences. The bicycle is equipped with two 6-axis force-torque sensors, one in the headset spacer and the other in the seat post. The input data for the CNN model consist of 30 s of data recorded during cycling (force, moment, speed, and lateral movement) and tabular data comprising personal information (sex, age, height, and weight). The results from 29 PD patients and 36 healthy controls, suggest the proposed CNN model presented an accuracy of approximately 86% in classification based on the 5-fold cross validation.