Detecting Parkinson’s Disease Using Machine Learning from Movement and Memory Data - Walking Task Analysis

Aug 5, 2026

Study evaluates machine learning methods, including decision trees and artificial neural networks, for detecting Parkinson’s Disease using data collected through the mPower mobile iOS app by Sage Bionetworks. The work centers on the walking task in version 1.0 build 7, requiring participants to walk unassisted for about 20 seconds, stand still for 30 seconds, then return for 20 seconds. Smartphone accelerometer and gyroscope signals capture 3D rotation during outward walking, return walking, and rest, enabling a convolutional neural network to assess confirmed PD cases.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Abstract

Methods and Data Source

Walking Task Protocol

Sensor Signals and Model Approach

Related Work

Prior mPower-Based Studies

Canadian Journal for the Academic Mind 1.1 (2023) Detecting Parkinson’s Disease Using Machine Learning from Movement and Memory Data Emma Richard a*, Mary Kelly b a Department of Cognitive Science, Carleton University 0000-0003-1164-6862 b Department of Cognitive Science, Carleton University 0000-0002-6541-2992 | ABSTRACT: This study explored the potential of machine learning techniques, specifically decision trees and artificial neural networks (ANNs), to detect Parkinson's Disease (PD) using data collected from the mPower mobile iOS app. Released in 2016 by Sage Bionetworks, mPower enables individuals, both with and without PD, to assess their cognitive and physical abilities through various tasks related to memory, tapping, voice, and movement. The main focus of this study is on the walking task within the app's version 1.0 build 7. Participants were required to walk unassisted for approximately 20 seconds in a straight line, followed by a 30-second period of standing still, and then returning with 20 seconds. The smartphone's accelerometer and gyroscope captured three-dimensional (3D) rotation data (x, y, z) during these movements, with the device placed in the participant's pocket or bag. A convolutional neural network was applied to the movement dataset to assess confirmed PD cases, utilizing accelerometer and gyroscope readings during outward walking, return walking, and rest periods. | | --- | | KEYWORDS: Parkinson’s, artificial intelligence, movement, memory | [https://doi.org/10.25071/2817-5344/55](https://doi.org/10.25071/2817-5344/55) * Corresponding Author-Email Address: emmarichard@cmail.carleton.ca Received 10 June 2023; Received in revised form 03 Aug 2023; Accepted 12 Aug 2023 © 2023 The Author(s) . This is an open access article under the CC BY-NC-SA 4.0 license Detecting Parkinson’s Disease…(Richard, Emma, et al.) PD occurs spontaneously, as it affects the nervous system based on motor and non-motor symptoms.1 The exact cause in most cases is not known or may not be attributed to a specific external factor, but it does not negate the presence of prodromal symptoms or other factors that can contribute to its development. It mainly affects older adults and is less common for people under 40 years old, and it is the second most common neurodegenerative disorder.2 If you are diagnosed with PD under 40 years old it is considered early onset and it tends to be because of genetics.3 The main risk factors related to PDare aging, genetics, and exposure to chemicals in the environment.4 A known environmental risk factor is exposure to the pesticide paraquat, which is used to control weeds.5 There have been animal models that show similar pathology to PD when they are exposed toparaquat. These models would show animals having an increase in anxiety one-week post-treatment and impaired motor skills 3 weeks post-treatment.6 There are other neurodegenerative diseases that are similar to PD, such as Dementia with Lewy Bodies or Multiple System Atrophy, but what separates PD is the tremor at rest, rigidity, slowed movement, and poor posture.7 There are about 60,000 new cases of PD diagnosed each year and 7-10 million people are affected worldwide.8 When diagnosing PD, it is mainly based on the symptoms because there are no definitive tests to classify it.9 It can be expensive to diagnose PD because doctors have to eliminate other neurological diseases to correctly diagnose the patient with PD, this can take hours to days.10 PD typically starts with non-motor symptoms such as depression or anxiety and progresses into the patient having motor symptoms such as slowed movement or akinesia.11 The later stages of PD can result in the patient having dementia due to the degradation of the midbrain.12 Released in 2016, Sage Bionetworks created a mobile iOS app called mPower to allow people with or without PD to measure their cognitive and physical capabilities. They created various tasks that people could compl

FAQs

Which machine learning techniques are evaluated for Parkinson’s disease detection?

The study explores decision trees and artificial neural networks, and applies a convolutional neural network to movement data to assess confirmed PD cases.

How does the walking task in mPower version 1.0 build 7 work in this study?

Participants walk unassisted in a straight line for about 20 seconds, stand still for 30 seconds, then walk back for another 20 seconds.

What sensor data is used during the task and how is it collected?

The smartphone accelerometer and gyroscope record 3D rotation signals (x, y, z) while the device is placed in the participant’s pocket or bag.

Aug 5, 2026

Upload by Emma Wilson