Digital Total - Computing & Data Science an der Universität Hamburg und in der Wissenschaftsmetropole Hamburg

Automated symptom evaluation using sensor data from wearables in patients with Parkinson's disease utilizing explainable AI

102

Nicht eingeplant

20m

Von-Melle-Park 4

Poster

Sprecher

Alexander Wiederhold (Universität Hamburg)

Beschreibung

Parkinson’s disease is a neurodegenerative disorder that affects primarily dopaminergic neurons often showing a characteristic tremor or akinesis. While the severity estimation remains a stationary UPDRS scoring, we aim for an additional sensor-driven approach to develop an automated symptom evaluation for a second opinion. Our team collected over 200 million samples of accelerometer data and over 5000 corresponding UPDRS labels at the neurology department of the University Hospital Hamburg-Eppendorf. We implemented an explainable AI algorithm as well as a high-performance baseline model. First results show a high accuracy even for explainable classifiers. Consequently, we determined that a simple algorithm might be superior to a complex model in clinical routine to support clinicians’ decision process.

Keywords

Parkinson
Neurology
Sensor-driven-approach
AI
Accelerometer-data