Abstracts

Theme 2 | E- & M-health in mental health care and research

Can personalized models improve naturalistic motor fluctuation detection in Parkinson’s disease? 

Jeroen Habets


Jeroen Habets 1, Ro’ee Gilron 2, Simon Little 2, Christian Herff 1, Mark Kuijf 3, Yasin Temel 1,4, Pieter Kubben 1,4

1 Department of Neurosurgery,  MheNs division III, Maastricht University
2 Department of Neuromodulation and Neurosurgery, University of California San Francisco
3 Department of Neurology, Maastricht UMC+
4 Department of Neurosurgery, Maastricht UMC+

Background
Motor fluctuations appear in the majority of Parkinson's disease (PD) patients within the first decade after diagnosis, due to disease progression and wearing-off of dopaminergic medication. The well-treated state without symptoms is called ON-medication, the state with burdensome motor symptoms is called OFF-medication. The evaluation of motor fluctuations is essential to manage therapy. Currently, motor fluctuations is heavily dependent on patient recall, labor-intensive examination, and questionnaires covering days till weeks. First generation PD monitor systems based on motion sensors augment clinical decision making based on multiple day naturalistic (during real-life) monitoring sessions. However, it is unclear how reliable motor fluctuations can be monitored over shorter time windows. Short-term motor fluctuation monitoring is required for dynamic therapy evaluation such as closed-loop deep brain stimulation. We investigate personalized models to improve short-term naturalistic motor fluctuation monitoring in PD.

Methods
We included 20 PD patients with hand-bradykinesia fluctuation between their ON- and OFF-states. An hour of daily life activities was recorded at home during the ON-state, and an hour during the OFF-state. We extracted 103 features from data of a wrist-worn accelerometer, such as maximum acceleration, variance of acceleration, spectral power in certain frequency ranges, and the jerkiness of movement. Features were calculated over different feature windows. Classification models using support vector and random forest classifiers were trained for every patient, based on their own individual data, or based on group data.

Results
Our results show that short-term (60 seconds) motor fluctuation detection is possible in the majority of the included PD patients. Individual models did not differ from group models, although there was a trend of higher results with the individual models. Sub analysis in the group models showed improved results with feature windows of 600 seconds, and with larger training data sets.

Conclusion
Our findings indicate that motor fluctuation monitoring is possible on short-term windows. Further research should be done with larger training data for individual models, including a valid activity classifier, and feature windows of 600 seconds.

The School for Mental Health and Neuroscience (MHeNs) strives to advance our understanding of brain-behaviour relationships by using an approach integrating various disciplines in neuro- and behavioural science, medicine, and the life sciences more widely. MHeNs performs high-impact mental health and neuroscience research and educates master's students and PhD researchers. MHeNs performs translational research, meaning practical collaboration between researchers in the lab and in the hospital. MHeNs is one of six graduate schools of the Faculty of Health, Medicine and Life Sciences (FHML) aligned to the Maastricht University Medical Centre+ (MUMC+).