paralysis agitans
Conditions
Interventions
None listed
Sponsors
Radboud Universiteit Nijmegen
Eligibility
Age
18 Years to 99 Years
Inclusion criteria
Inclusion criteria: In order to be eligible to participate in this study, a subject must meet all of the following criteria: 1. Self-reported diagnosis of Parkinson*s disease; 2. Use of levodopa or other Parkinson*s disease medication; 3. 18 years or older; 4. Freezing of gait episodes experienced on a daily basis (New freezing of gait questionnaire answer to question 2 * *How frequently do you experience freezing episodes?* * * Very often, more than once a day); 5. No cognitive or psychiatric impairment as judged by the researcher; 6. Possession of a smartphone with suitable Android operating system; 7. Able to provide informed consent.
Exclusion criteria
Exclusion criteria: A potential subject who meets any of the following criteria will be excluded from participation in this study: 1. Incapacitating dyskinesias or dystonia; 2. Comorbidities that cause severe gait impairment (e.g. severe arthrosis or neuropathy); 3. Usage of advance therapies such as Deep Brain Stimulation; 4. Freezing of gait episodes exclusively in ON period (because this is thought to have a different pathophysiologic mechanism than the more regular version of FOG which occurs predominantly during OFF).
Design outcomes
Primary
| Measure | Time frame |
|---|---|
| Machine learning and deep learning techniques will be applied to outcomes derived from clinical assessments, wearable motion sensors (e.g. gait related: abnormal gait classification, gait cadence, gait velocity, freezing index, number of FOG episodes, percentage time spent on FOG, heart rate variability, and skin conductivity), and video (presence of FOG) in order to create an algorithm aiming at predicting an upcoming FOG episode and detecting an existing FOG episode. Afterwards, ensemble technique will be applied to combine the results from multiple different algorithms or parameters into a single result. This step will increase algorithm precision. The same techniques will be applied to data collected during the follow-up, in order to develop an algorithm robust predicting an upcoming FOG episode and detecting an existing FOG episode under free-living circumstances. | — |
Secondary
| Measure | Time frame |
|---|---|
| To investigate the influence of different sensor types, sensor number, and sensor locations on the performance of a FOG detection and prediction algorithm performance we will create different prediction algorithms using data from clinical assessments (see table 1) and data from wearable sensors collected in different numbers and sensor location. Models will be created using Machine learning and/or Deep learning approaches. Machine learning algorithms build a mathematical model based on sample data, here outcomes extracted from the sensors and clinical assessments, in order to make predictions or decisions without being explicitly programmed to perform the task. Self-reported FOG, falls and balance problems, extracted from the smartphone application, will increase model performance. Afterwards, model performance measures from models including data from all sensors versus data from sensor from one location only, as well as models from semi-free-living condition versus free-living conditions will be compared in order to stablish the best model. Next, we will apply the model created using data from study visit 1 (see paragraph 10.1) to the data collected in study visit 2. We will then compare the performance measures of each model and determine whether the model is robust for a population that shows changes over time (i.e. the test-retest reliability of the algorithm for prediction FOG episodes). Finally, we will correlate balance-related outcomes (see table 1) to gait variability measures (i.e. stride time, stride length, gait cadence, and gait velocity). To achieve that we will apply extract from the raw sensor data outcomes such as: level of chaos in the signal (maximal Lyapunov exponent), stride time variability, stride length variability, and time in double stance. To increase reliability, all outcomes will be averages over a 3-days period. Those outcomes will them be used to detect a change on balance against clinical measurements (collec | — |
Countries
Netherlands
Outcome results
None listed