Purpose: Recent advances in wearable technology have enabled us to visualize how stroke patients spend time and how they behave during physical activity over a period of 24-h. This capability is expected to provide new insights into rehabilitation research. However, wearable electrocardiograph (ECG) monitoring devices generate noise and artifacts from external factors such as body movements, which causes misdetection of the R wave. To compensate for the loss of heart rate data due to misdetection, the present study focused on ensemble averaging as a method to impute missing data and validated this method of data imputation on 24-h recordings. Methods: First, to investigate the measurement period for data imputation, we continuously measured heart rates of six healthy participants for four days and applied ensemble averaging to the first two, three and all four days of measurement data. Next, we validated the imputation by ensemble averaging with 218 measurement data from 63 stroke inpatients in a rehabilitation ward. Results: For the healthy participants, the period with data losses decreased from 115 min (8.3% of the 24-h) to 5.5 min (0.4%), 0 min (0%), and 0 min (0%) when ensemble averaging was applied to two, three, and four days of measurement data, respectively. For data of stroke patients acquired in a two-day measurement session, ensemble averaging decreased the period with data losses more than the conventional method that selects data of the day with the least missing data among the measurement days did (0.17 and versus 1.7% of the 24-h). The median and maximum heart rate when ensemble averaging was applied was strongly correlated to the median and maximum heart rate when ensemble averaging was not applied. Conclusions: The results suggest that ensemble averaging is useful for imputing missing vital data such as heart rate.
All Science Journal Classification (ASJC) codes
- Biomedical Engineering