A novel approach to identify individual positioning in a range of supine postures
2019
INTRODUCTION: Pressure mapping provides visual feedback of the interface pressures between vulnerable tissues and supporting surfaces. However, the short-term nature of these measures provides limited insight into the temporal changes in pressure during evoked or self-induced movements. We examined the performance of selected parameters derived from continuous pressure monitoring and actimetry to detect postural changes. This yielded large data sets, which would benefit from intelligent data processing. This motivates the present study, which examines the accuracy of machine learning for the prediction of supine postures.
METHODS: Nineteen healthy participants adopted supine postures on a standard mattress, movements were evoked using the head of bed (HOB) angle and a tilting system to achieve sagittal (HOB between 0 and 60o) and lateral (left and right) postures, respectively. A series of time-related biomechanical parameters were estimated using a pressure monitor and actimetry placed on the sternum. Two supervised machine learning algorithms were assessed, namely K-nearest neighbors (KNN) and Naive-Bayes (NB), established with training data (n=9) and cross-validated with test data (n=10). KNN estimates the distance between a test data point and the nearest data point in the training phase, and NB the probability that a test data point belongs to specific cluster of postures.
RESULTS: Ranking of the biomechanical parameters revealed whole body contact area (>20mmHg) and trunk tilt angles provided the highest discrimination for postural changes. Separate clusters were identified for postures incorporating 20oHOB increments. The accuracy in predicting the range of sagittal and lateral postures was >80% for all subjects, for NB approach. By contrast, KNN accuracy resulted >70% for 8/10 subjects. An exemplar of both results are presented for one participant (Figure2). The NB algorithm was probably able to accommodate part of the non-linearity in the data, which could explain the differences in accuracy.
CONCLUSIONS: Accurate prediction of supine postures was achieved by combining machine-learning approaches with robust parameters estimated from two monitoring systems. This approach represents an advanced method of monitoring postures and mobility. Future work will combined evaluation of the local physiological response to these postures in order to create intelligent monitoring solutions. These technologies have the potential to identify pressure ulcer risk and efficient strategies for prevention in practice.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
0
References
0
Citations
NaN
KQI