| Abstract [eng] |
With an increasing number of patients with frailty being referred for surgery, there is a need for convenient tools that would help to improve the understanding of the effectiveness of exercise-based rehabilitation. Accordingly, this doctoral thesis proposes and investigates a wearable-based approach for unobtrusive assessment of frailty. Currently, the clinical practice for assessing frailty is limited to in-clinic evaluations. However, wearable technology has advanced to the point where frailty can be assessed outside of the clinical setting. Most previous research has focused on identifying frailty or pre-frailty in older adults. However, the feasibility of capturing subtle changes in the frailty status during exercise training has yet to be explored. This thesis aims to develop, investigate, and validate algorithms for unobtrusive monitoring of the frailty status in the activities of daily living. A derivative dynamic time warping-based algorithm has been proposed to detect physical stressors, namely, walking and stair-climbing, in wearable-based biosignals. For parametrization of frailty, algorithms that assess the kinematic properties and heart rate to physical stressors have been explored. A concept of interpretable machine learning has been proposed for identifying clinically informative features that would provide information on the frail physiological functions of an individual patient. |
