Analysis of movement patterns and quantitative parameters from inertial data to assess patients motor performances during tele-rehabilitation

Tele-rehabilitation is emerging as a new opportunity to deliver rehabilitation at patients' home, by extending healthcare services accessibility, ensuring continuity of care and remote monitoring. This is a consequence of COVID-19 pandemic, but it was already becoming particularly relevant for the management of a growing elder population and the higher incidence of chronic diseases. Exercise and specific rehabilitation programs are indeed known to lead to improved motor skills for various clinical conditions. Numerous efforts are being carried out, therefore, on the development of portable devices, suitable for home use, that exploit different technologies, such as cameras or wearable inertial measurement units (IMUs). In the literature, many studies can be found facing the challenge of exploiting IMUs data to provide clinically relevant information on patients health status. IMUs are being widely used in gait analysis, for example, with the goal of assessing gait in out-of-the-lab conditions and for its continuous monitoring. Another common application is the extraction of quantitative parameters during the execution of simple movements, often taken from clinical scales, to characterize patient motor capabilities and progresses. However, to foster adoptions of tele-rehabilitation solutions in clinical practice, there is still a need to assess patient performances and improvements from IMUs data while executing rehabilitation exercises commonly prescribed at home. Therefore, this thesis project aims at investigating inertial data from a wide set of exercises of variable complexity, included in a rehabilitation protocol proposed by the Neurorehabilitation Clinic, Ospedali Riuniti of Ancona, and widely adopted in clinical practice. The main objective is to explore a method and find out relevant parameters to support clinicians in the evaluation of remote exercise sessions. This project has been realized in collaboration with the research company Henesis, that made available two datasets: one on healthy volunteers and a second on pathological subjects, coming from a clinical trial. The inertial data were acquired using ARC intellicare, a medical device that allows motor and respiratory tele-rehabilitation. This thesis proposes a case study on 2 Parkinson and 2 Long COVID-19 patients: their data were processed and analyzed through a Python code, in order to find averaged movement patterns for each exercise. The patterns obtained from the patients have been compared to those from the healthy population, to look for any abnormalities or deviations. In addition, duration of exercises executions and a Dynamic Time Warping (DTW) score have been investigated to provide a complementary quantitative description. Average duration of repetitions for a subset of exercises have been computed and a DTW score, a distance metric used also in other studies, has been implemented to quantify similarity of patients' movement patterns to the obtained healthy reference. A comparison is presented between DTW scores computed on data from the first and last day of the rehabilitation program. Results show that this metric could potentially be used to analyse differences in motor performances of Parkinson and long COVID-19 patients. Future developments will include extension of the analysis to all patients and all exercises of the dataset. Correlation of results with clinical scales is then necessary to validate the proposed approach.