Predictive medicine : walking test outcome prediction prior to hip and knee surgery

Details

Supervisors

Absil, Pierre-Antoine

Degree label

Master [120] : ingénieur civil en science des données, à finalité spécialisée

Abstract

Smart connected objects including phones, tablets or watches have seen a fast-growing interest in the last years and share nowadays a significant part of our environment. The huge amount of data collected by these devices are various and follow many different purposes. Among them, the number of steps per minute is a particular type of data usually used for sport and physical training. However, this thesis aims to exploit it for medical follow-up of patients who have undergone hip or knee replacement surgery, in close collaboration with moveUP, a Belgian company developing an application for that specific objective. Firstly, different features were extracted and derived from the steps per minute variable such as the average step accumulation and the 6 minutes walking test, among others. Thereafter, a dashboard including the engineered data on an interface has been developed to allow physiotherapists the monitoring of their patients’ performances as well as to provide different statistics. A second phase of the presented work consisted in building a prediction model on the adapted 6 minutes walking test, considered as a performance indicator by the health specialists, in the sixth weeks following the surgery using all historical patient’s data (age, gender, BMI, etc.) in order to improve their rehabilitation. To that end, a classification model has been built by converting the quantitative data of steps per minute into 4 or 2 cadence categories. After investigation, the best model was binary and led to 65% accuracy. Further work could be to use more data, i.e. variables derived from the steps per minute or extracted from the moveUP database, to improve the model. An alternative could further consist in performing an unsupervised learning and verify the existence of any clusters emerging from patterns within the patients and to compare them with the classes used for cadences in this work.