Subject

One of the important factors contributing to an increased risk of falling is related to balance problems. There exists work that illustrates that a balance test on a force plate can discriminate between fallers and non-fallers (e.g. https://www.medrxiv.org/content/10.1101/2021.05.19.21257466v1.full.pdf) Such work either computes parameters related to postural sway and others from the force plate data and then trains a classifier with such features. Alternatively, the timeseries can just be fed directly into a deep learning topology. In that case the deep network is expected to learn the features and the decision boundary.

Force plates are typically available in gait labs only and are hence not very well suited for large screening or prevention programs. However, the Wii balance board used to be a low-cost widely available sensor with similar measuring capabilities. In the context of the Brussels Gerontopole, we collected longitudinal data of a large population of elderly.

Kind of work

You will use deep learning methods for time series classification to try to identify elderly with an increased risk of falling from a single Wii balance board recording. Besides fall risk, we will try to discriminate frail from pre-frail elderly and we will try to identify subjects that will develop frailty in the near future.

In a second phase, you will try to combine the balance board assessment collected every six months to investigate whether this temporal evolution of balance is a better predictor for frailty or fall risk compared to a single measurement.

Number of Students

1