Machine learning for prediction of walking abilities
in incomplete spinal cord injured patients
T. Bajd, M. Grobelnik, D. Mladenic, N. Lavrac,
V. Prodnik, H. Benko, R. Savrinc, P. Obreza
It is difficult to predict the outcome of the functional electrical
stimulation (FES) rehabilitation process when patients are admitted
to the spinal unit soon after an accident that caused incomplete
spinal cord injury (SCI). Similarly, it is not possible to decide what
rehabilitation aid the patient will need after recovering from
a spinal cord injury.
The aim of this study is to develop a diagnostic procedure which
will soon after the accident predict which incomplete SCI patients
are candidates for a permanent use of FES orthotic aid.
Based on data about 31 incomplete SCI patients, a classification tree
was developed using a machine learning tool CART.
The induced classification tree indicates that the candidates
for chronic use of FES are patients with weak ankle dorsiflexors
and sufficiently strong knee extensors.
It was further demonstrated that FES is less functional
in the incomplete SCI patients who are bound to wheelchair.