||According to World Health Organization, on average, one person died from stroke every six seconds. For people who survive the stroke, their quality of life is often compromised by its after-effects. Rehabilitation is thus a critical part of stroke recovery. However, not all rehabilitation methods are equally suitable for every patient. To maximize the cost-effectiveness of stroke rehabilitation, this study uses five demographic features (age, onset time, MMSE, education level and gender) and a number of scales to predict if the scores of these scales can be improved by rehabilitation.|
This study recruits 130 subjects. The conditions before and after the rehabilitation of these stroke patients were evaluated by using eight different scales. Since some of these scales can be divided into several sub-scales, we obtain 13 scale scores for each patient.
For each of the tested scales, depending on whether the score improved or not, patients are divided into success and failure groups. Via statistical tests, nearest neighbor and neural network classification methods, we have built and tested a number of classifiers to distinguish these two group of patients.
Experiment results show that the classification accuracy varies with the features and methods employed for building the classifiers. In predicting the success of a particular scale, the best result of in this study was obtained by using the scores of that scale and SIS hand function before the rehabilitation as well as age and onset time. The resulting accuracy is 0.684, Kappa Index is 0.355, PPV is 0.791, NPV is 0.560, sensitivity is 0.678, and specificity is 0.695.