Methodological approach in prediction of balance with machine learning applied on fMRI data

Machine learning is a useful method to assess biomedical data. However, due to the complexity of these data, it is important to rely on a robust methodology. Predictions with machine learning are not only an add-on to statistical analysis but are a real process of discovering and establishing clinical validity of the diagnostic techniques. This paper describes a machine learning approach to predict balance changes in seniors based on functional magnetic resonance imaging (fMRI) data. Functional MRI results in a huge amount data containing some noise. Thus, a noise filtering technique is proposed and subsequent analyses are conducted with various machine learning algorithms. The best prediction of the balance ability based on brain activation patterns was obtained by applying a neural network approach. This resulted in an area under the curve (AUC) of 0.91. Other learning algorithms such as the tree ensemble are less predictive but allow tracking and better understanding of the source of differentiation. The results are based on a real dataset.