Figure 3

Scheme of the proposed probabilistic prediction model. (A) Given a query subject X _q  ∈ TE, a set of EMG signals $F_{\mathit{qj}},j=\overline{1,4}$ are obtained as a response to repeated electrical stimulation of ten sites on the sole of the foot. (B) A probability distribution histogram P _qj is constructed from each signal F _qj (or combination of signals from multiple sites) to be used as classification feature. (C) The signal F _qj is labelled p (for patient) or h (for healthy), depending on the distances d _qj to the closest neighbouring histograms P _i , derived from the set of training subjects {X _i  ∈ TR}. (D) The final prediction for the subject X _q is carried out based on the labels l _qj derived from the individual assessment of all four signals. Query subjects X _q  ∈ V were used instead for all validation procedures (site combination and training set selection).