A decision tree classifier for postural and movement conditions in Essential Tremor patients

This paper proposes a decision tree based classifier to discriminate between movement and postural conditions in Essential Tremor (ET) patients when their Deep Brain Stimulator (DBS) is switched OFF and they do not yet present tremor symptoms. This aims to be the first stage of a fully automated closed-loop ON-OFF DBS system in which the algorithm for prediction of tremor onset uses optimized parameters depending on the patient's postural or movement condition. The classifier inputs are the power of the surface-electromyogram (sEMG) and accelerometer (Acc) signals recorded at the symptomatic extremities of the patients. The proposed classification tree uses Gini splitting rule and an optimized pruning scheme. The classifier achieves an overall accuracy of 96.55% by correctly classifying 112 out of 116 trials in four ET patients: 49 trials were in the movement condition and 67 were in postural condition. A classification accuracy of 100.00% (49 trials out of 49) and 94.03% (63 trials out of 67) is achieved for movement and posture conditions, respectively.