Classification of normal and pathological tremors using a multidimensionalelectromagnetic system

A new multidimensional movement analysis system was used to record limb tre mor over six degrees-of-freedom, and signal processing techniques were expl ored to develop a suitable classification method to distinguish between dif ferent types of tremor. The specific aims were to investigate the ability o f the system to screen for differences between normal subjects and a group of neurological patients, and then to differentiate between three diagnosti c groups of patients. Postural tremor at the hand was recorded in normal subjects (n = 24) and pa tients with essential tremor (n = 21), multiple sclerosis (n = 17) and park insonism (n = 19). Data were collected using a 3Space Fastrak(R) (Polhemus, Inc.) over six degrees-of-freedom (three translational directions and thre e rotations). Spectral estimates produced measures of tremor frequency and amplitude. Mathematical models of the data, using autoregressive modelling and K-nearest neighbour classification, produced parameters used to classif y, (1) the normal subjects and 24 patients (using the three rotational move ments), and (2) the three patient groups (using all six movement directions ). Results were given in terms of the probability of each subject belonging to the groups being classified. Tremor frequency and amplitude showed large overlap between the groups. The screening classification produced high probabilities of correctly classify ing normal subjects (>70%) and patients (>70%). The diagnostic classificati on produced clear differences between the patient groups (60% for essential tremor, 80% for multiple sclerosis and 60% for parkinsonism). The ability of this assessment technique to distinguish between postural tr emor in normal subjects and neurological patients suggests that it could be developed as a screening tool. Classification of tremors between the patie nts groups, with a high degree of sensitivity, indicates the potential for further development of the system as a diagnostic aid. (C) 2000 IPEM. Publi shed by Elsevier Science Ltd. All rights reserved.