RECURRENT MAP ANALYSIS OF DIGITAL SPIRAL DRAWING SIGNAL TO INVESTIGATE FINE MOTOR MOVEMENT IN MULTIPLE SCLEROSIS PATIENTS AND CONTROLS

Abstract

Recently, recurrent plot (RP) has been used as one of the analysis tools in complex system dynamics. In this paper, we hypothesize that complex features extracted from RP have superiority in discriminating the upper extremity performance in two groups of mulitiple sclerosis (MS) patients without tremor and healthy controls compared to statistical and power spectrum features. We define spiral drawing task for upper extremity and the position signals is recorded from subjects. Then, velocity profiles are extracted and the common statistical and spectral features are exported. To extract complex features from RP, a modified methodological approach based on density distribution is presented and the properties of distribution are calculated as complex features. Finally, the applicability and capabilities of these three groups of features are invested by a Neuro-Fuzzy Classifier. The performance of the Neuro-Fuzzy classifier is reported as sensitivity, specificity and accuracy criteria. The results of the analysis yield out that complex features have the highest performance comparatively. This hypothesis is proven and validated through the experiments and it is shown that the complex features have promising discriminating capabilities. To validate the classifier used, different structures of the neuro-fuzzy classifier are studied in terms of the number of membership functions and the type of fuzzy sets and the most efficient structure is extracted out. Furthermore, the efficiency of the Neuro-fuzzy classifier with its optimum structure and tuned parameters is compared with some other well-known and commonly used classifiers.