s of the SIAMOC 2008 Congress / Gait & Posture 29S (2009) e1–e31 e22 Cluster analysis: A useful data reduction technique in simultaneous kinematics–kinetics–plantar pressure analysis Z. Sawacha *, A. Guiotto , G. Guarneri , G. Cristoferi , A. Avogaro , C. Cobelli 1 Department of Information Engineering, University of Padova, Padova, Italy Department of Clinical Medicine & Metabolic Disease, University Polyclinic, Padova, Italy Introduction: To accurately evaluate the extent of diabetics gait deviations from normal gait, it is important to consider not only how each feature of the gait pattern has changed but also how the relationship between the features changed. In order to improve diabetic foot prevention treatments it is important to establish the correlation that exists between gait variables and foot biomechanics alterations. Least but not less important, the large amount of variables obtained in kinematics–kinetics and plantar pressure (PP) analysis need to be reduced in order to be clinically useful. Thus, in the light of these premises, the aims of this studywill be to apply cluster analysis either to agglomerate the patients with common gait deviations in the same cluster or to find which gait parameters characterizes the pathology. So far patients can be classified relatively to their gait pattern alterations and specific prevention protocols developed. Methods: Kinematics, kinetics and PP data of 35 subjects (14 normal (C), 9 diabetics (D), 12 diabetic neuropathics (DN)) have been collected by means of 6 cameras BTS Sr.l. motion capture system (60–120 Hz) synchronized with 2 Bertec force plates (FP4060-10) and 2 plantar pressure plates (Imago). A 3D foot markerset was used and gait and posture analysis performed [1]. Five homogeneous groups were created, in order to compare subjects with the same type of foot: cavus foot (CF), CF and valgus (V) heel, CF and normal heel, CF and normal hallux, CF and V hallux. 3D subsegments forces, plantar pressure, contact surface and joint rotation angles were calculated [1] and each group normative bands (nb) created (mean 1SD). COP pattern was analyzed in both anteroposterior (AP) and mediolateral direction and its integral calculated. Hierarchical and k-mean cluster analysis were performed: each one starting with all of the observations in one cluster and then proceeding to split them into smaller clusters. Results: k-Means cluster analysis [2] was successfully used as a data reduction method. Each group significant variables were identified as the ones that successfully agglomerate the same type of subjects (normal, diabetics, neuropathics) relatively to the same type of foot carachteristics. Concerning the kinematics variables these were: midfoot-hindfoot (M-H) prono-supination and abdadduction (AA) together with forefoot-midfoot (F-M) AA in cavus foot and hallux valgus groups, F-M dorsi-plantarflexion (DP) in cavus foot and valgus hallux groups, M-H DP in cavus foot group. Concerning kinetics and PP variables these were: the F contact surface, Mmean and peak pressure, F vertical, anteroposterior and mediolateral force in cavus foot and hallux valgus groups (Fig. 1). Fig. 1. The bar plot displays the percentage of subjects (blue C, red D, green DN) belonging to each group that have been agglomerated in the same cluster by the specific gait parameter, respectively: midfoot-hindfoot prono-supination angle (1), forefoot-midfoot dorsi-plantarflexion angle (2), forefoot-midfoot abductionadduction angle (3), during one gait cycle (kinematics analysis), forefoot contact surface (4) during stance phase (PP). Discussion: Cluster analysis allowed to classify subjects according to their gait alterations, and agglomerated correctly each group specific gait parameters. Therefore this variables subdivision can be used in planning prevention treatments. A larger sample of subject is being investigated.
Read full abstract