A Wavelet-Based Approach for Screening Falls Risk in the Elderly using Support Vector Machines

Abstract

Trip related falls are a prevalent and costly threat to the elderly. Early identification of at-risk gait helps prevent falls and injuries. The main aim of this study is to explore the effectiveness of a wavelet based multiscale analysis of a gait variable [minimum foot clearance (MFC)] in extracting features for developing a model using Support Vector Machines (SVM) for automated detection of balance impairment and estimation of the falls risk in the elderly. MFC during continuous walking on a treadmill was recorded on 11 healthy elderly and 10 elderly with balance problems (falls risk) and with a history of tripping falls. The multiscale exponents (beta) between successive wavelet (Wv) coefficient levels after Wnu decomposition of MFC series (512 points) into eight levels from level 2 (Wnu2) to level 256 (Wnu256), were calculated for healthy as well as falls-risk elderly adults. Using receiver operating characteristic (ROC) analysis, the most powerful predictor variable was found to be betaWnu16-Wnu8 (ROCarea = 1.0), followed by betaWnu16-Wnu8 (ROCarea = 0.92). These multiscale exponents were used as inputs to the SVM model to develop relationships between the intrinsic characteristics of gait control and the healthy/falls-risk category. The leave one out technique was utilized for optimal tuning and testing of the SVM model. The maximum accuracy was found to be 100% using a polynomial kernel (d = 4) with C = 10 and the maximum ROC = 1.0, when the SVM model was used to diagnose gait area patterns of healthy and falls risk elderly subjects. For relative risk estimation of all subjects, posterior probabilities of SVM outputs were calculated. In conclusion, these results suggest considerable potential for SVM gait recognition model based on multiscale wavelet features in the detection of gait changes in older adults due to balance impairments and falling behavior. These preliminary results are also encouraging and could be useful not only in the falls risk diagnostic applications but also for evaluating the need for referral for falls prevention intervention (e.g., exercise program to improve balance).