Parsimonious Modeling for Kinematic Data

• Activity Number: 423 - SPEED: Biopharmaceutical Statistics, Medical Devices, and Mental Health
• Type: Contributed
• Date/Time: Tuesday, August 1, 2017 : 2:00 PM to 2:45 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Md Islam* and Jonathan W. Stallings
• Companies: North Carolina State University and North Carolina State University
• Keywords: Functional model ; Historical functional covariate ; FLiRTI ; Group-lasso ; Kinematic data ; EMG

Abstract

Functional linear regression is a popular modeling approach but is prone to overfitting and often yields estimates that are difficult to interpret. To improve interpretability, James et al (2009) introduced the FLiRTI (functional linear regression that's interpretable) model. We extend FLiRTI approach to multiple functional covariates and introduce smooth regression functions approximated by a tensor product of basis. Our study is motivated by the work with biomedical engineers at North Carolina State University, who are interested in assessing the systematic association between electromyography signals measuring muscle contractions and hand-wrist movements across different postures of able-bodied subjects. The primary goal is to identify the influential signals and implement a predictive model in developing robotic prosthetics to project the intended movements of an amputee. Our proposed approach consists of two steps: determine influential covariate(s) using group-lasso and fit the generalized FLiRTI model on the selected covariate(s). An extensive simulation study suggests excellent numerical performance in terms of interpretation and identification of the pattern of movements.