Rule based behavior classification via accelerometry in transtibial amputees

Abstract

A high-performing and comfortable prosthesis follows from an understanding of how amputee activity modulates residual limb volume and socket fit. There are currently very few systems available which capture general long-term amputee behavior and so a computer algorithm for categorizing activity data is developed and validated. An accelerometer is fixed to the prosthesis pylon and reliably captures long-term natural transtibial activity. Data are downloaded and processed in a rule-based decision tree algorithm wherein regions are categorized into human behavior models, called schema, representative of a wide range of activities. After validation and optimization, algorithm overall effectiveness was graded at 92.36% with 100.00% active/passive activity differentiation. Through examining the activity of transtibial amputees and their prosthesis relationship made possible by this work, improvements can be made in prosthesis treatments, design, materials, and adaptive technology that will result in a better fitting, more secure, and empowering prosthesis.