Early recognition of upper limb motor tasks through accelerometers: real-time implementation of a DTW-based algorithm Rossana Muscillo, Maurizio Schmid n , Silvia Conforto, Tommaso D’Alessio Applied Electronics Department, Roma Tre University, Rome, Italy article info Article history: Received 18 February 2010 Accepted 15 January 2011 Keywords: Classification Dynamic Time Warping Motor function evaluation Inertial sensors abstract A new real-time implementation of a Dynamic Time Warping (DTW)-based classification scheme is presented here, and its performance evaluated on experimental data. Nine young adults were requested to perform instances of eight different purposeful movements described in the Wolf Motor Function Test, while wearing a three-axis accelerometer sensor placed on the inner forearm. Results include the correct recognition percentage, as compared to a classification scheme based on the traditional DTW measure, and the recognition percentage as a function of the time elapsed from the beginning of the performed movements. The Real-Time DTW basically performs with the same accuracy of the traditional DTW-based classification scheme (91.5% of correct recognition percentage), a figure that increases to 96.5% if the multidimensional scheme is adopted. Moreover, more than 60% of movements are correctly recognized before their end, thus setting the way for applications in rehabilitation and assistive technologies, where a real-time control scheme is able to interact with the user while the movement is being performed. & 2011 Elsevier Ltd. All rights reserved. 1. Introduction Stroke is the leading cause of long-term disability worldwide [1]. The recovery of motor function in post-stroke patients is thus a pivotal factor in rehabilitation of post-stroke patients. It is acknowl- edged that motor performance is one of the greatest predictors of motor recovery in stroke survivors [2]. With the aim of assessing motor function recovery in stroke survivors, different scales have thus been introduced, able to document the efficacy of rehabilitation programs [3,4]: most of them usually entail, among others, upper and lower limb activities, which are administered in a multi-item fashion, with marks associated with the extent of success in executing each single task, and/or with timings needed to perform them. A subset of these test sections generally refers to tasks that are commonly performed during daily living activities: one of the advantages of this approach over single or multiple-joint non-functional tasks is that they can capture the amount of disability that is experienced by each subject when facing real-life activities. Among them, the Wolf Motor Function Test is a complete set of 15 different functional tasks that the patient is requested to perform while seating in front of a desk. It specifically targets upper limb functional tasks, and both time to perform each task and quality of movement are extracted and classified according to a pre-determined scale. Even if this kind of tests is usually administered in a hospital facility, the increase of home-based programs for rehabilitation, promoted by the advantages of de-hospitalization both in eco- nomical and in clinical effectiveness terms, makes it possible to envision technological platforms able to run tests that can be performed autonomously by the patient in her/his home setting. In order to achieve this, two elements are thus needed: 1. Minimally intrusive devices able to capture kinematics when performing the tasks; 2. Unsupervised methods able to early recognize among the different tasks being performed and, possi- bly, to predict the scores that would otherwise be subjectively determined by the rehabilitation staff. In regards to the technological aspect, different possibilities are now at hand: camera-based motion capture is a valid tool to provide patients with an artificial environment for rehabilita- tion [5], even if the kinematics that can be extracted with this kind of technology may not be as reliable as needed for the evaluation; sensing fabrics are also a useful tool to gather this kind of kinematics as well, and they have been profitably introduced in rehabilitation settings [6], but the data can suffer from temporal drifts and hysteresis [7]. The inertial sensor technology is the key technology: it has been proven as a reliable tool in home-based rehabilitation for different pathologies, ranging from stroke [8] to cerebral palsy [9], and it has already been tested in this scenario. As far as the methodological aspects are concerned, while the topic regarding the prediction of scores is a well-studied process in stroke rehabilitation [10], the issue related to distinguishing among classes corresponding to different exercises can be effi- ciently tackled if real-time classification methods are at hand. The availability of a real-time classifier for this kind of applications is Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/cbm Computers in Biology and Medicine 0010-4825/$ - see front matter & 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.compbiomed.2011.01.007 n Corresponding author. Tel.: + 39 06 5733 7309; fax: + 39 06 5733 7026. E-mail address: schmid@uniroma3.it (M. Schmid). Computers in Biology and Medicine 41 (2011) 164–172