Implementing Pattern Recognition and Matching techniques to automatically detect standardized functional tests from wearable technology

Abstract

Wearable sensor technology is often used in healthcare environments for monitoring, diagnosis and recovery of patients. Wearable sensors can be used to detect movement throughout measurement of standardized functional tests, which are considered part of the assessment criteria for Activities of Daily Living (ADL). The volume of data collected by sensors for long term assessment of ambulatory movement can be very large in tuple size since they may contain detailed 3-D sensor information. Extracting recorded movement data corresponding to standardized functional tests from an entire data set is complex and time consuming. This paper examines whether standardized functional tests can be automatically detected from long term data collected by wearable technology devices using Artificial Intelligence (AI) techniques. The current research work is aligned with clinical trial data generated by patients who are suffering from Axial Spondylo Arthritis (axSpA). These datasets contain Inertial Measurement Unit (IMU) values corresponding to individual patient functional tests for axSpA. Rotation angles with respect to each functional test are plotted against time. Individual movements that form part of a functional test are constructed for training and testing the AI system. Individual movement patterns are split into training and testing data inputs and are used to train the Neural Network (NN) system and to estimate overall prediction accuracy of the NN system. NN model is trained in such a way that the learned system can predict new functional test patterns with respect to the trained data and it is compared with expected data set and returned the accuracy of prediction. Once the semi supervised learning phase of AI system has successfully finished with adequate amount of data, it is capable for automatically detect gait and posture changes of patients at home.