Comparing algorithms for assessing upper limb use with inertial measurement units

Abstract

The various existing measures to quantify upper limb use from wrist-worn inertial measurement units can be grouped into three categories: 1) Thresholded activity counting, 2) Gross movement score and 3) machine learning. However, there is currently no direct comparison of all these measures on a single dataset. While machine learning is a promising approach to detecting upper limb use, there is currently no knowledge of the information used by machine learning measures and the data-related factors that influence their performance. The current study conducted a direct comparison of the 1) thresholded activity counting measures, 2) gross movement score,3) a hybrid activity counting and gross movement score measure (introduced in this study), and 4) machine learning measures for detecting upper-limb use, using previously collected data. Two additional analyses were also performed to understand the nature of the information used by machine learning measures and the influence of data on the performance of machine learning measures. The intra-subject random forest machine learning measure detected upper limb use more accurately than all other measures, confirming previous observations in the literature. Among the non-machine learning (or traditional) algorithms, the hybrid activity counting and gross movement score measure performed better than the other measures. Further analysis of the random forest measure revealed that this measure used information about the forearm’s orientation and amount of movement to detect upper limb use. The performance of machine learning measures was influenced by the types of movements and the proportion of functional data in the training/testing datasets. The study outcomes show that machine learning measures perform better than traditional measures and shed some light on how these methods detect upper-limb use. However, in the absence of annotated data for training machine learning measures, the hybrid activity counting and gross movement score measure presents a reasonable alternative. We believe this paper presents a step towards understanding and optimizing measures for upper limb use assessment using wearable sensors.

Measurement of Functional Use in Upper Extremity Prosthetic Devices Using Wearable Sensors and Machine Learning

“…For example, automatic detection of vestibular gait is possible with good accuracy if data is restricted to a specific task [ 51 ]. For upper extremity tasks, several studies have found that wrist-worn IMUs and Random Forest Classifiers are superior to other processing methods for detecting functional arm use in stroke [ 52 , 53 ]. Another group is working to provide a more detailed description of the upper extremity function, by first identifying five primitives of the upper extremity function: reaching, transporting, repositioning, stabilizing, and idling [ 54 ].…”

GMAC: A simple measure to quantify upper limb use from wrist-worn accelerometers

“…Nevertheless, several measures have been proposed in the literature to detect UL use from a single IMU [5, 7, 8, 10]. These measures can be broadly categorized into traditional [5, 10, 11] and machine learning(ML)-based measures [6,10]; we use the terms measure, model, and algorithm interchangeably in the rest of the manuscript. The traditional measures are simple, hand-crafted algorithms with pre-specified parameter values that use specific signal features to detect UL use.…”

“…On the contrary, ML-based measures are algorithms trained on a set of labeled data to detect UL use from IMUs. Random forests, support vector machines, and multilayer perceptrons have been reported previously [6, 10], with the random forests [6, 10] offering the best performance to date. Additionally, intra-subject (i.e.…”

“…The TAC and the GM measures are the two most popular measures for quantifying UL use. Previous studies have shown that the TAC is a highly sensitive measure, while the GM is highly specific [10, 12]. We recently proposed a hybrid measure, called the GMAC , that combines the TAC and GM measures to balance out their respective high sensitivity and specificity [10].…”

“…Previous studies have shown that the TAC is a highly sensitive measure, while the GM is highly specific [10, 12]. We recently proposed a hybrid measure, called the GMAC , that combines the TAC and GM measures to balance out their respective high sensitivity and specificity [10]. The GMAC showed a better overall performance than TAC or GM, as quantified using the Youden index, but had a lower performance than the inter- and intra-subject ML measures.…”