Automatically evaluating balance using machine learning and data from a single inertial measurement unit

Anything longer than two months for your hospital to implement this and your board of directors needs to be fired. If your hospital doesn't even have EXACT FALL PREVENTION PROTOCOLS, that's even worse, reconstitute the whole hospital. Dead wood needs to be removed.

Automatically evaluating balance using machine learning and data from a single inertial measurement unit

• Fahad Kamran,
• Kathryn Harrold,
• Jonathan Zwier,
• Wendy Carender,
• Tian Bao,
• Kathleen H. Sienko &
• Jenna Wiens 

Journal of NeuroEngineering and Rehabilitation volume 18, Article number: 114 (2021) Cite this article

• Metrics details

Abstract

Background

Recently, machine learning techniques have been applied to data collected from inertial measurement units to automatically assess balance, but rely on hand-engineered features. We explore the utility of machine learning to automatically extract important features from inertial measurement unit data for balance assessment.

Findings

Ten participants with balance concerns performed multiple balance exercises in a laboratory setting while wearing an inertial measurement unit on their lower back. Physical therapists watched video recordings of participants performing the exercises and rated balance on a 5-point scale. We trained machine learning models using different representations of the unprocessed inertial measurement unit data to estimate physical therapist ratings. On a held-out test set, we compared these learned models to one another, to participants’ self-assessments of balance, and to models trained using hand-engineered features. Utilizing the unprocessed kinematic data from the inertial measurement unit provided significant improvements over both self-assessments and models using hand-engineered features (AUROC of 0.806 vs. 0.768, 0.665).

Conclusions

Unprocessed data from an inertial measurement unit used as input to a machine learning model produced accurate estimates of balance performance. The ability to learn from unprocessed data presents a potentially generalizable approach for assessing balance without the need for labor-intensive feature engineering, while maintaining comparable model performance.

Introduction

Balance training leverages the ability of the central nervous system to “reweight” functioning sensory inputs to compensate for sensory loss [1]. Balance exercises have been shown to be effective for preventing falls in at-risk individuals with balance concerns, such as older adults and individuals with vestibular deficits [1,2,3,4]. But, the most effective programs require the supervision of a physical therapist (PT) [5,6,7]. Without direct supervision from a PT, remote supervision scenarios (e.g., home-based balance training) provides limited benefits [8, 9]. In such scenarios, PTs typically provide individuals with paper-based instructions along with guidance regarding how often to perform the exercises [10]. Progression through a home-based balance training program is commonly informed by both an individual’s self assessment of their performance in addition to in-person evaluation during clinic-based training sessions [11,12,13]. However, the lag in real-time feedback associated with this approach could potentially hinder rehabilitation progress and negatively affect rehabilitation outcomes. Moreover, self-assessments may be inaccurate relative to PTs’ assessments, resulting in suboptimal or even unsafe exercise training [14, 15].

To improve the effectiveness of remote rehabilitation training, automated techniques for assessing balance that do not rely on self-assessments or PT supervision are needed. Recently, researchers have investigated the utility of machine learning (ML) tools applied to data collected from inertial measurement units (IMUs) for automatic balance assessment [16]. Such tools have the potential to support quick and accurate estimates of balance performance, thereby improving in-home balance training outcomes and maintaining safe exercising practices. Wearable sensors that measure motion throughout an individual’s daily routine provide a unique opportunity to detect, assess, and evaluate body movement [17, 18]. Like Bao et al. [16], we used data from a single IMU capturing trunk sway data to automatically assess balance. However, in contrast to Bao et al. who relied on an approach that required hand-engineered features, we explored the utility of ML approaches that leveraged unprocessed data. We hypothesized that ML models that leveraged the spatial information during a balance exercise would lead to more accurate assessments of balance compared to other representations. Overall, we focused on the development of methods for providing real-time assessments of balance that were more accurate than self-assessments to provide real-time feedback and improve the effectiveness of remote rehabilitation training methods.