Recurrent network based automatic detection of chronic pain protective behavior using MoCap and sEMG data

Abstract

In chronic pain physical rehabilitation, physiotherapists adapt exercise sessions according to the movement behavior of patients. As rehabilitation moves beyond clinical sessions, technology is needed to similarly assess movement behaviors and provide such personalized support. In this paper, as a first step, we investigate automatic detection of protective behavior (movement behavior due to pain-related fear or pain) based on wearable motion capture and electromyography sensor data. We investigate two recurrent networks (RNN) referred to as stacked-LSTM and dual-stream LSTM, which we compare with related deep learning (DL) architectures. We further explore data augmentation techniques and additionally analyze the impact of segmentation window lengths on detection performance. The leading performance of 0.815 mean F1 score achieved by stacked-LSTM provides important grounding for the development of wearable technology to support chronic pain physical rehabilitation during daily activities.

Sensor Technologies to Manage the Physiological Traits of Chronic Pain: A Review

“…Another way of identifying the presence of pain is through movements and physical activity, which are essentially different in subjects with chronic pain [25]. It is commonly considered that people who feel impaired and report restrictions in daily-life due to pain are less physically active [123][124][125]. People with chronic pain of physical origin adopt specific behaviors in response to pain, or when this is expected to occur, to protect the body [11,124].…”

Learning Temporal and Bodily Attention in Protective Movement Behavior Detection

“…Their study suggests the applicability of LSTM networks for the detection of protective behavior, based on MoCap and sEMG data. An obvious limitation of both [6] and [15] is that they considered all body-parts with equal importance across time and activity whereas protective behavior may occur in a specific stage of the activity and involve only a specific set of body parts, possibly different stages and body parts across activity types and across the CP population. Hence, by processing the full-body MoCap (and sEMG) data in a traversal manner, redundant information and less informative data are retained, possibly constituting noise and thus reducing the performance of the model.…”

“…Second, we are keen to understand what such modality alone (without sEMG) can achieve using attention mechanisms with respect to the state of the art. Still, as previous studies [15] [33] have shown that sEMG data are critical for high performance, future work should explore the combination of the two types of data.…”

“…Recorded activities included: One-leg-stand, Stand-to-sit, Sit-to-stand, Reach-forward and Bend -typical everyday activities that are generally challenging for those with CLBP [7] [9]. Previous studies based on Mocap and sEMG data of the EmoPain dataset mainly employed vanilla neural networks [15] and feature engineering approaches [6] [13] [14], where the dynamic biomechanics of movements are only used to guide feature design. Unlike acute rehabilitation where a goldstandard movement trajectory (and its deviation) informs intervention, in chronic pain, fear of injury, fear of pain, and anxiety lead the person to engage body parts in ways that are not biomechanically necessary or efficient but may increase sense of control and reduce fear.…”

“…performed by a subject during one trial (each participants underwent two trials of two difficulty levels, for details see [6]). Given that there are five types of protective behaviors (hesitation, guarding, stiffness, bracing and support) [6] but limited volume of the full dataset, we use the experience learned from [15] about treating all five behaviors as one single category called protective behavior. The input of BANet are local joint angles and their energies (the square of the angular velocity), with each angle calculated from three relevant anatomical points (e.g., the knee angle formed by joining joints on the neck, knee and ankle) as shown in Fig.…”