Keywords

lumbar spine, vertebral kinematics, wearable, skin strain, joint kinematics, cadaver test, supervised machine learning, electromagnetic tracking, low back pain

Abstract

Introduction: Joint kinematics have been shown to be strongly diagnostic of underlying pathology in the knee, hip, shoulder, and ankle. However, the anatomical complexity of the spine has made obtaining accurate, segmental kinematics extremely challenging. Current approaches required specialized hardware (e.g., a dual x-ray system), or are highly invasive (e.g., bone pins implanted into the vertebrae). Recently, our lab has developed a wearable array of skin-mounted strain sensors that hopes to address this challenge. The purpose of the present work study was to quantify the utility of this array through comparison with real-time lumbar vertebral kinematics.

Materials and Methods: A wearable array of 16 nanocomposite, wide-range, strain gauges with a kinesiology tape substrate was attached to the skin proximate to the lumbar spine of a cadaver specimen (Female, 79 y.o., BMI: 18.5). The cadaver specimen was positioned in standing, partially supported by a ceiling mounted harness, and manually manipulated in both primary and combined modes of bending. Bone pins were inserted into the spinous processes of each of the L1 – S1 vertebrae. A 6 DOF electromagnetic tracking system (trakSTAR) was attached each bone pine to provide real-time, correlated rigid body motion tracking for each vertebra. Electrical signals from each of the 16 strain sensors during 10 different exercises, including flexion (Figure 1), were recorded, processed with a 0.3 Hz low-pass filter, and subsequently plotted over time. A linear model of the spinal motion was created with sensor data as the input and vertebral kinematics data as the supervised training output, with a 70/30 split. The average RSME was calculated for each strain sensor and exercise.

Results and Discussion: The linear model for flexion-extension achieved a root-square mean error (RSME) of 9.1%, with the other combined flexion-related exercises (right flexion-extension, left flexion-extension) achieving RSME values less than 10%. Results for stand-alone axial rotation were less satisfactory, likely due to the skin laxity of the cadaveric specimen which was very high, potentially due to rapid weight loss prior to death. A significantly correlated change in sensor gain with strain was recorded for 6 of the 10 exercises, with 3 exercises also reporting a low RSME.

Conclusion: The signal from wearable strain sensor array was significant correlated with spine segment kinematics for most loading modes. Future work will improve both the correlation model, as well as the sensor array, allowing for more robust measurements in the context of axial rotation.

Acknowledgements: Research reported in this abstract was supported by NIAMS of the National Institutes of Health under award number: UH2AR076723. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Document Type

Poster

Publication Date

2021-10-07

Language

English

College

Ira A. Fulton College of Engineering

Department

Mechanical Engineering

University Standing at Time of Publication

Graduate Student

Download

Share

COinS