Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Biomechanics

Biomechanics - Poster Session B

Poster E3 - Effects of task and sensor location on tremor identification in the upper extremities

Thursday, October 24, 2024
2:45 PM - 3:45 PM EST
Location: Exhibit Hall E, F & G

Introduction:  Parkinson’s Disease (PD) is a disease often characterized by the presence of tremor, a rhythmical involuntary oscillation. Tremors can be broken down into 2 categories: rest tremor and essential tremor. Rest tremor is more prevalent when the individual is stationary, and tends to present itself asymmetrically across the body with a lower frequency of approximately 3-6 Hz, while essential tremor tends to be elicited from voluntary actions and has a higher frequency of around 5-12 Hz [1]. The presence of rest tremor is often used in conjunction with other symptoms to diagnose PD. Currently, tremor symptoms are assessed using the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS), where a trained neurologist uses a 5 point scale to assess the severity of tremor through visual observation; however, this test is subject to rater error and inter-rater variability due to the reliance on human observation [4]; thus, wearable sensors are being considered for tremor assessment. Previous research on wearable accelerometers found them to be generally applicable in regards to detecting tremor, but there is a lack of standardization on data collection and analytical protocols [6]. This study aims to investigate the impact of task (with/without distractor), sensor location (finger or wrist), and sensor side (left or right) on objective tremor characteristics when compared to neurologist assigned tremor ratings using the UPDRS. Previous findings using the UPDRS found that mental load [3] and dominant side [5] had significant impacts on tremor, so it can be expected that wearable sensors should produce similar results.

Materials and

Methods:  Tri-axial accelerometers were attached using tape on the wrist and index finger of thirty participants, who were rated according to the UPDRS while performing resting tremor upper (RTU) and resting tremor upper counting (RTUC) tasks [2]. These tasks were chosen as they keep the limbs stationary, thus eliciting rest tremor. The RTUC task consists of a similar protocol as the RTU task but with an added distractor for the participant in the form of counting numbers out loud. Accelerometer values were converted to resultant jerk to control for the unknown rotational alignment of the sensor. The derivative of each axis was obtained, then the norm of all three axes was calculated for a resultant vector. Using Python, the power spectral density (PSD) of the jerk data was then taken to obtain mean amplitude for the two frequency bins of interest: 3-6 Hz and 5-12 Hz, or rest and essential tremor respectively (Baumann C.R. 2012, S90). Tremor ratings were converted to a binary classification, where 0 was classified as tremor absent while 1-4 was classified as tremor present. A wilcoxon ranked test was performed on the data in these bins, comparing the frequency bin values obtained with the new binary tremor rating assigned for the upper limb tasks. Mean PSD frequency bin amplitude was then analyzed using a linear mixed effects model to determine the effects of other metrics recorded (location, side, task, binary tremor classification).

Results, Conclusions, and Discussions: Results
The significant results of the unpaired wilcoxon tests are presented in Table 1. Unpaired wilcoxon ranked tests revealed significant differences in jerk tremor amplitude between tremor present and tremor absent rating in both rest and essential tremor frequency bands. Significant differences in tremor were found in both tasks for both the wrist and finger sensor, but only on the left side. In Table 2, linear mixed effect analysis revealed that the presence of a distractor had a significant impact on jerk tremor frequency (p < 0.05) and a possible significant interaction between the side of the body in which the sensor was located and the assigned rating was found in both rest and essential tremor in Table 2. No other effects were found to be statistically significant.

Discussion:
Currently, few studies address the impact of factors known to impact PD tremor using wearable sensors. This study examined the impact of factors such as sensor location, side, and task on rest and essential tremor when compared to tremor rating from the MDS-UPDRS, and found that the task performed and sensor side can impact rest tremor. Tremor in PD was found to be asymmetric in nature in [5], generally occurring on the side of dominance; while the wearable sensor data agrees that tremor tends to be asymmetric, we lack data to confirm if this is associated with hand dominance or site of earliest PD symptom onset. In comparison to the findings in [5] that were obtained using a larger sample size (n=134) and only the UPDRS, this study used a smaller sample while wearing wearable sensors, which may provide a more sensitive measure of tremor changes. Lastly, the presence of a distractor was found to have a significant impact on tremor amplitude, similar to the findings in [3]. Future studies should investigate further the impact of hand dominance on tremor between sensor and UPDRS outcomes, as well as integrating task and sensor location into machine learning classifiers for tremor identification and monitoring.

Acknowledgements (Optional):  Acknowledgements:
I would like to thank Dr. James Brasic at New York University for insights on PD, as well as Dr. Bodo Winter at the University of Birmingham for his guides on linear mixed effects models in R.