Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session B

Poster A6 - DECODING HAND MOVEMENTS FROM HEALTHY ELECTROMYOGRAPHIC SIGNALS TOWARD A TRANSFERABLE NEAR-NATURAL PROSTHESIS

Thursday, October 24, 2024

2:45 PM - 3:45 PM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

Caly Ferguson

Undergraduate Researcher
University of Maryland, Baltimore County, United States

Co-Author(s)

DP

Dinyi Pei

PhD Candidate
University of Maryland, Baltimore County, United States
PO

Parthan Olikkal, MS

PhD Candidate
University of Maryland, Baltimore County
Arbutus, Maryland, United States
Zainab Idowu

Undergraduate Researcher
University of Maryland, Baltimore County, United States

Last Author(s)

RV

Ramana Vinjamuri

Principal Investigator
University of Maryland, Baltimore County, United States

Introduction: Stroke, injuries, diabetes, cardiovascular diseases, and amniotic band syndrome are a few unfortunate situations that have caused approximately 50 million people worldwide to have a limb amputated. Over the years, new technological advancements have allowed missing limbs to be replaced by mechanical prosthetic limbs that act as assistive devices but do not offer the full range of motion and sophistication to make one’s day-to-day life natural. The specific objective of this project is to create a computational neuroscience model using machine learning that can synthesize non-amputated individuals’ electromyography (EMG) signals from forearm muscle groups, and predict diverse hand movements based on these signals as a transferable method of predicting amputated individuals’ desired hand movements.

Materials and

Methods: The raw data for this research came from an online repository that collected and labeled EMG signals from non-amputated individuals performing different hand gestures using EMG sensors that span the circumference of their forearms. These recordings were processed using a combination of root-mean-square and moving average smoothing methods to eliminate the noise from the raw data. Then the Classification Learner Application within MATLAB was used to simultaneously train multiple machine learning models that can effectively differentiate the differences in EMG signal amplitudes from each sensor on the recording area from the fist hand gesture and each of a healthy individual's finger flexions. With this, the combined activations of different muscles in the forearm, called synergies, can be used to decode specific hand gestures.

Results, Conclusions, and Discussions: After training and testing, the most accurate model with a 99.2% testing prediction accuracy, used Linear Discriminant Analysis to differentiate dimensionality between the fist and each of the different finger flexion signals. The Hiwonder Robotic Hand was coded using MATLAB's Arduino Support Package feature to actuate the fingers to a predicted position based on the classification model to simulate how a robotic prosthetic would operate. The next step of this research project is to apply this model to a real-time simulation to test if it can make real-time predictions on the classified movements of a healthy individual. After this, the model will be tested on an amputated individual, aided by a visual cue like virtual reality or a video to see if the model is transferable to people without a hand. Currently, the model cannot predict positions between full flexion of the fingers. Still, future aims are to improve the versatility of this model so that it can actuate more complex hand positions, and apply this model to a 3D printed prototype myoelectric prosthetic to test how it can improve an amputated individual’s life.

Acknowledgements (Optional): A special thanks to the National Science Foundation (CAREER Award HCC-2053498), the Meyerhoff Scholars Program at UMBC, the UMBC HHMI Undergraduate Scholars Program, the Department of Computer Science and Electrical Engineering at UMBC, the Vinjamuri Lab and Dr. Ramana Kumar Vinjamuri for supporting this research. Additionally, this research was supported in part by a grant to UMBC from the Howard Hughes Medical Institute through the Driving Change Initiative, the USM Louis Stokes Alliances for Minority Participation Program, and NIH’s Undergraduate Research Training Initiative for Student Enhancement.