AI BASED MUSCLE ACTIVATION PATTERNS IN DAILY GRASPING MOVEMENTS FROM EMG DATA

Abstract

Electromyography (EMG) data gives a useful
insight into muscle activation patterns during
everyday grasping activities. This work seeks
to create interpretable and actionable machine
learning models for analyzing EMG data in
order to better understand muscle activation
patterns during these motions. We gathered
and analyzed a dataset of EMG recordings
from several muscles used in grasping
activities performed by a group of healthy
people. The dataset contains numerous
characteristics, including muscle activation
amplitude, frequency components, and
temporal activation sequences. We used many
machine learning algorithms to anticipate
certain grasping actions and determine the
most important factors that contribute to these
predictions. The feature importance analysis
revealed the crucial function of various
muscles and their activation times in
discriminating between different kinds of
grasps. In addition, we used Shapley Additive
Explanations (SHAP) to verify that our models
were interpretable, enabling us to understand
how specific muscle activations contributed to
the overall grasping action prediction. Our
models showed great prediction accuracy and
gave useful information about the underlying
muscle activation patterns. Furthermore, we
created a real-time decision-support tool to
help clinicians and researchers evaluate EMG
data and make sound judgements on muscle
function and rehabilitation procedures. Future
research will concentrate on verifying these
models in clinical settings and investigating
their potential uses in personalized
rehabilitation programs and the development
of improved prosthetic devices