Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 27

EMG Characterization Through

Visual Biofeedback
Enhancing Muscle Control and Real-Time Assessment

Prepared by Muhammad Nashman Husain [21122054]

1
Content
1. Introduction 2. Methodology 3. Results​
o EMG Overview o Overall Project Flow o Preprocessing Results​
o Visual o Project Aspects o Window Size​
Biofeedback o Overall Design Flow o Extracted Features
o Purpose of EMG o The circuit diagram o PCA​
Characterization o The biofeedback ​
system 4. Discussion​
o EMG Data ​
2. Project Road Map 5. Limitations and Challenges​
Acquisition
o Data Recording ​
6. Future Work​
o Feature Extraction
o PCA

2
Introduction : EMG Overview

EMG (Electromyography) is the


It is a diagnostic technique that
measurement of the electrical
measures, detects, records the
activity of the skeletal muscles
myoelectric signals
(myoelectric signals)

EMG observation details the Helps clinicians to diagnose and


muscle’s physiological treat neuromuscular diseases
condition and issues related to it

3
Introduction : Visual Biofeedback

01 02 03
Visual biofeedback is a It allows the clinician and This is a crucial concept
technique that displays patient to monitor and used in rehabilitation and
and provide a real-time facilitate muscle control prosthetic for patient to
visual representation of a through observation have more control of their
patient’s physiological muscles/prosthesis
response, in this case
EMG

4
Introduction : Purpose of EMG Characterization

EMG characterization is important in many application in the medical fields

Characterizing EMG can lead to patterns of muscle activation which can


classify and distinguish between various gestures and movements

These patterns is widely applied in prostheses training, neuromuscular


disorder diagnoses and rehabilitating muscle control

5
Methodology: Overall Project Flow

6
Methodology: Project Aspects

The project consists of 2 aspects : (a) the technical aspect and (b) the
programming aspect

The technical aspect consists of circuit connection, electrode


placement and data recording/collection

The programming aspect consists of processing the data and


extracting useful features of the EMG for characterization

7
Methodology: Overall Design Flow

8
• 1st Order High Pass Filter (fc = 16Hz)
Methodology :
• Non-Inverting Op-Amp (Gain = 16)
EMG Circuit • 2nd Order Low Pass Filter (fc = 482Hz, Gain = 1)
Diagram • Non-Inverting Op-Amp (Gain = 11)

9
Methodology :
The
Biofeedback
System

• Done by interfacing spacebar control using Arduino


Leonardo
• The spacebar will be pressed when muscle contracts and
released when muscle relaxes
10
Methodology : EMG Data Acquisition

The electrodes was placed on the extensor carpi ulnaris (upper


forearm) and data was sent to Arduino Leonardo

An app called CoolTerm was used to interface with the


Arduino and capture raw EMG data of several hand gestures
and converted into text files

The text filed was imported into Excel and plotted to check for
errors

11
Methodology : EMG Data Acquisition
Hand Gestures
These hand gestures involve the use of the extensor carpi ulnaris.

5. ROCK SIGN
1. FIST 2. 4-FINGER 3. KEY GRIP (KGR) 4. REST
(ROCK)
EXTENSION (4FEX)
12
Methodology :
Data Recording

13
Methodology : Python Processing

Jupyter Notebook was used as the Python IDE for future use

The raw EMG data was imported was preprocessed using built-in Python
libraries

The preprocessing steps consists of zero-centering(ensures the baseline is


at zero),rectification,squaring and moving average(smoothing) of the raw
EMG data
14
Methodology :
Feature Extraction
• The EMG data was then extracted
per batch into several features to
differentiate and better classify the
EMG of the hand gestures

15
The EMG Features
Features Equation Description
Integrated EMG(IEMG) The area under the rectified signal

Mean Absolute Value (MAV) The average mean of the signal

Simple Square Integral (SSI) The summation of the squared signals

Root Mean Square (RMS) The arithmetic mean of the squared values

Variance (VAR) The measurement of the spread of data

Waveform Length (WL) The cumulative length of the signal

Zero-Crossing Rate The rate at which the signal crosses the zero amplitude

16
Methodology : PCA (Principal Component
Analysis)

THE EXTRACTED FEATURES PER BATCH FOR PCA USES A DIMENSIONALITY REDUCTION PCA WAS DONE ON THESE EXTRACTED
EACH GESTURES WAS STORED AND PCA WAS TECHNIQUE TO SIMPLIFY COMPLEX FEATURES TO VISUALIZE AND ANALYSE THE
PERFORMED ON EACH OF THEM DATASETS BY TRANSFORMING THE DATASETS VARIABILITY AMONG THE DIFFERENT HAND
INTO A LOWER-DIMENSIONAL SPACE GESTURES

17
Results: Raw EMG Output

FOREAEM EMG
1200

1000

800
EMG Amplitude

600

400

200

0
28.8477 29.2947 29.7417 30.1886 30.6357 31.0827 31.5297 31.9768 32.4236 32.8697 33.3177 33.7637 34.2107 34.6576 35.1046 35.5516 35.9987 36.4459 36.8927 37.3396 37.7866 38.2336 38.6797 39.1267 39.5736 40.0206

18
esults:
ython Preprocessing

19
• 1 second of measurement contains approximately 440
samples
Results : EMG • Feature extraction was iterated throughout all the

Window Size
samples at window intervals
• The sampled iteration is stored as Python array
variables REST,FIST,4FEX,ROCK,KGR

All
Samples

440 Samples

440 Samples

20
Results : Feature Extraction of all Samples

IEMG MAV SSI RMS VAR WL ZC Repetitions

REST 5728583.81 467.42 45550842836 468.47 985.19 3916081 4 300

FIST 15502796.15 467.11 54825501807 477.32 9640.90 17040014 390 300

4FEX 16908513.43 467.24 59969476913 477.21 9409.31 25990397 172 300

ROCK 11874135.76 467.19 52524773703 471.72 4249.61 11995875 4 300

KGR 7287350.23 467.25 49988916210 469.36 1973.05 9997921 0 300

21
• PCA results dictate the pattern similarity and
relationship with the data.
Results : PCA • Features with similar pattern and behaviour will
be clustered and grouped together

22
Results : PCA • The PCA Variance tells us which Principal Component captures
more variability with the data

Variance Ratios • The PC with higher values dictate that the component captured the
most important patterns in the data

23
Discussion
PCA results show that almost all of my hand gestures have similar patterns and
behaviours.

The features of different gestures are clustered and grouped together

The PCA Variance tells us that PC1 has captured more variability of the data
than PC2

This may be due to similar patterns occurring on the PC1 axis than PC2

24
Limitations & Challenges
Data was only taken from a single EMG channel and from a single patient.

Each user’s hand gestures vary in terms of speed, amplitude and methods

More complex gestures with more sensitive movements depicts a real-life degree of
freedom in prosthetic control but proves more challenging than just 5 hand gestures

A real-time gesture recognition and EMG characterization requires more precise and
efficient algorithm and computation

25
Future Work

• Test real-time feature extraction and EMG


characterization on patient with the trained PCA model
• Develop a multichannel EMG on the flexor carpi
radialis(lower forearm)
• Characterize more gestures to allow more degree of
freedom in the prosthetic control
• Build a printed circuit board and mass produce them
for ease of use and mobility
• Configure the EMG features to control the visual
biofeedback rather than an EMG amplitude for a more
accurate control

26
Thank you

27

You might also like