Student ID: 0000674327
Student Name: Sabahat Waqar
Paper ID: 03
Title
Conformal, stretchable, breathable, wireless epidermal
surface electromyography sensor system for hand gesture
recognition and rehabilitation of stroke hand function
Yang et al. Introduce the development of a conformal, soft, breathable and wireless
epidermal-sEMG (surface electromyography) sensor system for hand gesture recognition
(HGR) as well as clinical rehabilitation applications. The sensor itself is made by a "laser cut
and paste" method in which the material ensures good stretchability and lower impedance,
making it safer for long-term skin application without causing inflammation. The system
exhibits a high signal-to-noise ratio (SNR) during dynamic states like sweating and moving.
The sensor is able to recognize seven unique hand gestures with an accuracy of 97.96% using
a CNN-based deep learning model. Nonetheless, when assessed on hybrid data that contained
inputs for multiple subjects the accuracy goes down to 81.02 % which highlights individual
muscle position variability impediments. The paper showcases the system and its use in
stroke rehabilitation research, where muscle activity can be quantified to monitor neural
innervation response along with fatigue levels during tasks such as grasp motions.
Interestingly, the integrated apparatus has been used in mirror Th[14]（MT to a hand
rehabilitation glove which enables this device to increase stroke patients motor function
recovery. The authors claim that improving the location of electrodes and increasing the
number of sensors in an array could increase accuracy results with direct effect on
rehabilitation outcomes. This epidermal sEMG sensor system, which is low cost and exhibits
excellent performance for clinical rehabilitation/human-machine interface applications.
.
Fig. 3.1. Hand g e s t u r e recognition of the integrated epidermal sEMG sensors s y s t e m . (a) Schematic illustration of Hand g e s t u r e
recognition related to forearm su- perficial muscle groups, and the position of the epidermal sEMG electrode. (b) The architecture of a CNN
that takes in sEMG raw data and outputs its probabilities of classes. (c) The confusion matrix of subject 1 from the CNN model. (d) The
accuracy of various machine learning models and a CNN model based on the dataset from subject 1. (e) The ROC curves of each
classification and the corresponding AVC based on the dataset from subject 1. (f) (i) The accuracy and (ii) ROC curves of each classification
and the corresponding AvC based on the dataset from only one channel. (g) The accuracy of HGR based o n the dataset from five subjects.
The confusion matrix Fig. 3.1. represents the classification performance of CNN algorithm
for identifying hand gestures, i.e. number of correctly and incorrectly classified gestural
classes Results show high accuracy of the proposed system in simple hand gesture
recognition using gyroscope as shown next.
Fig. 3. 2. Clinic application of the integrated epidermal sEMG system in hand function rehabilitation. (a) Photograph of the stroke
upper limb h e m i p l e g i a patient during grip training. (b) (i) illustration of clinical evaluation setting and (ii) sEMG data recording by the
integrated epidermal sEMG system. The comparison of (c) number of complete counts, (d) MAV, and (e) mean frequencies between the
intact and affected side hand. (f) (i) Photograph of the MT with the epidermal sEMG sensor for grip exercises and (ii) a session including
closed and open affected hand processes.
Fig. 3.2. Clinical applications with the integrated sEMG system (stroke rehabilitation), grip
training, stroke patientvalidation. It shows differences in metrics such as total grasps and
mean absolute value (MAV) between the intact side versus affected-affected sides. Results:
The patient had rote more grasps on the able side that was a demonstration of monitoring
muscle activity with this system and also to record the rehabilitation progress.