Development of spike train algorithms for physiotherapy assessment using deep learning approaches
Physiotherapy nowadays has become a demanding medication for curing bones related injuries and pain to restore someone's to health in order to gain back the ability to cope with daily living tasks. As the technologies of sensors have risen, smart physiotherapy monitoring systems become trend...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English English English |
Published: |
2021
|
Subjects: | |
Online Access: | http://eprints.uthm.edu.my/1783/2/FADILLA%20%27ATYKA%20BINTI%20NOR%20RASHID%20-%20declaration.pdf http://eprints.uthm.edu.my/1783/1/FADILLA%20%27ATYKA%20BINTI%20NOR%20RASHID%20-%2024p.pdf http://eprints.uthm.edu.my/1783/3/FADILLA%20%27ATYKA%20BINTI%20NOR%20RASHID%20-%20fulltext.pdf |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Physiotherapy nowadays has become a demanding medication for curing bones related
injuries and pain to restore someone's to health in order to gain back the ability to cope
with daily living tasks. As the technologies of sensors have risen, smart physiotherapy
monitoring systems become trendy researches due to its potential to enhance the
quality of physiotherapy assessment. However, varied sensor technologies of
physiotherapy assessment have lacked versatility and robustness. This research
proposed a spike train feature extraction for physiotherapy assessment to enhance the
patient's progression. However, the concerns are how capable is spike trains in
achieving high accuracy as other related works on recognising and assessing
rehabilitation movements. In this context, spike trains are defined as sequences of
recorded times when neurons fire spikes or also known as action potentials. This study
implemented a spike train as a primary method of feature extraction that illustrated a
significant pattern for each exercise performed.Three datasets, UI-PRMD dataset,
K3Da dataset, and Self-Collected dataset have been adopted in the studies to be
encoded into spike trains formal representation which resulting to an average of 415
spike patterns. Next, the patterns of raster plots were being trained as the input into a
deep learning framework to evaluate the accuracy of the pattern's uniqueness.
Furthermore, this study makes use of the occurrence of spikes' number, which is
known as firing rate, to distinguish movements' correctness and being compared with
the deep learning evaluation measures to prove the efficiency of deep learning
prediction. The proposed framework achieved recognition rates of 99.44%, 98.21%,
and 100.00% for UI-PRMD, K3Da, and self-collected datasets, respectively. These
results proved that the proposed framework achieved targetable accuracy for all
datasets trained with various CNN architectures. Next, the experimental results of
physiotherapy assessment indicate that the correctness prediction by the proposed
framework closely follows the ground-truth value for the movements. This study is
among the first successful attempts of implementing spike train into a deep learning framework for a real-time-based rehabilitation session case study with promising
results. Hence, spike train is the foremost choice as features that are hugely rewarding
towards deep learning as it can visually differentiate each of the physiotherapy
movements with unique patterns. |
---|