Uwamahoro, Raphael (2025) Estimation of elbow flexion torque from NMES MMG signals and anthropometric variables using GLEO-RFR. Doctoral thesis, Universiti Teknikal Malaysia Melaka.
![[img]](http://eprints.utem.edu.my/style/images/fileicons/text.png) |
Text (Full Text)
Estimation of elbow flexion torque from NMES MMG signals and anthropometric variables using GLEO-RFR.pdf - Submitted Version Available under License Creative Commons Attribution. Download (8MB) |
Abstract
This thesis presents the development of an elbow joint flexion torque (TQ) estimation model that integrates neuromuscular electrical stimulation (NMES) induced mechanomyography (MMG) signals from the biceps brachii (BB) muscle across three forearm postures and four elbow flexion angles, together with anthropometric variables of the arm. 36 healthy male participants received NMES at 30Hz frequency, 110µs pulse width, 30mA current amplitude, 1s ramp time, and a 6s on over 2s off duty cycle, which induced TQ levels below 15% of maximum voluntary isometric contraction (MVIC). 30s recordings of TQ and MMG signals were collected at forearm positions of neutral, pronation, and supination, and at elbow flexion angles of 10°, 30°, 60°, and 90°. In addition, each participant performed voluntary contractions in 3 randomly selected combinations of elbow angle and forearm posture to provide data for torque estimation model validation. MMG, TQ and anthropometric measurements were recorded into a computer through a data acquisition device for offline analysis. 12 MMG features were extracted and assessed for reliability using Two-Way Random Effects, Single Measurement and Absolute Agreement Intraclass Correlation Coefficient ICC (2,1), at 95% confidence interval. Also, 7 anthropometric variables were validated via intra-test percentage reliability (%R) and technical error of measurement (TEM). Further, Grey Relational Degree (GRD) analysis quantified the correlation of MMG and anthropometric++ features with TQ output. These features were subsequently employed to develop a random forest regression (RFR) based TQ estimation model, optimized via the general learning equilibrium optimizer (GLEO) for feature selection and hyperparameter tuning. Test–retest ICC (2,1) values for TQ and MMG ranged from 0.6880 to 0.8230, indicating moderate to high reliability. Forearm posture and elbow angle significantly affected TQ RMS (p < 0.05), with notable variations in MMG RMS, MMG MPF and MMG MDF. MMG RMS and TQ RMS increased from 10° to 60° and then declined at 90° (p < 0.05), whereas MMG MPF and MMG MDF progressively decreased with increasing joint angle (p < 0.05) along the lateral and transverse muscle axes. Since the behaviour of the transverse axis was statistically significant across a majority of postures and angles, data from it was used for model development. GRD analysis showed TQ and MMG correlation coefficients from 0.5734 to 0.8173. The optimized RFR model achieved 33% of feature reduction (from 12 to 8), yielding 6.25% of improvement in the R2 values (from 0.7228 to 0.7853) and 0.5232 on the testing and validation datasets respectively. Similarly, anthropometric variables exhibited TEM values between 0.0079 and 0.2417, with %R ranging from 97.9294 to 99.9567. GRD analysis showed TQ and anthropometric++ features correlation coefficients ranging from 0.5808 to 0.8708. Anthropometric++ features based RFR model also achieved 33% of feature reduction (from 9 to 6), 6.10% improvement in the R2 values (from 0.6560 to 0.7170) and R2 of 0.4437 on the testing and validation datasets respectively. These findings support the integration of MMG and anthropometric features towards enhanced TQ prediction accuracy, with MMG features demonstrating superior performance. This hybrid approach holds significant implications for ergonomic design, assistive technologies, and sports rehabilitation strategies.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | Mechanomyography, Neuromuscular electrical stimulation, Anthropometry, Random forest regression, General learning equilibrium optimizer |
| Subjects: | T Technology T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Divisions: | Library > Tesis > FTKEK |
| Depositing User: | Norhairol Khalid |
| Date Deposited: | 17 Mar 2026 07:16 |
| Last Modified: | 17 Mar 2026 07:16 |
| URI: | http://eprints.utem.edu.my/id/eprint/29658 |
| Statistic Details: | View Download Statistic |
Actions (login required)
 |
View Item |