Design of upper limb muscle strength assessment system based on surface electromyography signals and joint motion

Abstract

Purpose: This study aims to develop a assessment system for evaluating shoulder joint muscle strength in patients with varying degrees of upper limb injuries post-stroke, using surface electromyographic (sEMG) signals and joint motion data.

Methods: The assessment system includes modules for acquiring muscle electromyography (EMG) signals and joint motion data. The EMG signals from the anterior, middle, and posterior deltoid muscles were collected, filtered, and denoised to extract time-domain features. Concurrently, shoulder joint motion data were captured using the MPU6050 sensor and processed for feature extraction. The extracted features from the sEMG and joint motion data were analyzed using three algorithms: Random Forest (RF), Backpropagation Neural Network (BPNN), and Support Vector Machines (SVM), to predict muscle strength through regression models. Model performance was evaluated using Root Mean Squared Error (RMSE), R-Square (R ²), Mean Absolute Error (MAE), and Mean Bias Error (MBE), to identify the most accurate regression prediction algorithm.

Results: The system effectively collected and analyzed the sEMG from the deltoid muscles and shoulder joint motion data. Among the models tested, the Support Vector Regression (SVR) model achieved the highest accuracy with an R ² of 0.8059, RMSE of 0.2873, MAE of 0.2155, and MBE of 0.0071. The Random Forest model achieved an R ² of 0.7997, RMSE of 0.3039, MAE of 0.2405, and MBE of 0.0090. The BPNN model achieved an R ² of 0.7542, RMSE of 0.3173, MAE of 0.2306, and MBE of 0.0783.

Conclusion: The SVR model demonstrated superior accuracy in predicting muscle strength. The RF model, with its feature importance capabilities, provides valuable insights that can assist therapists in the muscle strength assessment process.

Keywords: feature extraction; feature importance; muscle strength assessment; regression prediction; surface electromyographic signals; upper limb movement disorders.

Figure 1

Placement of upper limb testing devices. (A) sEMG sensors are placed on the anterior, middle, and posterior deltoid muscles. (B) MPU6050 sensor is placed at the distal end of the limb.

Figure 2

A method for upper limb muscle strength regression prediction based on sEMG and motion capture. RF, Random forest; BPNN, backpropagation neural network; SVR, support vector regression; RMSE, root mean squared error; R², R-square; MAE, mean absolute error; MBE, mean bias error.

Figure 3

sEMG acquisition device.

Figure 4

Noise reduction processing results. (A) Comparison graph of original sEMG (blue) and sEMG after noise reduction with the Savitzky-Golay filter (red). (B) Blue waveform represents the variation of the original sEMG. (C) Red waveform represents the sEMG after noise reduction with the Savitzky-Golay filter.

Figure 5

Feature extraction results of different sample points (anterior deltoid muscle).

Figure 6

Decision tree error curve for training and testing sets.

Figure 7

RF model prediction results. (A) Training set—predicted values vs. actual values. (B) Testing set–predicted values vs. actual values.

Figure 8

Distribution of feature importance.

Figure 9

BPNN model prediction results. (A) Training set—predicted values vs. actual values. (B) Testing set—predicted values vs. actual values.

Figure 10

SVR model prediction results. (A) Training set—predicted values vs. actual values. (B) Testing set—predicted values vs. actual values.

Figure 11

Comparison of regression models. (A) Comparison of R² results for three models. (B) Comparison of RMSE results for three models. (C) Comparison of MAE results for three models. (D) Comparison of MBE results for three models.