Sliding-Window CNN + Channel-Time Attention Transformer Network Trained with Inertial Measurement Units and Surface Electromyography Data for the Prediction of Muscle Activation and Motion Dynamics Leveraging IMU-Only Wearables for Home-Based Shoulder Rehabilitation

Abstract

Inertial Measurement Units (IMUs) are widely utilized in shoulder rehabilitation due to their portability and cost-effectiveness, but their reliance on spatial motion data restricts their use in comprehensive musculoskeletal analyses. To overcome this limitation, we propose SWCTNet (Sliding Window CNN + Channel-Time Attention Transformer Network), an advanced neural network specifically tailored for multichannel temporal tasks. SWCTNet integrates IMU and surface electromyography (sEMG) data through sliding window convolution and channel-time attention mechanisms, enabling the efficient extraction of temporal features. This model enables the prediction of muscle activation patterns and kinematics using exclusively IMU data. The experimental results demonstrate that the SWCTNet model achieves recognition accuracies ranging from 87.93% to 91.03% on public temporal datasets and an impressive 98% on self-collected datasets. Additionally, SWCTNet exhibits remarkable precision and stability in generative tasks: the normalized DTW distance was 0.12 for the normal group and 0.25 for the patient group when using the self-collected dataset. This study positions SWCTNet as an advanced tool for extracting musculoskeletal features from IMU data, paving the way for innovative applications in real-time monitoring and personalized rehabilitation at home. This approach demonstrates significant potential for long-term musculoskeletal function monitoring in non-clinical or home settings, advancing the capabilities of IMU-based wearable devices.

Keywords: deep learning; inertial measurement unit; musculoskeletal analysis; shoulder rehabilitation; surface electromyography.

Figure 1

Experimental setup for IMU and sEMG data acquisition: (a) placement of the IMU sensor; (b) data acquisition checkpoints for IMU and sEMG sensors; (c) a subset of raw signals captured by the system.

Figure 2

Data process result. The Euler angles obtained from IMU transformation and the processed multi-channel EMG signals are plotted for three preset SJ movements.

Figure 3

Dataset organization structure.

Figure 4

SWCTNet model architecture. The model consists of the SW-CNN Block, CTAT Block, and Downstream Task Block.

Figure 5

Structure of the SW-CNN Block.

Figure 6

Structure of the CTAT Block.

Figure 7

Results of the ablation study conducted on four public datasets.

Figure 8

Radar chart comparing the performance of different models on the feature prediction task, normalized to the range [0, 1]: (a) results for SWIFTIES dataset; (b) results for personal dataset.

Figure 9

Visualization of the sequence generation task, showing the actual and predicted EMG feature time-series signals, where (a,b) are the results of the RMS feature, (c,d) are the results of the MPF feature, (a,c) are the results for healthy individuals, and (b,d) are the results for patients.