sEMG–IMU Fusion with Transfer Learning for Knee Flexion and Adduction Moment Estimation During Tai Chi

Knee flexion moment (KFM) and knee adduction moment (KAM) are important indicators for assessing knee joint loading in patients with knee osteoarthritis (KOA). Tai Chi is commonly used as an exercise-based rehabilitation intervention for KOA, and accurate assessment of knee joint moments during Tai Chi is important for rehabilitation training. However, due to individual and cohort-related differences, the performance of knee joint moment prediction models often decreases markedly in cross-subject and cross-population applications. To address this issue, this study proposes a knee joint moment prediction method based on surface electromyography (sEMG) and inertial measurement unit (IMU) signals combined with transfer learning. sEMG and IMU data were collected from healthy participants and patients with KOA while performing multiple Tai Chi movements. A multimodal CNN-BiLSTM regression model was developed to simultaneously predict KFM and KAM, and three transfer learning strategies were designed for target-domain adaptation. The results showed that, under the more stringent repetition-level within-subject evaluation, the mean R2 values for KFM and KAM were 0.876 and 0.628, respectively, in the healthy group, and 0.750 and 0.651, respectively, in the KOA group. Direct prediction results in the cross-subject and cross-population tasks indicated substantial domain shifts across individuals and populations. Target-domain fine-tuning substantially improved prediction performance relative to direct prediction. In the transfer from healthy participants to patients with KOA, the R2 values for KFM and KAM increased from −0.627 and −0.138 under direct prediction to 0.660 and 0.497, respectively. However, target-only scratch training achieved R2 values of 0.687 and 0.574, indicating no additional benefit from healthy-source pretraining under the present protocol.

Authors

Institutions

Publication Details

Journal
Sensors
Published
2026-11-17
DOI
https://doi.org/10.3390/s26227251
Primary Topic
Muscle activation and electromyography studies
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

sEMG–IMU Fusion with Transfer Learning for Knee Flexion and Adduction Moment Estimation During Tai Chi

Pan Xu, Meijin Hou, Yueming Gao, Xiaofang Wu et al.
2 links
Sensors
Muscle activation and electromyography studies
article

sEMG–IMU Fusion with Transfer Learning for Knee Flexion and Adduction Moment Estimation During Tai Chi

Pan Xu, Meijin Hou, Yueming Gao, Xiaofang Wu, Yiying Zheng, Jiazhi Sun, Xiangbin Wang
article en

Abstract

Knee flexion moment (KFM) and knee adduction moment (KAM) are important indicators for assessing knee joint loading in patients with knee osteoarthritis (KOA). Tai Chi is commonly used as an exercise-based rehabilitation intervention for KOA, and accurate assessment of knee joint moments during Tai Chi is important for rehabilitation training. However, due to individual and cohort-related differences, the performance of knee joint moment prediction models often decreases markedly in cross-subject and cross-population applications. To address this issue, this study proposes a knee joint moment prediction method based on surface electromyography (sEMG) and inertial measurement unit (IMU) signals combined with transfer learning. sEMG and IMU data were collected from healthy participants and patients with KOA while performing multiple Tai Chi movements. A multimodal CNN-BiLSTM regression model was developed to simultaneously predict KFM and KAM, and three transfer learning strategies were designed for target-domain adaptation. The results showed that, under the more stringent repetition-level within-subject evaluation, the mean R2 values for KFM and KAM were 0.876 and 0.628, respectively, in the healthy group, and 0.750 and 0.651, respectively, in the KOA group. Direct prediction results in the cross-subject and cross-population tasks indicated substantial domain shifts across individuals and populations. Target-domain fine-tuning substantially improved prediction performance relative to direct prediction. In the transfer from healthy participants to patients with KOA, the R2 values for KFM and KAM increased from −0.627 and −0.138 under direct prediction to 0.660 and 0.497, respectively. However, target-only scratch training achieved R2 values of 0.687 and 0.574, indicating no additional benefit from healthy-source pretraining under the present protocol.

SensorsVol. 26(22)
Fujian University of Traditional Chinese Medicine (CN), Fuzhou University (CN)
Good health and well-being
Muscle activation and electromyography studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon