A multi-label deep residual shrinkage network for high-density surface electromyography decomposition in real-time.
Jinting Ma, Lifen Wang, Renxiang Wu and 8 others
PMID 40340912WHAT IT FOUND
A new EMG decoding method reduced processing delay to 15.15 ms, but another model was more accurate at longer signal windows.
Both were tested only on static calf muscle contractions in 16 participants, not on patient outcomes.
Key findings
01At its best settings, ML-DRSNet correctly identified 10.90 ± 10.51 motor units, 70.87% of the average motor units found by the reference CBSS method, with precision 0.93 ± 0.12 and sensitivity 0.84 ± 0.25.
02At its optimal longer window, ML-DCNN scored 1.00 ± 0.00 on precision, sensitivity and a combined accuracy score (F1-score), and correctly identified 14.71 ± 8.23 motor units, 95.66% of the average reference total; it outperformed the other two models at optimal settings.
03ML-DRSNet was fastest, with a prediction time including its optimal window size of 15.15 ms per window, versus 69.36 ms for ML-DCNN and 76.96 ms for MT-DCNN.
STILL TO COME
How it was doneWhat they found
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The participants were physically active male participants, not patients, so the paper does not show how the method behaves in people with weakness, pain, spasticity or neurological conditions. Only 16 of the recorded participants were analysed because two lacked usable plantarflexion data. The models were tested only during static plantarflexion contractions, not during walking, balance tasks, fatigue or dynamic movement. The reference motor-unit labels came from another decomposition method plus manual editing, so the models were judged against that method's output rather than an independent clinical truth. Models were trained separately for each participant and contraction level, so the paper does not show that one model works across people or sessions. The study did not test electrode displacement, and the reported speed did not include real-world acquisition, transmission or hardware delays.
Declared interests
Funding came from Shenzhen Medical Research Fund, Medicine Plus Program of Shenzhen University, Basic and Applied Basic Research Foundation of Guangdong Province, Shenzhen Science and Technology Program, and National Natural Science Foundation of China. The supplied text does not report author competing interests.
The easy way to misread this
Do not conclude that this algorithm is ready for clinical use or that it improves patient outcomes. It was tested on static calf contractions in physically active participants, and its accuracy was measured against another EMG decomposition method, not against function or treatment results.