Systematic analysis of success of lower limb muscle combinations in the prediction of ankle biomechanics during stair descent: Guiding input selection in sEMG-based prosthetic control.

OBJECTIVE

Surface electromyography (sEMG)-driven control is a promising modality for powered lower-limb prostheses. However, high variability of residual musculature and the necessity for low-complexity systems require a systematic optimization of muscle inputs. Moreover, for stair descent motion, the performance of lower extremity muscles individually or collectively, but systematically was not studied. To address this, we evaluated and ranked all combinations of eight muscle inputs for their ankle biomechanics predictive success and specifically studied three use cases: (i) economical (total number of muscles minimized), (ii) flexible (lower leg muscles minimized), and (iii) practical (upper leg muscles used exclusively).

METHODS

The muscles analyzed using a long short-term memory network were: medial gastrocnemius (MG), soleus (SOL), peroneus longus (PL), rectus femoris (RF), vastus medialis (VM), biceps femoris (BF), and gluteus maximus (Gmax). As an evaluation index, an overall error score was defined and computed by using Pearson's correlation coefficient (successful for r > 0.70, very successful for r > 0.90) and root-mean-squared error.

RESULTS

The best-performing combination included four muscles: TA + MG + PL + VM (rposition=0.9091 ± 0.0061, rmoment=0.9534 ± 0.0047). TA muscle alone showed successful performance (rmoment=0.9275 ± 0.0015, rposition=0.8881 ± 0.0062). An added knee extensor, i.e., TA + VM, enhanced the success (rposition=0.9003 ± 0.0021, rmoment=0.9291 ± 0.0095) labeled as the flexible case. RF + VM + BF + Gmax is the practical case.

CONCLUSION

Our results confirm the potential of systematically and strategically selected sEMG sensors in the control of powered ankle prostheses and show that targeting the combined inclusion of knee extensors and, ankle plantar- and dorsiflexors do enhance prediction robustness and estimation performance for stair descent motion.