Reconstructing limb function represents a shared goal between researchers and amputees.However,the development of human-machine interfaces for decoding multi-degree-of-freedom(multi-DoF)movements remains challenging d...Reconstructing limb function represents a shared goal between researchers and amputees.However,the development of human-machine interfaces for decoding multi-degree-of-freedom(multi-DoF)movements remains challenging due to muscle crosstalk,co-activation,and incomplete extraction of motor unit(MU)activities in surface electromyography(sEMG)signals.To address these issues,this study proposes an enhanced neural-driven musculoskeletal model(MM)by integrating MU classification into the decoding process.Six sequential two-DoF movement tasks were designed and a classification framework containing eight task-specific separation matrices was established based on the selective activation of the MUs.The interference between multi-DoF movements was significantly reduced by refining the separation matrices,which effectively removed the MUs co-activated by multiple DoFs.The refined separation matrices were used to derive neural drives,which were subsequently integrated into the proposed four-DoF MM,and the accuracy loss resulting from reduced MU counts was compensated through the iterative optimization of physiological parameters.The proposed method was evaluated by an offline experiment involving 13 participants,and then compared with both classical neural-driven and non-negative matrix factorization(NMF)-driven MMs.Results demonstrated significant improvements in both correlation coefficient and normalized root mean square error,especially in complex four-DoF movement tasks.This study offers a novel and biologically grounded decoding strategy that enhances multi-DoF movement prediction and provides a promising direction for advanced prosthetic control.展开更多
基金supported by the National Key R&D Program of China(Grant No.2022YFB4700801)the National Natural Science Foundation of China(Grant No.52525504)the Emerging Frontiers Cultivation Program of Tianjin University Interdisciplinary Center。
文摘Reconstructing limb function represents a shared goal between researchers and amputees.However,the development of human-machine interfaces for decoding multi-degree-of-freedom(multi-DoF)movements remains challenging due to muscle crosstalk,co-activation,and incomplete extraction of motor unit(MU)activities in surface electromyography(sEMG)signals.To address these issues,this study proposes an enhanced neural-driven musculoskeletal model(MM)by integrating MU classification into the decoding process.Six sequential two-DoF movement tasks were designed and a classification framework containing eight task-specific separation matrices was established based on the selective activation of the MUs.The interference between multi-DoF movements was significantly reduced by refining the separation matrices,which effectively removed the MUs co-activated by multiple DoFs.The refined separation matrices were used to derive neural drives,which were subsequently integrated into the proposed four-DoF MM,and the accuracy loss resulting from reduced MU counts was compensated through the iterative optimization of physiological parameters.The proposed method was evaluated by an offline experiment involving 13 participants,and then compared with both classical neural-driven and non-negative matrix factorization(NMF)-driven MMs.Results demonstrated significant improvements in both correlation coefficient and normalized root mean square error,especially in complex four-DoF movement tasks.This study offers a novel and biologically grounded decoding strategy that enhances multi-DoF movement prediction and provides a promising direction for advanced prosthetic control.
