Analysis and classification of corticomuscolar coactivation features during upper limb movement

Neural Interfaces facilitate human-machine communication. These technologies enable humans to express their intentions to external devices. By acquiring and interpreting biological signals, these interfaces allow for the control of computers, robots, or prosthetics. Such technologies are particularly relevant in rehabilitation and assistive contexts, as they can enhance the quality of life for individuals with motor disabilities. A more innovative and comprehensive application of these methodologies involves the use of hybrid neural interfaces, which combine two or more signals to merge and amplify the individual information they convey. This study aims to highlight the co-activation between brain and muscular signals during the movement of the right upper limb. Thereby providing an integrated view of the user’s motor intentions and enabling their identification and classification within the first second of movement. Various fusion methods have been considered, including some based on traditional techniques such as power spectral analysis and signals coherence, as well as more innovative approaches based on Riemannian geometry. The classification is performed using deep learning algorithms. Data is divided into training, validation, and test sets to evaluate the models’ ability to generalize across different acquisition sessions. The results indicate that the feature structures employed yield information that can be effectively recognized by simple neural network models that require minimal training time. Furthermore, despite the complex classification context involving eleven different upper limb movements, the study demonstrates promising performances. These encouraging results suggest that the analysis of neuromuscular co-activation in cross-frequency provides a good representation of the internal dynamics of the human neuromotor system. This could represent an area of interest for further investigation, aiming to enhance the understanding of neuromuscular mechanisms and their application in Human-Machine Interfaces. In this study, Convolutional Neural Networks (CNNs) were used to process these combined EEG and EMG inputs. CNNs, a form of deep learning, are particularly effective for recognizing patterns in complex data, making them ideal for analyzing bioelectrical signals. By examining EEG and EMG signals simultaneously, CNNs can identify relationships between brain and muscle activity, offering a more complete understanding of motor functions. The main goal is to determine whether CNN models can effectively extract and interpret useful information from these coactive signals. If successful, this approach could enhance HMIs, leading to better applications in neurorehabilitation, prosthetic control, and treatment of mobility disorders like stroke or spinal cord injuries. EEG–EMG fusion could provide deeper insights into motor intentions, improving interventions for various conditions. In this study, Convolutional Neural Networks (CNNs) were used to process these combined EEG and EMG inputs. CNNs, a form of deep learning, are particularly effective for recognizing patterns in complex data, making them ideal for analyzing bioelectrical signals. By examining EEG and EMG signals simultaneously, CNNs can identify relationships between brain and muscle activity, offering a more complete understanding of motor functions. The main goal is to determine whether CNN models can effectively extract and interpret useful information from these coactive signals. If successful, this approach could enhance HMIs, leading to better applications in neurorehabilitation, prosthetic control, and treatment of mobility disorders like stroke or spinal cord injuries. EEG–EMG fusion could provide deeper insights into motor intentions, improving interventions for various conditions.

Neural Interfaces facilitate human-machine communication. These technologies enable humans to express their intentions to external devices. By acquiring and interpreting biological signals, these interfaces allow for the control of computers, robots, or prosthetics. Such technologies are particularly relevant in rehabilitation and assistive contexts, as they can enhance the quality of life for individuals with motor disabilities. A more innovative and comprehensive application of these methodologies involves the use of hybrid neural interfaces, which combine two or more signals to merge and amplify the individual information they convey. This study aims to highlight the co-activation between brain and muscular signals during the movement of the right upper limb. Thereby providing an integrated view of the user’s motor intentions and enabling their identification and classification within the first second of movement. Various fusion methods have been considered, including some based on traditional techniques such as power spectral analysis and signals coherence, as well as more innovative approaches based on Riemannian geometry. The classification is performed using deep learning algorithms. Data is divided into training, validation, and test sets to evaluate the models’ ability to generalize across different acquisition sessions. The results indicate that the feature structures employed yield information that can be effectively recognized by simple neural network models that require minimal training time. Furthermore, despite the complex classification context involving eleven different upper limb movements, the study demonstrates promising performances. These encouraging results suggest that the analysis of neuromuscular co-activation in cross-frequency provides a good representation of the internal dynamics of the human neuromotor system. This could represent an area of interest for further investigation, aiming to enhance the understanding of neuromuscular mechanisms and their application in Human-Machine Interfaces. In this study, Convolutional Neural Networks (CNNs) were used to process these combined EEG and EMG inputs. CNNs, a form of deep learning, are particularly effective for recognizing patterns in complex data, making them ideal for analyzing bioelectrical signals. By examining EEG and EMG signals simultaneously, CNNs can identify relationships between brain and muscle activity, offering a more complete understanding of motor functions. The main goal is to determine whether CNN models can effectively extract and interpret useful information from these coactive signals. If successful, this approach could enhance HMIs, leading to better applications in neurorehabilitation, prosthetic control, and treatment of mobility disorders like stroke or spinal cord injuries. EEG–EMG fusion could provide deeper insights into motor intentions, improving interventions for various conditions.