Gesture Classification in Electromyography Signals for Real-Time Prosthetic Hand Control Using a Convolutional Neural Network-Enhanced Channel Attention Model

Comments 1: The title should be in terms of the gesture classification procedure in EMG signals for a further hand control instead of the obtention of a data-based model of the hand, which can confuse the reader. In this sense, a different name can be helpful.

Response: Thank you for your kind suggestion. The title emphasizes the key elements of our research, highlighting the gesture classification procedure, the use of EMG signals, and the implementation of a real-time system for prosthetic hand control. In the revised manuscript, we changed the title of the article. (Page 1)

Comments 2: When stating the low complexity model in the introduction, the computational cost is proven to be competitive, but the complexity is not emphasized. A further analysis (discussion in the results section) should be provided.

Response: Thank you for your kind suggestion. While the computational cost was proven to be competitive, it is crucial to delve into the complexity of the model.  The low complexity of our model can be attributed to streamlined convolutional layers and the incorporation of the ECA module, which enhances the model's attention mechanism without introducing unnecessary computational burden. Specifically engineered for efficiency, the ECA module strategically attends to relevant channels within the data, emphasizing crucial information without imposing excessive computational overhead. This ensures that the model becomes more adept at capturing key patterns and features without a proportional increase in computational requirements. This low complexity is advantageous for real-time applications, particularly in the context of prosthetic hand control, where quick and responsive processing is para-mount. Notably, the achieved enhancement in accuracy comes without a substantial in-crease in computational complexity. Now, we provide a further analysis in the Discussion section of the main text. (Page 14)

Comments 3: Is the EMG data available for further research on the basis of the (possible) published work? The bibliographic analysis is up to date, it is important to justify the advantages of the proposal.

Response: The EMG data used in this study was collected with the approval of the Ethics Committee of Fujian Provincial Hospital (approval Number: K2022-09-015). All these participants have not suffered from upper extremity muscle pain or orthopedic diseases in the past year and were informed about the content and purpose of the experiment.  This dataset is available for further research, through additional studies and refinements to the proposed model, we believe that further insights and improvements can be achieved.

Comments 4: The bibliographic analysis is up to date, it is important to justify the advantages of the proposal.

Response: We appreciate the reviewer's acknowledgment of the up to date bibliographic analysis. In justifying the advantages of our proposal, we emphasize that our study builds upon the latest advancements in deep learning and EMG signal processing. The introduction of the CNN-ECA model, as demonstrated in our experiments, showcases improved accuracy, recall, and F1 score compared to classical machine learning approaches and existing deep learning models. The low complexity of our model, as discussed in the manuscript, is a key advantage, making it well-suited for real-time applications such as prosthetic hand control. We believe that our approach contributes to the current state of the art in EMG-based gesture recognition systems, providing both methodological and practical advancements.

Comments： In general, this manuscript is well-written and provides insight into developing the control system for prosthetic hands.

Response: Thank you very much for taking the time to review this manuscript, and for giving us the opportunity to revise the manuscript.

Suggest 1: Line 167: please give an approximate number of repetitions instead of 'many times'

Response: Thank you for your kind suggestion. In the context of our study, each gesture was repetitively performed approximately 15-20 times within a span of one minute. In the revised manuscript we have made changes and are highlighted in yellow. (Page 4, Line 173)

Suggest 2: Please explain why a random sequence of gestures was not used in the acquisition.

Response: Thank you for your kind suggestion. The decision to avoid a random sequence of gestures was driven by the necessity for precise and accurate labeling of each gesture in the dataset. In order to train our deep learning model effectively, it was crucial to have a one-to-one correspondence between the collected data and their respective labels. This approach ensures that the model learns and generalizes well by associating specific muscle activation patterns with their corresponding gestures. The structured, non-random sequence allowed for a systematic and controlled acquisition process, contributing to the overall reliability of the dataset for training purposes.

Suggest 3: Institutional Review Board Statement: the authors should mention that the study was approved by the ethics committee.

Response 3: Thank you for your kind suggestion. The experiment protocol was reviewed and approved by the ethics committee of Fujian Provincial Hospital (approval number: K2022-09-015). All these participants have not suffered from upper extremity muscle pain or orthopedic diseases in the past year and were informed about the content and purpose of the experiment. We have made changes according to your comments (Page 4, Line 163)