Classification and Regression of Muscle Neural Signals on Human Lower Extremities via BP_AdaBoost

Round 1

Reviewer 1 Report

1.      The introductory part requires a better theoretical foundation, so please include more bibliographic sources.

2.      In this paragraph, The existing researches of gait recognition technology indicate that the stronger the stability of knee motion, the more reliable the gait features obtained from knee joint, and the higher the gait recognition rate relying on knee motion., you discuss existing research without citing any sources.

3.      I think the following sentence should be sent to the results: The results show that the BP_AdaBoost strong classifier designed in this paper can improve the recognition of different motion patterns, and the recognition rate is improved by 14.7% compared with the simple BP weak classifier algorithm. Simultaneously, when the EMG signal is used as the input, the BP_AdaBoost strong regressor can well predict the knee angle, BP_AdaBoost algorithm improves that rate of BP neural network from 78.82% to 93.52%.

4.      The following paragraphs must reach the Material and Method in a particular form: We selected 10 healthy students (5 males and 5 females) in this experiments, which had no motor dysfunction or congenital diseases. Before the test, let experimenters get enough rest for 48 hours, remove their hair on the test muscle surface, wipe the muscle surface with an alcohol cotton ball, stick the positive and negative electrode patches of the EMG sensor firmly along the long axis of the biceps femoris and rectus femoris, and place the reference electrode in the tensor fascia lata bundle. The EMG signal acquisition equipment is IWorx-BIO8, which is 8 channel EMG signal collector. The upper computer adopts LabScribe integrated software, which can be output directly .csv format file for classification algorithm recognition. According to the motion pattern of human walking, a timer is used to ensure that 5 actions are completed within 2 seconds to ensure the synchronization of data sampling. Each movement was collected for 10 cycles independently. In order to let the experimenter's muscles rest fully, the experimenter rested for 10 278 minutes between the test groups each time.

5.      The discussion part is very little covered in the paper, so please compare the results obtained by you with other results from similar studies.

6.      Please rephrase the part of the conclusions, where a maximum of two conclusions appear, short and to the point.

7.      Some words/paragraphs/phrases need to be changed or similar to other studies on some crucial paragraphs in the paper. In this way, please reformulate everything that is marked with bold and italic, and strikethrough:

·        Based on this, this paper focuses on the knee movement and muscle neural signals. 57 Firstly, according to the anatomical principle, this paper analyzes the thigh muscles 58 related to the knee movement by OpenSim, and selects the rectus femoris muscle in the 59 front of the thigh and the biceps femoris muscle in the back of the thigh as the research 60 object through SPSS correlation analysis; Then, the EMG signals of the two muscles are 61 obtained experimentally.

 

·        According to the relevant literature of pattern recognition, multi-dimensional EMG 128 signals will lead to information redundancy and increase the burden of classifiers. 129 Simultaneously, because the femoral intermediate muscle belongs to deep muscle, its 130 surface EMG signal cannot be measured effectively. Therefore, the rectus femoris muscle 131 in the front and biceps femoris muscle in the back of the thigh are selected as the 132 research objects to collect muscle EMG signals in this paper.

Author Response

Please see the attachment

Author Response File: Author Response.doc

Reviewer 2 Report

In this study, the authors explore the classification of five knee movements using muscle neural signals. The authors use the BP-AdaBoost algorithm to improve knee movement classifications. However, there are a few points that would need to be addressed as below.

1. The authors assert that because the EMG signal precedes human motion by 30-150 ms and may have broad application prospects as in power assisted exoskeletal systems. It is not clear how the authors can ensure that the classification of a knee movement from muscle neural signals to be less than 30 ms? It is noted that each of these all five actions would be completed within 2 s.

2. Why does the correlation between the rectus femoris muscle signal and biceps femoris muscle signal provide a theoretical basis for the EMG signal acquisition of lower limbs? It is not clear why only these two signals are selected as input? If they are correlated, why both signals are used for the classification of knee movements? Are these two signals correlated strongly to the five knee movements?

3. In this study, 10 healthy students are selected for the experiment. There are 50 cycles (samples) for each subject and I believe the total number of cycles is 500. The authors do not show what percentage of these cycles are used for training, validation and testing respectively. As a consequence, it is difficult to make an assessment on how effective the BP-AdaBoost algorithm is. For the classification of five knee movements, it is not clear whether the number of sample size is adequate.

Overall, this is an interesting article, but it contains some issues that would need to be resolved or clarified. Compared to other recent publications where there exists a significant correlation between the neuromuscular control framework and the gait quality measures in chronic post-stroke individuals (Shin 2021), the ideas and results in this paper are still at a preliminary stage. More innovative work would need to be done to ensure adequate contribution to the current literature.

Author Response

Please see the attachment

Author Response File: Author Response.doc