| Muscle force estimation is an important issue in the biomechanics,which is of great importance to many research fields,such as clinical medicine,rehabilitation,biomaterials or sport training.Surface electromyogram(sEMG)signals,which can reflect the muscle activation level and can be easily recorded during human movement,have been widely used in indirectly estimating muscle force.High-density sEMG(HD-sEMG)grids,which are capable of collecting hundreds of monopolar sEMG signals over a larger portion of the muscle and providing 2-dimensional spatial distribution of muscle activity during muscle contractions,can provide more accurate quantitative analysis of muscle activation than the conventional single/bipolar electrode.Besides,individual human skeletal muscles are attached to a bone on each end by tendons.According to human anatomical studies,many skeletal muscles have multiple tendons,which suggests that skeletal muscles can be subdivided into smaller segments called muscle-tendon units(MTUs).Some researches have shown that multiple MTUs in one muscle were controlled via different strategies and thus different portions of the muscle may be recruited separately depending on the task demands.This finding can be used to explain heterogeneity of muscle activation.Therefore,when estimating muscle force,it is necessary to analyze the main active regions of MTUs using HD-sEMG grids based on the relationship between the movement characteristics of joint and MTUs,which can help improve the quality of muscle force estimation.The aim of the dissertation is to,by means of a HD-sEMG recording,explore how to determine the specific regions of the MTUs and the temporal variation of MTUs activation level,which can then be applied in muscle force estimation.The main research work and innovation points of the dissertation include the following aspects:(1)A non-negative matrix factorization(NMF)algorithm was applied to HD-sEMG signals obtained from a single skeleton muscle for locating MTUs and determining their activation levels.The HD-sEMG envelope matrix was factorized by the NMF algorithm to get a matrix of basis vectors with each column representing an activation pattern and a matrix of their corresponding time varying coefficients.The area of each MTU was identified through a gray-scale map of weighting factors derived from its activation pattern vector,which was different from conventional methods using the signal's amplitude such as mean absolute value(MAV)or root mean square(RMS).In order to verify the feasibility of the proposed method,two gastrocnemius related and one biceps related motion tasks were defined in our study,and eight healthy male subjects participated in the data collection experiments.In the gastrocnemius related Task 1,subjects were required to keep standing on their toes and change the feet from toe-out to toe-in,which was supposed to target the medial head firstly and then lateral head.The gastrocnemius related Task 2 was defined with motions in a opposite order to those of Task 1.In the biceps brachii related motion task(Task 3),subjects were instructed to keep contracting the biceps brachii muscle to perform elbow flexion and meanwhile rotated their wrist in the same side from initial supination to a neutral position,and then returned back to supination.This task was designed to target the medial head at the first stage,then the lateral head and the medial head at the end.It was hypothesized that each task involved two main activation patterns,one representing that the lateral head was preferentially activated and the other representing that the medial head was preferentially activated.The results showed that specific distinguishable area of MTUs could be identified through the extracted activation patterns and their corresponding time-varying coefficient curves could efficiently depict how the activation level of each MTU changes through the task performance.Compared with conventional methods identifying the MTU activity using RMS method,the results demonstrated that the proposed method could separate the areas of MTUs successfully even at the low degree of compartmentalization for the biceps brachii muscle.Another advantage of the method is that the factorized result of NMF is quite robust and insensitive to noise,whereas the results of RMS method is sensitive to the length of the window and influenced significantly by the noise.(2)A new method in the estimation of muscle force using sEMG was proposed.The major activation pattern was selected from all the extracted activation patterns,and the channels with top 32(the first quarter)weighting factors were selected as the input activation signal to a force estimation model.In order to verify the feasibility of the proposed method,the biceps brachii muscle was examined,and two isometric contraction tasks were designed.In the first task,the contraction level of the force was increased linearly from zero to a targeted level(20%,40%,and 60%maximum voluntary contractions(MVCs)respectively).In the second task,the wrist was changed between the supination and neutral positions,and the contraction level of the force was changed randomly from zero to 60%MVC.The performance metric used in the validation process was expressed as the root mean square difference(RMSD),the correlation coefficients and Bland and Altman.Polynomial fitting technique and Hill model were both adopted as the force estimation model.In the first task,all the 12 subjects recruited in this study were subdivided into two groups(Group 1 and Group 2).For the 8 subjects in the Group 1,the activation intensity of the major pattern was significant greater than the other pattern.The results showed that the proposed method could significantly improve the quality of force estimation compared with the conventional method using all the electrodes from the grid.For subjects in Group 2,by contrast,both extracted activation patterns had similar activation intensity,and therefore no significant difference was observed in force estimation quality between the proposed and conventional method.However,the electrode number used in force estimation was substantially reduced when using the proposed method.In the second task,eight subjects were recruited.The framework was changed:the whole muscle activity was firstly divided into eight segments,and the major activation pattern of each segment was extracted through NMF.It was found that the areas of the channels selected for force estimation of each segment varied between the regions of the medial and lateral head of biceps brachii.The results showed that the proposed framework yielded improved accuracy of force estimation,as compared with the method using the channels selected from the major activation pattern extracted from the whole activity.The study of this dissertation can be used to detect the activity of MTUs within a skeletal muscle in various training sessions,which provides a way to find proper electrode placement for muscle force estimation.The method proposed here can be employed on multiple noteworthy applications such as analysis of muscle activation heterogeneity,muscle fatigue examination based on MTUs,myoelectric control of prosthetic and robotic devices for assisted rehabilitation and guidelines for a more reasonable and efficient training protocols for athletes,fitness enthusiasts or patients with neuromuscular disorders or injuries. |
PDF Full Text Request