A Simultaneous Gesture Classification and Force Estimation Strategy Based on Wearable A-Mode Ultrasound and Cascade Model

Abstract

The existing Human-Machine Interfaces (HMI) based on gesture recognition using surface electromyography (sEMG) have made significant progress. However, the sEMG has inherent limitations as well as the gesture classification and force estimation have not been effectively combined. There are limitations in applications such as prosthetic control and clinical rehabilitation, etc. In this paper, a grasping gesture and force recognition strategy based on wearable A-mode ultrasound and two-stage cascade model is proposed, which can simultaneously estimate the force while classifying the grasping gesture. This paper experiments five grasping gestures and four force levels (5-50%MVC). The results demonstrate that the performance of the proposed model is significantly better than that of the traditional model both in classification and regression (p<0.001). Additionally, the two-stage cascade regression model (TSCRM) used the Gaussian Process regression model (GPR) with the mean and standard deviation (MSD) feature obtains excellent results, with normalized root-mean-square error (nRMSE) and correlation coefficient (CC) of 0.1049 ± 0.0374 and 0.9461±0.0354, respectively. Besides, the latency of the model meets the requirement of real-time recognition (T<15ms). Therefore, the research outcomes prove the feasibility of the proposed recognition strategy and provide a reference for the field of prosthetic control, etc.

Estimation of joint torque in dynamic activities using wearable A-mode ultrasound

“…In contrast, prior A-mode ultrasound studies have primarily used more advanced machine learning algorithms, such as deep neural networks, by training them directly on the raw ultrasound data. Despite enabling various classification applications 39 , 41 – 43 and continuous joint kinematics 66 or static grip force 44 , 45 estimation, these algorithms often require substantial amounts of training data due to the large number of tunable parameters within the model. In comparison, we alleviated the data requirement by training the relatively simple quadratic fit on the extracted muscle thickness, rather than on the raw A-mode data.…”

“…In practice, this challenge has led to a rather empirical approach in defining the relationship between muscle deformation measurements and joint torque estimates. Specifically, a spectrum of models, including quadratic 28 , cubic 25 , 70 , exponential 71 , and even more complex machine learning 44 – 47 models have been suggested to describe such relationship. In this work, we chose the quadratic fit, mainly for its simplicity and interpretability, for describing the mapping from muscle thickness and joint kinematics to joint torque.…”

“…While A-mode ultrasound has been used to classify hand gestures 38 – 41 , muscle states 42 , and ambulation modes 43 , continuous estimation of muscle mechanical loads is less explored. Only a few studies have investigated estimating hand grip force 44 , 45 and knee torque 46 during static isometric contractions or elbow torque during controlled isokinetic elbow flexions 47 . The use of A-mode ultrasound for reliable muscle deformation measurement and joint torque estimation during dynamic and unconstrained functional tasks remains unexplored.…”