ProRuka: A highly efficient HMI algorithm for controlling a novel prosthetic hand with 6-DOF using sonomyography

TL;DR

This paper introduces ProRuka, a low-cost 6-DOF prosthetic hand controlled by sonomyography and machine learning, demonstrating high accuracy in gesture classification and effective real-time control for daily activities.

Contribution

The study presents a novel integration of sonomyography with a 6-DOF prosthetic hand and evaluates multiple machine learning algorithms for gesture recognition, including transfer learning.

Findings

Nine hand gestures classified with 100% success rate off-line.

Prosthesis successfully controlled in real-time for daily activities.

System demonstrates high reliability and precision in gesture recognition.

Abstract

Sonomyography (SMG) is a novel human-machine interface that controls upper-limb prostheses by monitoring forearm muscle activity using ultrasonic imaging. SMG has been investigated for controlling upper-limb prostheses during the last two decades. The results show that this method, in combination with artificial intelligence, can classify different hand gestures with an accuracy of more than 90%, making it a great alternative control system compared to electromyography (EMG). However, up to now there are few reports of a system integrating SMG together with a prosthesis for testing on amputee subjects to demonstrate its capability in relation to daily activities. In this study, we developed ProRuka, a novel low-cost 6-degree-of-freedom prosthetic hand integrated with the control provided by a SMG system with a wearable ultrasound imaging probe. The classification of hand gestures using different machine learning classification/regression algorithms including KNN, nearest neighbor regression, random forest, decision tree classifier, decision tree regression, support vector regression and support vector machine in combination with a transfer learning model (VGG16) was first evaluated off-line to determine its reliability and precision. Additionally, the developed controlling system were evaluated on two amputees, in real-time experiments using a variety of hand function test kits. The results from an off-line study including ten healthy participants indicated that nine different hand motions can be classified with a success rate of 100%. In addition, the hand function test in real time (using 4 different hand gestures) confirmed that the designed prosthesis with the SMG controlling system can assist amputees to perform a variety of hand movements needed in daily activities.