Flexible and Wearable Ultrasonic Sensors and Method for Classifying Individual Finger Flexions

Abstract

Ultrasound imaging technology has recently been proven to achieve higher classification accuracies than surface electromyography when predicting hand motions. However, typical designs involve a large linear array ultrasonic probe or bulky multichannel ultrasonic transducers. In this study, we constructed wearable ultrasonic sensors (WUS) using 110$-\mu$m thick flexible piezoelectric polymer film for an ergonomic strategy for prosthetic and human machine interface applications. We attached the three WUSs on the forearm of a healthy subject, 5 cm away from the wrist, to monitor the tissue motions associated with the finger flexions. An experiment to predict 100 ms time intervals of individual finger flexions was investigated using novel feature extraction methods involving the discrete wavelet transform. We achieved an accuracy of 92.5±7.6% for classification of finger flexions using a multilayer perceptron with a hidden layer of 15 nodes. The F1 score for classifying the five fingers ranged between 86-99% across all fingers using uniformly distributed class sample sizes. The results strongly support the utility of the ergonomic WUS system for continuously predicting individual finger flexions in prosthetic and human machine interface applications.