Effect of Lateral Resolution on Classifying Individual Finger Flexions using Ultrasound


  • Alexander Fernandes Carleton University
  • Yuu Ono Carleton University
  • Eranga Ukwatta Carleton University


B-mode ultrasound imaging has recently shown promise in achieving higher classification accuracies than sur-face electromyography for predicting discrete hand gestures and individual finger movements. This preliminary study inves-tigates the performance in classifying finger flexions when re-ducing the lateral sampling interval resolution of a conventional clinical ultrasonic imaging probe with data collected from one subject. An experiment using spatial and temporal features, ex-tracted from ultrasound radio-frequency (RF) signals are used with linear discriminant analysis to classify individual thumb, index, middle, ring and pinky finger flexion movements. The spatial lateral sampling interval is increased from 315 μm to 10 mm (reduction in lateral resolution) by averaging four groups of 32 consecutively acquired A-mode ultrasound RF signals from a 40 mm probe. The results for the four averaged RF ul-trasound signals with a 10 mm lateral sampling interval had an F1 score ranging between 77-91% with a classification accuracy of 84% for all five finger flexions. This classification accuracy was similar when using the acquired 315 μm lateral resolution and decreases to a classification accuracy of 32% for no lateral resolution, when the full 40 mm width is averaged into a single RF signal. The results show motivation for using a wearable multichannel ultrasound device for predicting individual finger flexions for prosthetic devices.




How to Cite

A. Fernandes, Y. Ono, and E. Ukwatta, “Effect of Lateral Resolution on Classifying Individual Finger Flexions using Ultrasound”, CMBES Proc., vol. 42, May 2019.