[3Yin-A] Poster 5

[3Yin-A-63] Estimation of Five-Finger Isometric Flexion Forces Using CNN and LSTM from Forearm B-mode Ultrasound Images

〇Shunya Sasaki¹, Noriki Mochizuki¹ (1. Nippon Institute of Technology)

Keywords:

Ultrasound,Force estimation,Convolutional neural network

In estimating finger motion and force from muscle activity, surface electromyography (sEMG)–based approaches are widely used. Because sEMG measures electrical signals during muscle activity using electrodes attached to the skin surface, it provides information reflecting the activity of multiple muscles. In contrast, ultrasound imaging can spatially capture morphological changes associated with muscle contraction, including those of deep muscles, and has been applied to tasks in which multiple muscles are simultaneously involved, such as finger-related actions. Prior studies using forearm ultrasound images have addressed hand gesture classification, estimation of finger joint angles and hand posture, grip/prehension force estimation, and single-finger fingertip force estimation. Therefore, this study extends the target to five-finger isometric flexion forces and continuously and simultaneously estimates them over time using a regression model that combines a convolutional neural network (CNN) and a long short-term memory network (LSTM) with forearm B-mode ultrasound images as input. As a result, experiments with nine subjects achieved estimation errors of 0.048–0.090 in MVIC ratio, confirming that continuous simultaneous estimation of five-finger isometric flexion forces is feasible.