Prediction of finger movements using ultrasound images

Ultrasound imaging as a diagnostic device is a widespread, non-invasive technique used to investigate the interior of the human body. It uses high-frequency sound waves to visualise organs as well as musculoskeletal structures. In this thesis, ultrasound images of the human forearm are used to predict nger positions. A possible application for the presented method is the control of dexterous hand prostheses or robotic hands. Local rst-order features extracted from ultrasound images, and position values gathered by a sensor glove, provide the basis for regression analysis. Preliminary experiments prove the feasibility of this approach and reveal that there is a linear relationship between image features and nger joint angles. This holds uniformly for several subjects and enables on-line computation. Speci c nger movements, such as nger exion, thumb rotation and thumb adduction, can be reconstructed up to sub-degree precision. In order for the system to correctly estimate the hand posture, movements of the ultrasound transducer with respect to the skin must be kept to a minimum. An optical ow based technique is implemented to compensate the inevitable movements and prevent drift errors. Several experiments conducted with an amputee also suggest that the system is able to reconstruct the imaginary limb and therefore may be used to treat phantom limb pain.