This paper concerns the fully automatic direct in vivo measurement of activeand passive dynamic skeletal muscle states using ultrasound imaging. Despitethe long standing medical need (myopathies, neuropathies, pain, injury,ageing), currently technology (electromyography, dynamometry, shear waveimaging) provides no general, non-invasive method for online estimation ofskeletal intramuscular states. Ultrasound provides a technology in which staticand dynamic muscle states can be observed non-invasively, yet currentcomputational image understanding approaches are inadequate. We propose a newapproach in which deep learning methods are used for understanding the contentof ultrasound images of muscle in terms of its measured state. Ultrasound datasynchronized with electromyography of the calf muscles, with measures of jointtorque/angle were recorded from 19 healthy participants (6 female, ages: 30 +-7.7). A segmentation algorithm previously developed by our group was applied toextract a region of interest of the medial gastrocnemius. Then a deepconvolutional neural network was trained to predict the measured states (jointangle/torque, electromyography) directly from the segmented images. Resultsrevealed for the first time that active and passive muscle states can bemeasured directly from standard b-mode ultrasound images, accurately predictingfor a held out test participant changes in the joint angle, electromyography,and torque with as little error as 0.022{\deg}, 0.0001V, 0.256Nm (root meansquare error) respectively.
展开▼
