2019/07/02 by Ryan Cunningham, Cunningham, Ryan J., Ian D. Loram +1
Engineering · Medicine · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Infrared Thermography in Medicine #Muscle activation and electromyography studies #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1907.01649
openalex publication_date 2019/07/02 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
Objective: To test automated in vivo estimation of active and passive\nskeletal muscle states using ultrasonic imaging. Background: Current technology\n(electromyography, dynamometry, shear wave imaging) provides no general,\nnon-invasive method for online estimation of skeletal intramuscular states.\nUltrasound (US) allows non-invasive imaging of muscle, yet current\ncomputational approaches have never achieved simultaneous extraction nor\ngeneralisation of independently varying, active and passive states. We use deep\nlearning to investigate the generalizable content of 2D US muscle images.\nMethod: US data synchronized with electromyography of the calf muscles, with\nmeasures of joint moment/angle were recorded from 32 healthy participants (7\nfemale, ages: 27.5, 19-65). We extracted a region of interest of medial\ngastrocnemius and soleus using our prior developed accurate segmentation\nalgorithm. From the segmented images, a deep convolutional neural network was\ntrained to predict three absolute, drift-free, components of the\nneurobiomechanical state (activity, joint angle, joint moment) during\nexperimentally designed, simultaneous, independent variation of passive (joint\nangle) and active (electromyography) inputs. Results: For all 32 held-out\nparticipants (16-fold cross-validation) the ankle joint angle,\nelectromyography, and joint moment were estimated to accuracy 55+-8%, 57+-11%,\nand 46+-9% respectively. Significance: With 2D US imaging, deep neural networks\ncan encode in generalizable form, the activity-length-tension state\nrelationship of muscle. Observation only, low power, 2D US imaging can provide\na new category of technology for non-invasive estimation of neural output,\nlength and tension in skeletal muscle. This proof of principle has value for\npersonalised muscle diagnosis in pain, injury, neurological conditions,\nneuropathies, myopathies and ageing.\n