Machine learning to extract muscle fascicle length changes from dynamic ultrasound images in real-time

B-mode ultrasound has become one-off, if not the main way of measuring muscle fascicle fiber lengths non-invasively. Yet, the gold standard for tracking these is still time-intensive hand-tracking, and even with semi-automated approaches, the process takes time and has to be done post hoc. Hence, towards greatly improving current processing capabilities by tracking these muscle fasicle lengths in real-time, we trained and optimized machine learning models with collected B-mode ultrasound data.

We focused on soleus muscle ultrasound data given the relationships existing between soleus and whole body energetics while walking and our intention to use these measurements in the loop. To ensure these data were representative of different muscle fiber loading and displacement levels, we collected B-mode ultrasound data from the soleus muscle of six participants performing five defined ankle motion tasks: (a) seated, constrained ankle plantarflexion, (b) seated, free ankle dorsi/plantarflexion, (c) weight-bearing, calf raises (d) walking, and then a (e) mix. We collected 24-seconds of 60 frames-pre-second data for each task using Telemed, ArtUs EXT-1H, LV8-5N60-A2 ultrasound probe wrapped to the right calf with 3M Vetrap Bandaging Tape. The probe was aligned so that both aponeuroses were as close to horizontal as possible in the live ultrasound video feed.