Fig 1.
Summary of steps in processing of ultrasound images.
Raw images undergo manual preparation to select regions (step 0), and image filtering to highlight line-like structures (step 1). Aponeuroses are detected from the filtered images using either object- or line detection procedures (step 2). An ellipsoid region between detected aponeuroses is used to determine fascicle angle with a line detection procedure (step 3). Geometric muscle features are estimated from detected aponeuroses and determined fascicle angle (step 4). Pennation angle φ is defined as the difference between fascicle angle and superficial aponeurosis angle. Muscle thickness Tmuscle is defined as the perpendicular distance from the deep aponeurosis to the detected superficial aponeurosis. Fascicle length Lfas is calculated from pennation angle φ and muscle thickness Tmuscle using trigonometry. Muscle thickness Tmuscle and therefore fascicle length Lfas can be evaluated at any horizontal location selected by the user (dotted yellow line).
Fig 2.
Image filtering steps for highlighting line-like structures.
Fascicle filtering includes highlighting thin lines using a vessel enhancement filter (step 1.1f) and subsequent thresholding (step 1.2f). Aponeurosis filtering includes highlighting thick lines (step 1.1a), masking with a thresholded version of the original image (step 1.2a and step 1.3a), Gaussian smoothing (step 1.4a) and thresholding (step 1.5a). The results of the fascicle and aponeurosis filtering process are inputs for the fascicle angle determination (step 3) and aponeurosis detection respectively (step 2).
Fig 3.
Fascicle angle and fascicle length estimates of TimTrack, UltraTrack and manual observers for two typical example trials.
Fascicle angle and -length changed considerably during both movements, captured by manual observers (red dots), TimTrack (blue lines) and UltraTrack (yellow lines). Upper row: Two repetitions of a range-of-motion ankle movement from dataset 2. Lower row: Two repetitions of the 30 deg/s isokinetic ankle movement from dataset 3. UltraTrack overestimated fascicle length near the end of the trial, potentially due to integration drift. Manual estimation was performed on the concentric phase of each contraction, in accordance with the analysis by Drazan and colleagues [13].
Fig 4.
Estimates of TimTrack algorithm and manual observer on geometric muscle features in two datasets (1–2).
Estimates of TimTrack algorithm (blue dots) and manual observers (red dots) for fascicle angle (A-C), superficial aponeurosis angle (D-F), pennation angle (G-I), muscle thickness (J-L) and fascicle length (M-O) of vastus lateralis (VL) during isometric torque production (left column), gastrocnemius lateralis (GL) during jumping (middle column) and gastrocnemius lateralis (GL) during range-of-motion movement (right column). Vertical red lines indicate the range of manual estimates for each image. When manual estimates are similar, this connecting line may not be visible. Similarly, when manual and TimTrack’s estimates are nearly identical, the latter may not be visible. Ultrasound images for manual estimation were selected such that they were spaced equidistantly and included the entire range of fascicle angles and -lengths present in each condition.
Fig 5.
TimTrack algorithm estimates vs. mean manual observer estimates for all datasets.
A. TimTrack algorithm estimates of muscle thickness increased linearly with mean of manual estimates (slope = 1.01±0.01, linear regression, dashed red line). B. TimTrack algorithm estimates of pennation angle increased linearly with mean of manual estimates (slope = 0.97±0.01, linear regression, dashed red line). C. TimTrack algorithm estimates of fascicle length increased linearly with mean of manual estimates (slope = 1.01±0.01, linear regression, dashed red line). Overall, TimTrack’s were closely related to mean observer estimates (R2 = 0.93–0.97).
Table 1.
Muscle thickness.
Table 2.
Pennation angle.
Table 3.
Fascicle length.