Development of a fascicle tracking algorithm for the evaluation of muscle architecture in dynamic tasks

Ultrasound (US) imaging represents the most used technique for assessing pennate muscle architecture and its functionality. Fascicle tracking is a method that provides information about changes in architecture during muscle contractions. It consists of continuously tracking one or more fascicles within a target muscle throughout a sequence of US images. The main aim is to follow the fascicle position and orientation during passive or active movements. To estimate this information, fascicle length (FL), and pennation angle (PA) changes are typically computed along the analyzed US sequence. Fascicle tracking potentialities cover many fields, such as rehabilitation, medicine and sport science. Various fascicle tracking algorithms have been presented in literature, to offer an alternative to time-consuming and subjective manual tracking. However, they have several drawbacks, such as their high dependency on the investigated muscle, the US device and the analyzed task. Concerning the latter aspect, current tracking algorithms have been developed for controlled, uniform and constrained isokinetic or isometric contractions. More complex conditions, such as unconstrained dynamic movements (e.g. walking and running) are more challenging to be analyzed in terms of fascicle tracking because of the complex muscle shape changes occurring during the contraction. Achieving accurate fascicle tracking during dynamic activities remains therefore a key challenge for practical applications, since fascicle orientation and velocity may show a higher variability. Starting from these premises, the thesis project focuses on the development of an algorithm, specifically designed to perform fascicle tracking, on gastrocnemius medialis, in dynamic unconstrained conditions, such as during walking. The algorithm is based on the Kanade-Lucas-Tomasi (KLT) algorithm to estimate the displacement of feature points within a region of interest. The tracking is facilitated by the introduction of spatial FFT processing, which provides fascicle enhancement. Due to the challenging conditions, the proposed solution provides the user with the possibility to supervise the automatic tracking by an on-line detection of possible non-physiological estimates of the fascicle changes. The algorithm was tested on video acquisitions (9 s) from 5 subjects, performing two cyclic but unconstrained tasks: heel rise and walking. Algorithm accuracy was quantified by calculating the RMSE between the estimated variables of interest (FL and PA and those obtained from manual tracking (reference gold standard)). Moreover, since the algorithm allows to correct a wrong fascicle inclination by manually retracking it, the number of manual interventions (N) was analyzed too. Results show lower RMSE values for the heel rise (FL RMSE=4.54±1.26 mm, PA RMSE=1.51±0.41°), compared to the walking task (FL RMSE=6.32±2.07 mm, PA RMSE=1.65±0.55°). These values are comparable with those obtained by other algorithms tested on more controlled tasks (i.e. isokinetic and isometric). The average number of manual interventions was 7.9±5.1, which is relatively low value considering the total number of frames (video duration*frame rate) that would be required to be manually analyzed by the operator in case of manual tracking. In conclusion, the algorithm represents a step forward with respect to manual tracking and the state of the art of fascicle tracking algorithms, opening new possibilities in the assessment of muscle structure in unconstrained, dynamic conditions.