Evaluation of image registration methods used to estimate three-dimensional deformation of human skeletal muscle during isometric contractions

We assessed the accuracy of non-rigid image registration algorithms to quantify the three-dimensional deformation of the human medial gastrocnemius muscle from MRI scans of the muscle at rest and during isometric contractions at 10% and 20% of maximum voluntary isometric plantarflexion torque. Demons and b-spline registration were used to predict the displacements of anatomical landmarks within muscles of four participants. The mean registration errors (distance between predicted and true locations as identified manually by three raters) ranged across algorithms from 0.4 to 0.9 mm for 10% contractions and from 1.0 to 1.9 mm for 20% contractions. Even the best-performing algorithm predicted only half of the non-rigid component of the muscle deformation. The limited accuracy of non-rigid registration algorithms to predict the deformation of human muscles in vivo should be considered when designing new studies and interpreting previous studies using these methods.