Automated three-dimensional tracking of the left ventricular myocardium in time-resolved and dose-modulated cardiac CT images using deformable image registration.

BACKGROUND: Assessment of myocardial deformation from time-resolved cardiac computed tomography (4D CT) would augment the already available functional information from such an examination without incurring any additional costs. A deformable image registration (DIR) based approach is proposed to allow fast and automatic myocardial tracking in clinical 4D CT images.

METHODS: Left ventricular myocardial tissue displacement through a cardiac cycle was tracked using a B-spline transformation based DIR. Gradient of such displacements allowed Lagrangian strain estimation with respect to end-diastole in clinical 4D CT data from ten subjects with suspected coronary artery disease. Dice similarity coefficient (DSC), point-to-curve error (PTC), and tracking error were used to assess the tracking accuracy. Wilcoxon signed rank test provided significance of tracking errors. Topology preservation was verified using Jacobian of the deformation. Reliability of estimated strains and torsion (normalized twist angle) was tested in subjects with normal function by comparing them with normal strain in the literature.

RESULTS: Comparison with manual tracking showed high accuracy (DSC: 0.99±0.05; PTC: 0.56mm±0.47 mm) and resulted in determinant(Jacobian)>0 for all subjects, indicating preservation of topology. Average radial (0.13 mm), angular (0.64) and longitudinal (0.10 mm) tracking errors for the entire cohort were not significant (p > 0.9). For patients with normal function, average strain [circumferential, radial, longitudinal] and peak torsion estimates were: [-23.5%, 31.1%, -17.2%] and 7.22°, respectively. These estimates were in conformity with the reported normal ranges in the existing literature.

CONCLUSIONS: Accurate wall deformation tracking and subsequent strain estimation are feasible with the proposed method using only routine time-resolved 3D cardiac CT.