Patients with atrial fibrillation (AF) are at risk for cardioembolic stroke. The recent studies suggest that reduced two-dimensional left atrial (LA) deformation, such as longitudinal strain, may be a potential indicator of stroke risk. We aim to evaluate three-dimensional (3-D) LA global and local strain derived from four-dimensional computed tomography angiography in patients with persistent atrial fibrillation, and compare those with and without cardioembolic stroke history. 22 patients with persistent atrial fibrillation (11 with documented cardioembolic stroke) were included in this study. The LA strain was calculated by our novel mesh-regularised sub-volume tracking method. LA morphology, CHADS-VASC2, and 3-D global and local strain in 8 LA regions (anterior, lateral, appendage, roof, posterior, mitral isthmus, floor, and septum) were analysed. To explore the potential performance of the collected parameters for stroke risk evaluation, we used logistic regression to assess the outcomes of using these parameters to identify patients with cardioembolic stroke. Compared to those without previous stroke events, the patients with cardioembolic stroke presented lower global strain (0.090 vs 0.101; p = 0.006). Patients with cardioembolic stroke also had lower strain values in the roof (0.087 vs 0.106, p = 0.030), posterior (0.055 vs 0.072, p < 0.001) and floor (0.083 vs 0.097, p = 0.022). Results from logistic regression indicated that the accuracy of LA posterior strain as the indicator for previous stroke events was 0.8958. Therefore, 3-D LA global and local strain (especially in the posterior wall) could be useful in identifying patients with persistent AF and at high risk for a cardioembolic stroke.
Atrial strain and motion play important roles in evaluation of stroke risks for patients with atrial fibrillation. While cardiac computed tomographic angiography (CTA) provides detailed left atrial morphology with unparallel image resolution, finding a suitable strain measurement method for CTA remains a considerable challenge. In this paper, for the first time, we introduced a mesh regularized image block matching method to estimate 3D left atrial (LA) surface strain with 4D CTA. A series of performance tests with ex-vivo phantom and in-vivo 4D-CTA data were deployed. In conclusion, our proposed method could provide reliable LA motion and strain data within limited time.