Abstract 4144233: Left Atrial Appendage Thrombus Prediction with Statistical Shape Modeling

Introduction: Shape of the left atrial appendage (LAA) has been proposed a predictor of stroke in patients with atrial fibrillation (AF). Prior literature employ qualitative assessments of LAA shape, which limits reproducibility. Statistical shape modeling is a quantitative method for assessment of shape. CT is increasingly utilized for identifying LAA thrombus prior to cardioversion in patients with AF. In this study we use particle based statistical shape modeling to quantify shape of the LAA, both independently and with the left atrium (LA), and evaluate if shape correlates with presence of LAA thrombus. Methods: We collected 132 cardiac CTs from consecutive studies of patients over 14 months obtained for evaluation of LAA thrombus prior to cardioversion. Of these, 16 patients were excluded. The LA and LAA were manually segmented independently from the systolic phase of the remaining 116 patients. Shape analysis was then performed using Shapeworks software (SCI, University of Utah) to compute shape parameters of the LAA in isolation as well as the LA and LAA in combination without controlling for scale or orientation. The shape parameters explaining the greatest shape variance were considered for the model until at least 80% of shape variance was included. A logistic regression model for prediction of LAA thrombus was created using these shape parameters with forward and backward stepwise model selection. Results: Of the 116 studies analyzed, 6 patients had thrombus in the LAA. Average shapes of the patients with and without thrombus differed in overall size as well as prominence of the LAA. Four shape parameters accounted for 81.2% of the LAA shape variance while six shape parameters accounted for 80.5% of the combined LA and LAA variance. The first shape parameter was predictive of LAA thrombus using both shape of the LAA only (p = 0.0258, AUC = 0.762), and when LAA shape was combined with LA shape in a joint model (p = 0.00511, AUC = 0.877). Conclusion: Statistical shape modeling of the LAA, with or without the LA, can be performed on CT image data, and demonstrates differences in shape of these structures between patients with and without LAA thrombus. Patients with LAA thrombus had a larger overall LAA size, LA size, and a more prominent LAA with distinctive morphology. Findings suggest that statistical shape modeling may offer a quantitative and reproducible approach for using LAA shape to assess stroke risk in patients with AF.