Abstract 4365282: Predicting Aortic Diameter Expansion in Acute Uncomplicated Type B Aortic Dissection Using a Novel Voxel-Based Statistical Shape Modeling Approach

Background: Type B aortic dissection (TBAD) occurs when a tear develops in the intimal layer of the descending aorta. Aortic expansion remains a major factor limiting acuate uncomplicated TBAD patients’ survival outcomes. Accurately predicting aortic diameter growth may guide the selection of optimal treatment strategies on a patient-specific basis, improving patient survival. We developed a voxel-based statistical shape modeling (SSM) approach for complex, multi-label TBAD anatomies and hypothesized that the extracted shape features could predict aortic growth. Methods: 55 clinical CT scans from 22 patients at diagnosis and follow-up were obtained. The true and false lumens, dissection flap, fenestrations, and thrombus were segmented from the CT images. Each patient’s geometry included three anatomical components: (1) true and false lumens represented by voxel segmentations, (2) thrombus modeled with a voxel grid within the false lumen, and (3) fenestrations mapped onto a surface pixel grid of the dissection flap. This representation enabled construction of three independent statistical shape models (SSMs) for the individual components. Principal component analysis was applied to each component after converting the voxel- and pixel-based representations into continuous forms. The resulting SSM-derived shape features were used to predict aortic diameter growth rates via linear regression with leave-one-out (LOO) cross-validation. Results: The first eight SSM shape modes from each of the true and false lumen, thrombus, and fenestration models explained 88%, 92%, and 66% of shape variation, respectively. Figure (A) shows the mean TBAD shape and the first mode of variation from each SSM. These anatomical shape features derived from the SSMs were used to predict aortic growth rates using linear regression. Fig. (B) shows the actual versus predicted growth rates from leave-one-out cross-validation. The results demonstrated a root mean square error (RMSE) of 1.29 mm, with 87.5% of the predicted growth rates falling within one standard deviation of the actual growth rates. Conclusion: The novel voxel-based shape representation enabled the construction of independent SSMs for distinct anatomical components of TBAD, allowing for detailed modeling of complex shape features. The combination of SSM-derived shape features from voxel-based representations with linear regression presents a promising strategy for predicting aortic growth rates.