Towards Automatic Risk Prediction of Coarctation of the Aorta from Fetal CMR Using Atlas-Based Segmentation and Statistical Shape Modelling

This paper proposes a fully-automated technique for estimation of an antenatal risk score for Coarctation of the Aorta (CoA) from fetal T2-weighted 3D cardiac magnetic resonance imaging (CMR). Our framework combines automated multi-class fetal cardiac vessel segmentation based on two fully-labelled atlases (control and CoA) with statistical shape analysis of the fetal arch. The segmentation framework is weakly-supervised, requiring only condition-specific atlas labels which are propagated to training subjects. The proposed shape analysis method utilizes the predicted segmentation to extract a set of centerlines and radii capturing the shape of the fetal arch. Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are then applied to derive a CoA risk score. The segmentation framework achieves a mean Dice of 0.86 +/- 0.03 for the aortic region. The CoA shape biomarker accurately discriminated between false positives (FP) and CoA cases (AUC 0.93) and showed good generalisability in an independent test set (AUC 0.87), achieving comparable performance to approaches using manual segmentations. Our proposed fully-automatic technique has the potential to improve the antenatal diagnosis of CoA from 3D fetal CMR data.