Abstract Introduction There are different surgical procedures used to treat aortic valve disease such as aortic valve replacement (AVR), the Ozaki procedure, the Ross procedure, and the valve-sparing procedure. There may be postoperative side effects associated with aortic morphology. Computational analyses that assess morphology are thus necessary in such scenario. Statistical shape modelling is used to assess three-dimensional morphology, discover unique shape features, create mean shapes, and create shape modes that display morphological variability. Hierarchical cluster analysis is a machine learning method used to classify a population into subgroups. Both of these statistical methods were applied to aortic valve replacement patients to evaluate morphological variability after aortic valve replacement (AVR) surgery. Purpose The purpose of this study is to assess the aortic morphology of AVR patients, identify morphological differences between different surgical procedures, identify subgroups postoperatively using statistical shape modelling and hierarchical cluster analysis, and assessing possible correlations between the resultant clusters and function i.e. ejection fraction. Methods Computed tomography and cardiac magnetic resonance images were used to reconstruct aortas into 3D models using segmentation and 3D reconstruction software. For each patient, two models were created (i.e. ascending aorta only and aorta including the arch and the descending aorta). N=35 patients were included in the study. Statistical shape analysis was run to create templates and shape modes. Statistical environment and language software was used to run hierarchical cluster analysis. Results Overall, the hierarchical cluster analysis did not classify the aortas based on type of surgery, but most patients in the Ross group demonstrated morphological differences from the other surgical groups and many of the aortas belonging to the Ross group were clustered together in both clustering analyses. No significant differences between clusters were seen (p=0.47 for ascending, p=0.19 for whole aorta). No correlation was found between clusters and ejection fraction (p=0.75, r2<0.01), possibly indicating that post-AVR aortic morphology in this sample does not reflect overt functional changes. Conclusion Both statistical shape modelling and hierarchical cluster analyses were successful in assessing morphological variability and identifying subgroups in this population. This framework could therefore be extended to larger samples to inform on surgical decision-making allowing surgeons to choose the appropriate surgery for the patient or design prosthetic valves that are more customised to the patient’s anatomy.Cluster-Ejection Fraction Correlation
Read full abstract