BACKGROUND: Sutureless valves require preoperative planning using computer tomography as a guide for sizing. We aimed to build a computer tomography based sizing model that mimics sizing procedure taking into account also two marginal non implantable sizes: smaller than small size available and bigger the largest size available.
METHODS: 202 patients who underwent Corcym Perceval implantation according to valve sizing recommendations and institutional experience (over 1300 implantations). All the patients had aortic valve annulus measurements in MPR mode as described by our group previously. All measurements and hemodynamic measurements were used to build a machine learning model with multiple cross-validation. Then the model used to predict valve size on unseen data. Labeling was undertaken taking into account four manufacturer provided sizes plus two un-implantable marginale sizes.
RESULTS: Model was performing with highest accuracy on the same database that was created with. On K-fold cross-validation cross validation was performed with 78 % of accuracy, 97.5% accuracy on train dataset, and 90.4% on unseen test data. Prediction on the old database (before 2017) demonstrated that all the valves were oversized almost on all the sizes.CONCLUSIONS: Current style of sizing is reproducible and can be used as an actual sizing in clinical settings. This algorithm can be practically applied to clinical work.
