Tavi Valve Size Prediction Using Machine Learning
Edoardo Zancanaro, Rafik Margaryan, Tommaso Gasbarri, Marcello Ravani, Anees Al Jabri, Sergio Berti, Marco Solinas.
Ospedale Del Cuore Fondazione 'G. Monasterio', Massa, Italy.
BACKGROUND: We aimed to build a machine learning model in order to predict TAVI size based on anatomical and functional measurements of the aortic valve. METHODS: 296 patients who underwent TAVI implantation in the catlab and had documented annular measurements using ProSize commercial software and viewed by cardiologist and cardiac surgeon. 216 of these patients implanted Sapien TM valves in four different sizes. RESULTS: Basic model was able to perform with high accuracy(85%) on the unseen data (10 time cross validation, see figure 1). After fine tuning the final model used for cross validation and was able to predict the correct size of the valve with 90.1 % accuracy, precision and recall about 0.995. Macro average ROC AUC was 0.99 (see figure 2). CONCLUSIONS:Machine learning models can predict with high accuracy new valve size and hence can be useful for clinical decision making.
Back to 2023 Abstracts