Background: Endoscopic mitral valve surgery is becoming standard of care. However it has steep learning curve and annular sizing is difficult with conventional sizers.
Methods: A retrospective analysis was performed on 662 patients who underwent mitral repair at one institution from July 2015 to December 2023. 120 of them had CT scans suitable for mitral valvular apparatus measurements (intertrigonal distance, anterior leaflet max length, etc). A machine learning pipeline was created in order to predict annuloplasty band size (from 26 to 42 mm). Mitral measurements were done in MPR modality.
Results: The final model had high accuracy on the traing dataset, about 99 %. The final model (best model among the trained) was able to predict the annular size with 94% accuracy on unseen data. A linear combination (PCA fo two components) of intertrigonal distance and anterior leaflet lengths was the main linear combination in the model
Conclusions: Linear combination of mitral valve measurements can predict annular size in clinical setting. Its value is highly valuable for young and inexperienced groups.