Recursive multiresolution convolutional neural networks for 3D aortic valve annulus planimetry

Comprehensive Summary

In this study, Theriault-Lauzier et al. investigate the use of 3D convolutional neural networks to determine the location and orientation of the aortic valve annular plane. Researchers used 1007 ECG-gated CT scans from 94 patients with severe degenerative aortic valve stenosis and the TensorFlow framework. K-fold cross-validation was performed to further evaluate performance. The model achieved an average localization error of 0.7 ± 0.6 mm for the training dataset and an error of 0.9 ± 0.8 mm for the evaluation dataset, which accurately corresponds to other published approaches. In 84.6% of the test cases, orientation errors were under 10˚. For the training dataset, the angular orientation error was (3.9 ± 2.3) ˚, while the evaluation dataset had an error of (6.4 ± 4.0)˚. The authors emphasize that this was the first application of convolutional neural networks to aortic valve planimetry, which achieved expert-level orientation accuracy.

Outcomes and Implications

This research is important because accurate determination of the aortic annular plane is crucial for correct transcatheter aortic valve replacement sizing and procedural success. This method, specifically, provides a fast and highly accurate alternative to manual annotation, which reduces operator variability and further error. From this initial study, potential extension into other anatomical measurements using convolutional neural networks could be achieved.