To develop and validate a multitask deep learning model for simultaneous aortic root segmentation and landmark localization on fluoroscopic images to support image-guided transcatheter aortic valve implantation (TAVI).
Approach:
Model Development: Developed BoundaryAwareMANet (BAMNet), a multitask architecture combining an EfficientNet-V2 encoder, an MA-Net-inspired decoder, a coordinate-aware landmark head, and an auxiliary boundary-guidance pathway.
Dataset: Utilized a retrospective dataset of 2,895 fully anonymized fluoroscopic frames from 83 patients who underwent TAVI between 2018 and 2024.
Performance Evaluation: Model performance was evaluated using patient-level five-fold cross-validation.
Key Findings:
BAMNet achieved Dice scores of X, IoU of Y, and Surface Dice@4 mm metrics across five folds.
Landmark localization reached median and mean errors of 2.03 mm and 2.66 mm, respectively.
The model produced both segmentation masks and landmark coordinates in a single forward pass at approximately 63 FPS.
Interpretation:
Joint segmentation of the aortic root and localization of anatomical landmarks on intraoperative fluoroscopy is feasible based on the model's performance.
Limitations:
The study is based on a retrospective dataset, which may limit generalizability.
Performance may vary with different fluoroscopic imaging conditions, potentially affecting clinical outcomes.
Conclusion:
The developed model demonstrates capabilities that could enhance real-time guidance during TAVI procedures.
