Clinical Report: Simultaneous Segmentation of the Aortic Root and Anatomical Landmark Identification Using Intraoperative Fluoroscopy for TAVI Assistance
Overview
This study presents a multitask deep learning model, BAMNet, for simultaneous segmentation of the aortic root and localization of anatomical landmarks during TAVI. The model demonstrated real-time inference capabilities.
Background
Accurate visualization of the aortic root and anatomical landmarks is crucial during transcatheter aortic valve implantation (TAVI) to ensure optimal valve positioning and minimize complications. Current reliance on fluoroscopy is limited by low soft-tissue contrast and motion artifacts. This study employs advanced deep learning techniques to enhance intraoperative imaging.
Data Highlights
Metric
Value
Dice Score
Not specified
IoU
Not specified
Surface Dice@4 mm
Not specified
Median Localization Error
2.03 mm
Mean Localization Error
2.66 mm
Real-time Inference Rate
63 FPS
Key Findings
BAMNet achieved significant accuracy in aortic root segmentation and landmark localization.
The model produced segmentation masks and landmark coordinates in a single forward pass.
Median and mean localization errors were 2.03 mm and 2.66 mm, respectively.
Real-time inference was maintained at approximately 63 frames per second.
The study utilized a dataset of 2,895 fluoroscopic frames from 83 patients.
Clinical Implications
The implementation of BAMNet may enhance the accuracy of anatomical landmark identification during TAVI.
Conclusion
The study demonstrates the feasibility of using a multitask deep learning model for real-time segmentation and landmark localization in TAVI, addressing significant challenges in intraoperative imaging.
