LocNeXt: Fast Automatic ConvNeXt-based Coronary Ostia Landmarks Detection in 3D CCTA Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Accurate localization of coronary ostia in CCTA is vital for clinical applications.
• Current methods may lack speed, efficiency, or robustness to anatomical variations.

Purpose of the Study:

• To introduce LocNeXt, a novel deep learning model for automated coronary ostia landmark detection in CCTA.
• To evaluate the performance and efficiency of LocNeXt compared to existing approaches.

Main Methods:

• Utilized the ConvNeXt architecture with modern design principles.
• Employed downsampled heatmaps for per-channel 3D classification, avoiding vector field regression.
• Trained and validated on public datasets (CAT08, ASOCA) with extensive data augmentation.

Main Results:

• Achieved median localization errors of 7.04 mm (LocNeXt-Q) and 5.24 mm (full version).
• Demonstrated fast inference times (3.85 s for LocNeXt-Q, 24.29 s for full version).
• Showed robust performance across anatomical variations and data efficiency.

Conclusions:

• LocNeXt offers a reliable, fast, and automated solution for coronary ostia landmark detection in CCTA.
• The method has the potential to enhance clinical workflow efficiency for various cardiovascular applications.