Bifurcation detection in intravascular optical coherence tomography using vision transformer based deep learning.

Area of Science:

• Cardiovascular Imaging
• Medical Artificial Intelligence
• Interventional Cardiology

Background:

• Bifurcation detection in intravascular optical coherence tomography (IVOCT) is crucial for percutaneous coronary intervention (PCI) strategies.
• Current methods may lack precision in identifying complex coronary artery structures.

Purpose of the Study:

• To develop and evaluate a novel deep learning-based method for bifurcation detection in IVOCT images using vision transformers (ViT).
• To improve the accuracy and efficiency of identifying coronary artery bifurcations and their ostia.

Main Methods:

• A ViT-based classification model was employed for direct bifurcation image identification, bypassing lumen segmentation.
• A ViT-based landmark detection model was utilized to estimate bifurcation ostium points.
• The models were trained and validated on 8640 clinical IVOCT images.

Main Results:

• The ViT-based method demonstrated superior accuracy and F1-scores for bifurcation identification compared to traditional non-deep learning approaches.
• Ostium distance error was significantly reduced by 68.5% compared to traditional methods and 24.81% compared to CNNs.
• The ViT method achieved higher success detection rates, particularly in close proximity sections (0.1 and 0.2 mm).

Conclusions:

• The proposed ViT-based deep learning method significantly enhances bifurcation detection performance in IVOCT images.
• This approach ensures high correlation and consistency between automated detection and expert manual results.
• The method holds significant potential for guiding optimal PCI treatment strategies.