TG-OT: New AI method fuses CCTA and IVUS scans without manual segmentation

TG-OT tackles the difficult task of registering coronary CT angiography (CCTA) with intravascular ultrasound (IVUS) — a fusion that could give cardiologists a more complete view of coronary artery disease. Traditional methods rely on pre-computed lumen or vessel wall segmentations, which often fail under IVUS acoustic shadowing from calcifications. The new approach, published by researchers from the University of Amsterdam and other institutions, replaces that fragile pipeline with lightweight convolutional neural networks (CNNs) trained to detect calcifications, bifurcations, and lumen radii directly on a topological (θ, z) cylinder. This design encourages topologically coherent detections without requiring explicit segmentation.

Registration is then formulated as an optimization over centerline warping parameters, driven by an unbalanced Sinkhorn optimal transport loss on the cylindrical geometry. This loss provides spatially informative gradients even when predictions are spatially disjoint, and it is complemented by a lumen matching term. In a 5-fold cross-validation on 47 paired CCTA-IVUS cases, TG-OT achieved strong longitudinal alignment (Dice = 0.99), rotational alignment (S_c = 0.96), and lumen alignment (Dice = 0.69) — all without manual interaction or prior segmentation. The work, submitted to MICCAI 2026, represents a meaningful step toward clinical integration of automatic CCTA-IVUS fusion.