Accurate coronary artery segmentation on coronary computed tomography angiography (CCTA) is essential for diagnosing coronary artery disease. Deep networks are conventionally trained and evaluated with the Dice coefficient, but volume-overlap metrics are poorly suited to thin, tubular anatomy: since most voxels belong to a few thickproximal segments, a missing distal branch barely affects Dice despite severely disrupting the connectivity required for clinical use. We introduce a surface metric that matches predicted and reference surface points via bipartite assignment under a localized, vessel-radius tolerance, reporting precision, recall, and F1 with decoupled false positives (spurious branches) and false negatives (missed branches) a distinction the symmetric Dice cannot make. With it we show that a strong Dice-trained baseline omits far more vessel surface than it hallucinates, an asymmetry its high Dice hides. Building on this, we propose a differentiable surface loss that simultaneously suppresses spurious mass and recovers absent structure, validated by fine-tuning three backbones (nnU-Net, SwinUNETR, NexToU) on two public benchmarks (Image-CAS, ASOCA). Against a matched-epoch control, it significantly improves surface F1 by recovering missed distal vessels at comparable Dice. Our findings argue for measuring and optimizing the vessel surface, not the volume it overlaps. Code
Figures & tables
Figure 1: Our two contributions, evaluated on identical error modes. (a) The surface metric matches predicted (blue, dashed) and reference (gold) surface points under a per-point radius tolerance ri , decoupling missed branches (FN) from spurious branches (FP) while accepted matches count as true positives (green). (b) The surface loss corrects both: a removal term drives the spurious branch down ( p^→0 ) and a creation term grows the missed branch ( p^→1 ), each weighted by a distance field (Eqs. 2 , 3 ). Both terms act directly on surface points, ensuring a thin distal branch receives equal gradient magnitude as a thick proximal trunk.
Vessel calibre
Missed (FN) rate ↓
Vol. share
Thin distal ( r<0.8 mm)
39.5%
1.2%
Medium ( 0.8≤r<1.5 mm)
12.5%
24.0%
Thick proximal ( r≥1.5 mm)
4.5%
74.9%
Table 1: Surface error by vessel calibre on the Dice+CE baseline ( n=200 ). The missed (false-negative) rate rises sharply toward thin distal branches, yet those branches carry a negligible share of the volume that drives Dice. Radius from the reference distance transform; per-voxel volume share assigned by nearest medial-axis radius.
Backbone
Loss
Dice ↑
clDice ↑
β0 -err ↓
FP ↓
FN ↓
F1↑
nnU-Net
Dice+CE
80.99
86.23
6.14
14,972
22,124
79.50
clDice [ 6 ]
81.64 ‡
86.78
5.14
14,180
21,542
80.20 †
cbDice [ 8 ]
81.92 ‡
86.99
4.53
13,835
21,089
80.50 †
Ours
82.05 ‡
87.12
4.05
13,215
15,798
82.60 †
SwinUNETR
Dice+CE
79.23
83.28
20.44
19,717
24,215
77.30
clDice [ 6 ]
80.33 ‡
85.14
14.42
19,354
23,985
78.50 †
Table 2: ImageCAS comparison across backbones. Dice, clDice, F1 higher is better; β0 -err, FP (spurious), FN (missed) lower is better. The decoupled FP/FN show which error each method reduces. Dice/clDice/ F1 on a 0 – 100 scale; bold = best per backbone group. ‡ / † : significant over Dice+CE on Dice/ F1 (Wilcoxon, p<0.05 , Holm).
Backbone
Method
Dice ↑
clDice ↑
β0 -err ↓
FP ↓
FN ↓
F1↑
nnU-Net
Best baseline
88.14
87.81
8.75
4,270
5,157
85.40
Ours
88.54
87.98
7.22
3,985
4,520
87.10
SwinUNETR
Best baseline
85.11
84.77
15.62
4,663
8,288
81.90
Ours
85.74
85.12
13.25
4,283
7,235
82.50
NexToU
Best baseline
86.89
86.27
7.88
4,559
5,639
84.30
Ours
87.25
86.90
6.54
4,231
4,725
84.90
Table 3: ASOCA second-dataset replication, strongest baseline vs. our method per backbone. Columns as in Table 2 . With n=8 test cases we report point estimates only; the consistent direction corroborates the ImageCAS findings.
Figure 2: Qualitative comparison across the three backbones (rows) and the competing objectives (columns: reference, our method, clDice, cbDice, and the Dice+CE baseline). Predictions are coloured against the reference: green = correctly segmented (TP), red = missed reference vessel (FN), blue = spurious predicted vessel (FP). Our loss recovers missed terminal vessel (less red) across all three backbones while keeping over-segmentation (blue) rare.
Accurate segmentation of the coronary vessel lumen is a prerequisite for quantitative assessment of atherosclerotic plaque and perivascular adipose tissue in coronary computed tomography angiography (CCTA). Cardiologists rely on semi-automated methods for this task because manual vessel tracing and segmentation are labour-intensive. Although many automated methods have been proposed, their validation remains limited by the lack of large, high-quality publicly available datasets. We provide a new dataset of voxel-wise annotations of the vessel lumen and coronary segments, alongside centerlines, and mesh surfaces for 800 scans from the publicly available ImageCAS dataset. Using this dataset, we benchmark established lumen segmentation methods against inter-observer variability, stratifying performance by disease, image quality, coronary dominance, coronary segment, vessel diameter, and lumen attenuation. These labels allow segmentation accuracy to be described in anatomical and clinical context rather than reported as a single aggregate score. The dataset supports the development and validation of methods for lumen segmentation, plaque and perivascular quantification, and haemodynamic modelling.
Kit M. Bransby, Esther Øksnebjerg, Kristoffer Kjær +7
DTU Compute, Technical University of Denmark, Kongens Lyngby, Denmark · Cardiovascular Research Unit, Copenhagen University Hospital – Rigshospitalet, Copenhagen, Denmark
Accurate pixel-level classification of coronary angiograms is critical for cardiovascular disease assessment, yet the field lacks standardized evaluation protocols. In this work we demonstrate a new benchmark for the assessment of deep learning models which densely classify pixels of coronary angiograms to one of SYNTAX classes (or background). The evaluation covers 24 distinct architectures starting with classic convnets to recent state-space-based vision algorithms. We release CARDIAG - a multi-center, multi-label dataset which we carefully split to reliably compute metrics, accounting for diameter error, overlap, centerline quality and calibration. The data contains SYNTAX labels, binary, uncertainty and segmentation masks as well as intermediate frames together with the selected non-sensitive DICOM metadata. From the multitude of algorithms, we nominate ConvNeXt V2 encoder with DeepLab V3 Plus decoder as the best performing, achieving macro F1=0.456, which we then ensemble with Mamba U-Net and Feature Pyramid Network, for an increased F1=0.479. We demonstrate all the architectures to be well calibrated and determine the generalization of the top 5 methods, together with the data efficiency of these architectures. We highlight the importance of both high-resolution and low-resolution features in encoding. We also demonstrate the model correctness in the context of patient demographic, vessel sides and projection angle configurations. Overall the released benchmark allows for future studies to robustly and rigorously assess the proposals, not only for SYNTAX segmentation, but lesion detection and many more.
Dominik Bernard Lau, Hubert Malinowski, Jerzy Szyjut +5
Gdańsk University of Technology · NASK National Research Institute · Medical University of Gdańsk
Accurate segmentation of coronary arteries from computed tomography angiography (CTA) images is of paramount clinical importance for the diagnosis and treatment planning of cardiovascular diseases. However, coronary artery segmentation remains challenging due to the inherent multi-branching and slender tubular morphology of the vasculature, compounded by severe class imbalance between foreground vessels and background tissue. Conventional convolutional neural network (CNN)-based approaches struggle to capture long-range dependencies among spatially distant vascular structures, while Vision Transformer (ViT)-based methods incur prohibitive computational overhead that hinders deployment in resource-constrained clinical settings. Motivated by the recent success of state space models (SSMs) in efficiently modeling long-range sequential dependencies with linear complexity, we propose MDSVM-UNet, a novel two-stage coronary artery segmentation framework that synergistically integrates multidirectional snake convolution (MDSConv) with residual visual Mamba (RVM). In the encoding stage, we introduce MDSConv, a deformable convolution module that learns adaptive offsets along three orthogonal anatomical planes -- sagittal, coronal, and axial -- thereby enabling comprehensive multi-view feature fusion that faithfully captures the elongated and tortuous geometry of coronary vessels. In the decoding stage, we design an RVM-based upsampling decoder block that leverages selective state space mechanisms to model inter-slice long-range dependencies while preserving linear computational complexity. Furthermore, we propose a progressive two-stage segmentation strategy: the first stage performs coarse whole-image segmentation to guide intelligent block extraction, while the second stage conducts fine-grained block-level segmentation to recover vascular details and suppress false positives..