Attenuation artifact detection and severity classification in intracoronary OCT using mixed image representations

TL;DR

Attenuation artifacts in intracoronary OCT caused by blood residues and gas bubbles can obscure vessel features and prompt re-acquisition. The authors introduce ArcNet, a two-branch network that fuses Cartesian and polar image representations to classify each $A$-line into none, mild, or severe artifacts. They deploy a loss combining cross-entropy, soft Dice, and total-variation regularization, along with stratified sampling to balance training. On 7,543 annotated frames, ArcNet attains strong frame-level F-scores (0.77 for mild, 0.94 for severe) with approximately 6 seconds to process a full pullback, outperforming polar-only baselines. The study demonstrates that mixed Cartesian–polar representations enhance automatic artifact assessment and could guide image-acquisition decisions in intracoronary OCT.

Abstract

In intracoronary optical coherence tomography (OCT), blood residues and gas bubbles cause attenuation artifacts that can obscure critical vessel structures. The presence and severity of these artifacts may warrant re-acquisition, prolonging procedure time and increasing use of contrast agent. Accurate detection of these artifacts can guide targeted re-acquisition, reducing the amount of repeated scans needed to achieve diagnostically viable images. However, the highly heterogeneous appearance of these artifacts poses a challenge for the automated detection of the affected image regions. To enable automatic detection of the attenuation artifacts caused by blood residues and gas bubbles based on their severity, we propose a convolutional neural network that performs classification of the attenuation lines (A-lines) into three classes: no artifact, mild artifact and severe artifact. Our model extracts and merges features from OCT images in both Cartesian and polar coordinates, where each column of the image represents an A-line. Our method detects the presence of attenuation artifacts in OCT frames reaching F-scores of 0.77 and 0.94 for mild and severe artifacts, respectively. The inference time over a full OCT scan is approximately 6 seconds. Our experiments show that analysis of images represented in both Cartesian and polar coordinate systems outperforms the analysis in polar coordinates only, suggesting that these representations contain complementary features. This work lays the foundation for automated artifact assessment and image acquisition guidance in intracoronary OCT imaging.

Figures (4)

• Figure 1: Examples of two optical coherence tomography images in the Cartesian domain and their polar counterpart, with annotated artifacts. The first and third image from the left show frames in Cartesian coordinates and the second and fourth in polar coordinates. Mild artifacts are indicated in yellow and severe in blue. Arcs in the Cartesian domain correspond to straight segments in polar coordinates, where the columns of the image are equivalent to the original A-lines. In polar coordinates the artifacts and their shadows are vertically aligned.
• Figure 2: Schematic of the ArcNet model. The top branch takes as input the Cartesian image, while the polar transformed image is fed to the bottom branch. Blocks represent feature dimensions (blue for Cartesian features and green for polar). $[\, H,W\, ]$ and $[\, \varrho, \vartheta\, ]$ are image sizes, F is feature size and it refers to the output of the last convolution before the concatenation with the features from the other domain. Output has size $[\, \vartheta, 3\, ]$, where each element in $\vartheta$ represents an A-line and 3 represents the number of classes.
• Figure 3: Example predictions of the ArcNet. Mild artifacts are indicated in yellow and severe in blue. External ring shows reference annotations and internal shows the prediction of our method. 1) Red thrombus does not trigger false positives, 2) Complex sample with no artifact, but presenting sidebranches, plaques, highly-mixed non-attenuating blood and tangential signal drop-off, 3-4) Mild attenuation due to gas bubbles, 5-6) Severe blood artifacts, 7) Small mild artifacts at 10 o'clock are not predicted, likely due to low input image resolution, 8) Imprecise segmentation of part of the severe artifacts 9-10) Near-perfect segmentation of very severe blood artifacts.
• Figure 4: Confusion matrices of the four ArcNet models (ArcNet Polar, ArcNet Single, ArcNet One-Way, and ArcNet) showing the predicted class percentages for each true class (none, mild, severe) at the A-line level. The matrices are normalized to display percentages.