Automated Quality Control for Intravascular Optical Coherence Tomography with Limited Labels

Yi Chen, Jiawen Li, Minh-Son To, Peter James Psaltis

Keywords: intravascular optical coherence tomography, quality control, artifacts, label- efficient learning

TL;DR: Automated Quality Control for Intravascular Optical Coherence Tomography with Limited Labels

Registration Requirement: Yes

Abstract: Artifacts in intravascular optical coherence tomography (IVOCT) imaging can obscure vessel-wall visualization and compromise downstream plaque assessment, yet automated quality control (QC) remains insufficiently developed. We present a label-efficient framework that retains expert-reviewed samples as anchors during progressive refinement of unlabeled data and uses a small calibration subset to derive class-specific decision thresholds, enabling more reliable decision-making under limited labels. On a clinical IVOCT dataset, it achieves strong artifact recognition together with a macro-specificity of 0.9536 and an expected calibration error (ECE) of 0.0678, with stronger specificity and calibration than seed-only supervision and generic semi-supervised learning baselines on the same backbone. These findings support automated QC as a practical front-end step for IVOCT analysis.

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 38