Abstract
BACKGROUND: Intravascular optical coherence tomography (IVOCT) enables detailed plaque characterisation in vivo, but visual assessment is time-consuming and subjective. AIMS: This study aimed to develop and validate an automatic framework for IVOCT plaque characterisation using artificial intelligence (AI). METHODS: IVOCT pullbacks from five international centres were analysed in a core lab, annotating basic plaque components, inflammatory markers and other structures. A deep convolutional network with encoding-decoding architecture and pseudo-3D input was developed and trained using hybrid loss. The proposed network was integrated into commercial software to be externally validated on additional IVOCT pullbacks from three international core labs, taking the consensus among core labs as reference. RESULTS: Annotated images from 509 pullbacks (391 patients) were divided into 10,517 and 1,156 cross-sections for the training and testing data sets, respectively. The Dice coefficient of the model was 0.906 for fibrous plaque, 0.848 for calcium and 0.772 for lipid in the testing data set. Excellent agreement in plaque burden quantification was observed between the model and manual measurements (R2=0.98). In the external validation, the software correctly identified 518 out of 598 plaque regions from 300 IVOCT cross-sections, with a diagnostic accuracy of 97.6% (95% CI: 93.4-99.3%) in fibrous plaque, 90.5% (95% CI: 85.2-94.1%) in lipid and 88.5% (95% CI: 82.4-92.7%) in calcium. The median time required for analysis was 21.4 (18.6-25.0) seconds per pullback. CONCLUSIONS: A novel AI framework for automatic plaque characterisation in IVOCT was developed, providing excellent diagnostic accuracy in both internal and external validation. This model might reduce subjectivity in image interpretation and facilitate IVOCT quantification of plaque composition, with potential applications in research and IVOCT-guided PCI.