Intravascular optical coherence tomography image analysis method

Intravascular optical coherence tomography (IVOCT) has the resolution and contrasts necessary to identify coronary artery plaques. Currently, segmentation of images and identification of plaque composition are typically done manually. We have created a method for automated plaque classification using tissue optical characteristics and textures. Altogether, we used over 13,500 images from both manually annotated clinical IVOCT data and ex-vivo IVOCT pullback data annotated accurately using a novel approach with 3D microscopic cryo-imaging. Using 5-fold stratified cross validation on user selected volumes of interest, accuracy was 92.5% with area under the curve of 0.98, 0.99, 0.99 for calcium, lipid and fibrous, respectively. With the classifier fixed, there was good agreement between pixel-based classification and annotated IVOCT ex vivo image data. Results encourage us to pursue fully automated processing of IVOCT.