Overview

Intravascular polarimetry complements the high-resolution images of conventional intravascular optical coherence tomography (OCT) with quantitative measures of tissue polarization properties related to tissue composition. However, additional metrics further complicate image interpretation and currently rely on tedious manual segmentation to leverage the quantitative polarization metrics. Here, we developed a customized convolutional neural network multi-class segmentation model to detect the intima-media boundary whenever visible and otherwise identify the presence of attenuating plaque that masks this boundary. Combined with the IV-OCT backscatter signal, the polarization metrics that enhance the discrimination of targeted features improve the accuracy and robustness of automated segmentation and lesion identification.