Introduction:
Fluorescence lifetime imaging (FLIm) allows label-free biochemical characterization of atheroma, however, multispectral FLIm produces massive data throughput making image interpretation problematic. We investigated whether machine learning algorithm can be applied to intravascular OCT-FLIm for automated biochemical characterization of coronary plaques.
Methods and Results:
We built a fully-integrated, high-speed OCT-FLIm system and a low-profile, dual-modal imaging catheter that can provide high-resolution OCT images and correctly co-registered multispectral fluorescence lifetime (FL) readouts: ch.1 and ch.2 FL, and FL intensity ratio (ch.2/ch.1). Rapid intracoronary imaging (10-20 mm/s, 100 rps) was safely performed in atheromatous pigs. There were significant differences in FL measurements according to plaque types (high-risk vs. fibrotic: p<0.001). Multispectral FL measurements sampled selectively from histologically-proven plaque components (lipid, macrophage, lipid+macrophage, SMC) were analyzed. Each component was distinguishable from one another either by difference of FLs or intensity ratio. Random forest classifier (RFC), trained with component-labeled multispectral FL dataset, accurately classified key biochemical components of atherosclerotic plaques. RFC-determined biochemical characteristics of target plaque were consistent across two repeated imaging data (intraclass correlation, p<0.0001) and corroborated closely with those derived from quantitative immunohistochemistries.
Conclusions:
Our OCT-FLIm incorporating RFC-based systematic multispectral FL analysis could provide high-resolution plaque imaging with automated biochemical characterization in beating coronary environment. The present imaging strategy enabling comprehensive characterization of multiple components of atherosclerotic plaques will open a new avenue in the field of cardiovascular imaging.
Timing and Operating Mode Design for Time-Gated Fluorescence Lifetime Imaging Microscopy