Background: Atherosclerosis, an underlying cause of cardiovascular diseases, is achronic inflammatory condition resulting in the accumulation of plaque composedof lipids and other bio compounds within artery walls. Thus, narrowing thearterial lumen and leading to the blockage of blood flow, and rupture of thearteries. Atherosclerosis is known to be an inevitably progressive disease, resultingin an increase in inflammation and lipid accumulation with age. Through thecombination of coherent anti-stokes Raman Scattering (CARS) microscopy, anon-linear optical microscopy modality, and an automated pipeline for plaqueclassification; the stages of plaque progression can be assessed in a label-freemanner. The pipeline provides a basis for recognizing changes in the progressionand/or stabilization of atherosclerotic lesions.
Results: The use of machine learning in microscopy has been increasinglyallowing the classification of large amounts of images based on specific featuresrelevant to different applications. Based on a set of label-free CARS images ofatherosclerotic plaques (i.e. foam cell clusters) from a rabbit model, wedeveloped an automated pipeline to classify lesions based on their majormorphological features. Through the combination of image preprocessing andsegmentation, feature extraction and supervised machine learning algorithms, theclassification pipeline showcased the ability to exploit relevant plaquemorphological features to accurately classify 3 pre-defined stages ofatherosclerosis: early fatty streak development (EFS), early fibroatheroma (EF)and advancing atheroma (AA), greater than 85% class accuracy. Conclusions: Minute changes in the morphology of plaque can often beoverlooked. Through the combination of CARS microscopy and computationalmethods, a powerful classification tool was developed to identify the progressionof atherosclerotic plaque in an automated manner. The ability to differentiateamongst EFS, EF and AA present the opportunity to classify the onset ofatherosclerosis at an earlier stage of development, as well as greatly improvingthe potential of tracking effectiveness of novel therapeutic interventions