Automated characterization of plaque tissue histology can be used to find distinct associations between tissue types and underlying biological processes and reveal spatial biomarkers using MIL for plaque vulnerability. CNNs are Deep learning (DL) models that learn to distinguish patterns specific to the training images, histology in our case. These learned patterns enable the classification of small patches in the histology into one of several classes, in our case tissue types that are of interest to plaques, such as LRNC, IPH, inflammation and calcification. The result is a segmented image in which every small tissue patch is assigned a class, revealing the spatial location of relevant tissue types and their organisation. From there, methods such as Multiple Instance Learning (MIL) allow for the creation of a map that indicates the location of the patches relevant to particular questions, such as that of plaque stability. All images in BiKE will be segmented for relevant tissue types, and a subset (n=200) will be analysed by correlating their segmentation with BulkRNA for relevant correlated genes, followed by MIL to create “attention maps” that will point to regions of interest to plaque stability and symptomaticity for CVD.