Intraplaque haemorrhage quantification and molecular characterisation using attention based multiple instance learning

Leveraging the Athero-Express biobank, an ongoing study comprising a comprehensive dataset of histological, transcriptional, and clinical information from 2,595 carotid endarterectomy patients, we developed an attention-based additive multiple instance learning (MIL) framework to automate the detection and quantification of IPH across whole-slide images of nine distinct histological stains. We demonstrate that routinely available Haematoxylin and Eosin (H&E) staining outperformed all other plaque relevant Immunohistochemistry (IHC) stains tested (AUROC = 0.86), underscoring its utility in quantifying IPH.

When combining stains through ensemble models, we see that H&E + CD68 (a macrophage marker) as well as H&E + Verhoeff-Van Gieson elastic fibers staining (EVG) leads to a substantial improvement (AUROC = 0.92). Using our model, we could derive IPH area from the MIL-derived patch-level attention scores, enabling not only classification but precise localisation and quantification of IPH area in each plaque, facilitating downstream analyses of its association and cellular composition with clinical outcomes.

By doing so, we demonstrate that IPH presence and area are the most significant predictors of both preoperative symptom presentation and major adverse cardiovascular events (MACE), outperforming manual scoring methods. Automating IPH detection also allowed us to characterise IPH on a molecular level at scale. Pairing IPH measurements with single-cell transcriptomic analyses revealed key molecular pathways involved in IPH, including TNF-α signalling, extracellular matrix remodelling and the presence of foam cells.

This study represents the largest effort in the cardiovascular field to integrate digital pathology, machine learning, and molecular data to predict and characterize IPH which leads to better understanding how it drives symptoms and MACE. Our model provides a scalable, interpretable, and reproducible method for plaque phenotyping, enabling the derivation of plaque phenotypes for predictive modelling of MACE outcomes. Project specific data For this project we share the DGE and GWAS results, as well as the unedited raw results that came directly from CellProfiler through the slideToolKit pipeline.

These data are stored on UMC Utrecht server for reference and sharing with others (see below Important notice on availability of data). About Athero-Express Biobank The AE started in 2002 and now includes over 3,500 patients who underwent surgery to remove atherosclerotic plaques (endarterectomy) from one (or more) of their major arteries (majority carotids and femorals). The AAA started in 2003 and now includes over 1,000 patients who underwent open surgery on arterial aneurysms, the majority on aortic aneurysms.

The staining protocols are described by Verhoeven et al. (AE) and Hurks et al. (AAA). About ExpressScan The ExpressScan is an ongoing, unfunded project to scan pathological slides of atherosclerotic plaques and aneurysm tissues at high-resolution using pathology scanners into whole-slide images (WSI). Here we describe these histological WSI data available for the Athero-Express (AE) and Aneurysm-Express (AAA) Biobank Studies.

Whole-slide images are available for the several commonly used stains, but note that these are not available in all samples in both studies. A table showing the approximate numbers of available WSI is given here. Associated ExpressScan projects - Glycophorin C: Mekke JM et al. and the associated GitHub repository. - slideEMask: entropy based tissue masker - slideNormalize: normalize WSIgithub.com/swvanderlaan/slideNormalize - ExpressScan-DeepZoom : local, on-campus webportal to inspect WSI (private, ongoing and unpublished) - ExpressScan_QC: quality control procedures to process results from slideToolKit (private, ongoing and unpublished) - EntropyMasker: improved method to automatically mask WSI using entropy (private, ongoing and unpublished); preprint available here. - DEEP-ENIGMA: a deep-learning image-segmentation project on plaques (private, ongoing and unpublished) About slideToolKit These data are also used for slideToolKit and other ExpressScan projects (see below).

Depending on the content of the project, a list of slides used is available to enable reproducible science. A link to the public GitHub repository for slideToolKit can be found here: github.com/swvanderlaan/slideToolKit. Important notice on availability of data The amount of data is huge: over 25,000 WSI on average 1Gb size per WSI.

There are also restrictions on use by commercial parties, and on sharing openly based on (inter)national laws and regulations and the written informed consent. Therefore these data (and additional clinical data) are only available upon discussion and signing a Data Sharing Agreement (see Terms of Access) and within a specially designed UMC Utrecht provided environment.