Clustering Methods for the Characterization of Synchrotron Radiation X‐Ray Fluorescence Images of Human Carotid Atherosclerotic Plaque

Abstract

This study employs computational algorithms to automatically identify and classify features in X‐Ray fluorescence (XRF) microscopy images. Principal component analysis (PCA) and unsupervised machine learning algorithms, such as Gaussian mixture (GM) clustering, are implemented to label features on a collection of XRF maps of human atherosclerotic plaque samples. The investigation involves the hard X‐Ray nanoprobe (NanoMAX) at MAX IV synchrotron radiation facility, utilizing scanning transmission X‐Ray microscopy (STXM) and XRF techniques. The analysis covers regions of interest scanned by the beam with a step size of 200 nm, yielding XRF maps of elements like calcium, iron, and zinc. These maps reveal intricate structures unsuitable for manual labeling. However, they can be accurately classified in an automated fashion using GM. Prior to clustering, PCA is used to deal with repeated patterns and background areas. The resulting clusters are associated with different types of features, which can be identified as specific tissues confirmed by histology. Regions of high concentrations of phosphorus, sulfur, calcium, and iron are found in the samples. These regions are also observed in the STXM results as spots of low transmission that typically are associated with calcium deposits only.

Keywords

• atherosclerosis
• carotid plaque
• clustering methods
• unsupervised machine learning
• X‐Ray fluorescence microanalysis