Model-free classification of X-ray scattering signals applied to image segmentation

In most cases, the analysis of small-angle and wide-angle X-ray scattering, SAXS and WAXS, respectively, requires a theoretical model to describe the sample's scattering, complicating the interpretation of the scattering from complex heterogeneous samples. This is the reason why in most cases, the analysis of the large number of scattering patterns, such as is generated by time-resolved and scanning methods, remains challenging. Here we introduce and demonstrate a model-free classification method to separate SAXS/WAXS signals based on their inflection points. We focus on the segmentation of scanning SAXS/WAXS maps, for which each pixel corresponds to an azimuthally integrated scattering curve. In such a way, the sample composition distribution can be segmented through signal classification without applying a model or previous sample knowledge. We employ dimensionality reduction and clustering algorithms to classify SAXS/WAXS signals according to their similarity. The number of clusters, i.e. the main sample regions detected by SAXS/WAXS signal similarity, is automatically estimated. From each cluster, a main representative SAXS/WAXS signal is extracted to uncover the spatial distribution of the mixtures of phases that form the sample. As examples of application, a mudrock sample and two breast tissue lesions are segmented.