Second-order Statistical Image Texture Features from Spectral X-ray Images when using Photon-Counting Spectral Detectors

Radiomics provides a promising solution by extracting high-dimensional quantitative features from medical images, enabling the capture of subtle patterns not discernible to the human eye. For example, in mammography, radiomic features such as gray-level co-occurrence matrices (GLCM), run-length and size-zone statistics, and deep-learning-derived texture features have been applied to assess breast tissue heterogeneity and density. Among these, GLCM features have been widely applied to describe parenchymal texture, capturing information on contrast, correlation, entropy, and higher-order statistics such as cluster shade, cluster prominence, and cluster tendency. In this work, we will explore the changes in these statistical features when a photon-counting spectral area detector is used. Laboratopry experiments with our benchtop system will be combined with in-house developed algorithms. These studies will offfer insights into separting texture features from detector noise and other related artifacts in spectral domain. Our studies may provide avenues to classify materual properties that are otherwise not readily apparent in images for heterogenous complex objects. Applications would be in medical or materials image classification such as for breast density classification from a single-shot in mammography.