Multifractal characterization of grayscale histopathological images: unveiling patterns linked to metastasis in breast cancer

Abstract

Introduction: Breast cancer, a pervasive global malignancy, demands precise prognostication of metastasis risk for personalized therapeutic strategies and enhanced survival rates. In pursuit of refined diagnostic methodologies, this study employs multifractal analysis on grayscale histopathological images, revealing distinctive patterns associated with metastasis occurrence.

Aim: to analyze the multifractal spectra of grayscale images for groups with and without metastasis to assess the utility of this analytical approach in enhancing the diagnostic process.

Materials and methods: The study included 102 female patients treated in the same year (1993) at the Institute for Oncology and Radiology of Serbia. Histopathological samples were immunostained with a pan-cytokeratin antibody and digitized with a high-resolution scanner, from which a specialist chose representative parts, thus leading to a total number of 519 images (418 in the no-metastasis and 101 in the metastasis group). Images were subjected to multifractal analysis, assessing the generalized dimension, Hölder exponent, and singularity spectra.

Results: Statistical comparisons between groups with and without metastasis unveil significant differences in the negative domains of both generalized dimension and Hölder exponent spectra, highlighting the influence of fine structures in tissue morphology that are linked to metastatic risk.

Conclusion: Multifractal analysis applied to images of histopathological samples from breast tumors demonstrates the ability to differentiate between groups of patients with and without metastasis. While caution is warranted regarding image resolution limitations and immunostaining sensitivity, this method is a non-training-dependent approach with potential diagnostic significance and possible synergies with advanced neural network approaches.