Segmentation and Multimodal Characterization of Metal Particles in the Human Hippocampus Using Discrete Segmentation Algorithms and Correlation Spectral Analysis

The accumulation of metallic micro- and nanoparticles in the human hippocampus is increasingly linked to neurotoxic processes and neurodegenerative disorders. Precise segmentation and detailed characterization of these particles are crucial to understanding their role. This study presents a novel method that combines discrete segmentation based on graph-cut theory with Dinic's algorithm for the computation of maximum flow. The images are modeled as directed weighted graphs, with pixel intensities and gradients defining edge capacities, enabling robust segmentation in electron microscopy data. To ensure robustness, the method is validated against ground-truth masks, achieving a Dice coefficient of 0.97898 ± 0.0172 and an Intersection over Union (IoU) of 0.9609 ± 0.0326. Morphometric parameters-area, perimeter, circularity, and Feret diameters-are automatically extracted. Concurrently, elemental analysis using Energy-Dispersive X-ray Spectroscopy (EDS) reveals a heterogeneous composition, including iron-rich particles and compounds containing nickel and chromium. The observed variability highlights the importance of single-particle analysis in better understanding the neurobiological impact of metallic deposits.