AI-Driven Analysis of Microscopy Data for Medical Decision Support

Abstract

Recent advances in deep learning have significantly expanded the possibilities of biomedical image analysis, particularly in automated and precise segmentation, classification, and tracking of cellular structures. This dissertation focuses on the computational analysis of microscopy image data acquired under real laboratory and diagnostic conditions. The analyzed datasets include both bright-field and fluorescence microscopy modalities, encompassing time-lapse imaging, low-contrast conditions, and complex biological environments.

The thesis proposes robust and interpretable methods for image data analysis aimed at cell detection, segmentation, classification, and counting—especially in challenging scenarios involving irregular cell shapes and high noise levels in the input data. It introduces a novel methodology for quantitative evaluation of wound-healing assays based on sector-wise spatial analysis and cell population tracking. Furthermore, it presents a comparative study of deep neural network architectures for the segmentation of chronic lymphocytic leukemia cells, a pipeline for automatic counting of fibroblasts and fibroblast-like cells in highly noisy bright-field microscopy images, and outlines future work focused on the application of hyperbolic neural networks for white blood cell classification from peripheral blood smears. This approach may contribute to faster and more accurate diagnostics of various hematological diseases, including leukemias.

Special emphasis is placed on fibroblast behavior during wound-healing assays. These cells exhibit heterogeneous morphology and migration patterns that require not only computational precision but also biological interpretability. By integrating deep learning techniques, classical image processing, and probabilistic modeling, the proposed approaches provide practical tools for cellular research and diagnostic applications. All proposed methods were validated on real microscopy data in collaboration with biomedical research institutions.