Nonlinear Optical Imaging for Tissue Characterization and Cell Classification

In this thesis, we investigated human skin's mechanical properties, microstructural organization, and nonlinear optical characteristics of collagen fibrils and fibers. It also introduces a rapid diagnostic method for differentiating leukocytes in bronchoalveolar lavage fluid (BALF) using advanced imaging techniques and deep learning. By integrating advanced imaging techniques, biophysical modeling, and computational analysis, this work establishes a novel framework that bridges fundamental biophysics research with clinical applications. Using two‑photon excited autofluorescence (2PEF) and second harmonic generation (SHG) microscopy, we reveal the three‑dimensional arrangement of collagen and elastin fibers in human dermal samples (Chapter 3 and chapter 4). By subjecting these samples to uniaxial stretching, we quantified key mechanical properties—including initial and elastic Young’s modulus, maximal stress and strain—uncovering significant variability among specimens. Correlation analysis between mechanical properties and microstructural features offered insights into how fiber organization influences tissue mechanics. Notably, our high-speed imaging captured, for the first time, the real-time reorientation of collagen and elastin fibers under tensile loading. We observed that collagen fibers continuously reorienting during the whole stretching, with sharp alignment in the linear phase, matching rising stress. In contrast, elastin fibers rearranged continuously from heel to linear phase, indicating their role in initiating deformation and modulating stress distribution. This challenges classical models that attribute skin mechanics to fiber unfolding and alignment in the heel region, followed by stretching in the linear phase. These findings refine our understanding of tissue behavior under stress and have implications for biomimetic material design, surgical planning, and wound healing optimization. Building on these findings, we further explored the collagen’s substructure by analyzing its nonlinear optical properties using polarization‑resolved second harmonic generation (PSHG) microscopy (Chapter 5). We developed a pixel-based quantification method in MATLAB. This method was applied to both artificial collagen matrices and tissue samples from healthy and scarred skin. Our focus was on assessment of fibril alignment and packing under varying collagen I/III ratios. By converting complex imaging data into quantitative, reproducible metrics, this method provides detailed structural insights into the extracellular matrix, which are critical for the early diagnosis and monitoring of fibrotic disorders. Additionally, it offers valuable guidance for tissue remodeling strategies following injury or surgery. Finally, in Chapter 6, we addressed a clinical challenge by developing a deep learning‑based approach for three‑dimensional leukocyte differentiation in BALF and blood samples. Leveraging label‑free higher harmonic generation microscopy along with advanced 3D deep learning models—including ResNet 3D and Vision Transformer 3D— we achieved a predictive accuracy exceeding 96%, with our results showing strong concordance with conventional cytospin analysis (differences were less than 3%). This innovative diagnostic method offers a rapid, cost‑effective, and portable alternative, enabling more efficient quantification of BALF and blood leukocytes and facilitating the diagnosis of inflammatory conditions in clinical settings. In summary, this thesis establishes an integrative framework that combines advanced imaging techniques, quantitative analysis, and computational methods. Our findings enhance our understanding of skin behavior under stress and its nonlinear properties while also opening new avenues for applications in connective tissue disorder research, biomimetic material design, and surgical planning. Furthermore, our deep learning framework facilitates rapid, cost‑effective leukocyte quantification, significantly advancing the diagnostic process for inflammatory conditions in clinical settings.