Development of Deep Learning Methods for Dendritic Spine Segmentation and Morphological Analysis

Dendritic spine morphology is a key indicator of synaptic function and neural plasticity, and its alterations are strongly associated with neurodevelopmental disorders such as Fragile X Syndrome. This thesis presents an experimental study on deep learning-based methods for the segmentation and morphological analysis of dendritic spines, focusing on human neuronal samples engrafted in murine models. The work was conducted following an internship experience and is based on the use of ZEISS arivis for the management, processing, and analysis of high-resolution microscopy datasets. The images analyzed derive from human neurons transplanted into mice, providing a biologically relevant in vivo model for the investigation of human neuronal morphology. A comparative evaluation was performed within the ZEISS arivis environment, contrasting deep learning-based segmentation workflows with conventional image processing approaches. The study focuses on assessing segmentation performance in terms of efficiency, robustness, and adaptability to complex three-dimensional biological structures. The results highlight that deep learning-based approaches significantly improve processing speed, scalability, and consistency in dendritic spine detection and morphological quantification. These methods demonstrate strong potential for enhancing the accuracy and reproducibility of large-scale morphological analyses. Overall, this thesis highlights the advantages of deep learning-based methods for automated image segmentation in computational neuroscience and underscores their potential for advancing the study of synaptic morphology in translational models of Fragile X Syndrome.