Brain surface analysis has emerged as a transformative approach in medical imaging, enabling detailed characterization of structural variations tied to neurodegenerative diseases, cognitive decline, and neurodevelopment. Unlike traditional volumetric neuroimaging, surface-based methods capture the cortex’s intricate geometry, offering superior interpretability and sensitivity to subtle morphological changes―critical for studying conditions like Alzheimer’s disease (AD), where early pathology manifests as localized cortical thinning or folding alterations.

In the first part of this thesis, we present "Enhancing Multimodal Image-Based Classification of Alzheimer’s Disease with Surface Information", where we demonstrated that integrating cortical surface features with multimodal neuroimaging data (FDG-PET, amyloid PET, and tau PET) significantly improves AD detection accuracy. Our findings underscored the untapped potential of surface representations to complement conventional volumetric and metabolic biomarkers, achieving state-of-the-art classification performance.

Building on this foundation, the second part introduces Mamba-Attention Surface Analysis (MASA), a novel framework that unites Mamba (a selective state-space model) with attention mechanisms to address fundamental limitations in cortical surface modeling. While existing deep learning approaches struggle to encode spatial relationships across the cortex, and Mamba-based methods ignore explicit neighborhood dependencies, MASA dynamically adapts to local surface geometry, preserving multiscale spatial hierarchies without compromising computational efficiency. This advancement not only refines surface-based feature extraction but also enables richer interpretations of disease-related cortical patterns.

Together, these contributions redefine the role of brain surface analysis in neuroimaging: the first establishes its empirical value in multimodal transformer-based diagnostics, while the second provides a scalable, anatomically grounded framework for future research. By bridging methodological innovation with clinical applications - from AD classification to developmental studies - this work opens new avenues for understanding brain structure and dysfunction at unprecedented resolution.