Advancing brain MRI analysis in aging and disease with deep learning

This thesis develops deep learning approaches to enhance brain MRI analysis across three key areas:
dMRI analysis, interpretable AI for aging and dementia studies that minimize the effects of confounding,
and clinical translation of AI-driven neuroimaging methods. We introduce automated, robust techniques
for white matter bundle segmentation and spatial registration of dMRI, improving tract-specific analysis
in both cross-sectional and longitudinal studies, achieving comparable accuracy and robustness to
methods developed for superior-quality dMRI data while ensuring applicability to large-scale population
studies. Beyond dMRI, we address challenges in mitigating confounding effects and enhancing interpretability in deep learning-based imaging studies. Lastly, we bridge the gap between AI
advancements and clinical applications by developing tailored frameworks for real-world challenges,
including longitudinal glioma analysis and cerebral small vessel disease detection. By advancing these
methodologies, this thesis aims to improve the accuracy, reproducibility, and clinical utility of AI-driven
neuroimaging analysis, supporting population studies and neurological disease research.