Title : Deepthickness: A novel deep learning method for estimating cortical thickness trajectories in healthy and alzheimer’s disease populations
Abstract:
Alzheimer's disease (AD) is a neurodegenerative disease that presents with critical challenges in diagnosis and treatment. Emerging research indicates that AD-related cortical changes, such as cortical thickness, can appear up to a decade before cognitive symptoms. Accurately measuring cortical thickness can therefore offer a significant avenue for early AD diagnosis and monitoring of clinical progression. Automatic techniques, such as FreeSurfer and CAT12 Toolbox, offer out-of-the-box cortical thickness estimates, but with an excessively long computational time (up to 10 hours per volume), systematic differences between approaches and significant errors when applied to clinical data. We propose DeepThickness; the first Deep Learning-based approach for estimating cortical thickness from structural MRI in just a few seconds. Our method utilises recent advances in deep learning to generate white matter and pial surface mesh reconstructions with cortical thickness estimates as an overlay. We report promising preliminary findings, highlighting our method’s similarity to FreeSurfer in mesh generation and cortical thickness estimations while accounting the software's identified limitations. Leveraging comprehensive clinical datasets, we also showcase our method’s use for mapping cortical thickness, cognition and other clinically relevant trajectories over time for healthy, MCI and AD populations.
