Efficient Atrophy Mapping: A Single-Step U-Net Approach for Rapid Brain Change Estimation

The estimation of brain atrophy is crucial for eval-uating brain diseases and analyzing neurodegeneration. Existing methods for computing atrophy maps often suffer from lengthy processing times due to the computational cost of multi-step processing. In this work, we propose a novel technique for atrophy map calculation using a single U-net-based architecture. This approach consolidates multiple traditional medical imaging processing steps into a single process, aiming to accelerate the computational time required. Specifically, our method estimates structural changes by generating a flow map from two longitudinal Magnetic Resonance Imaging (MRI) scans of the same subject. We trained and evaluated our system on a dataset comprising 2000 Tl-weighted MRI scans sourced from two different public datasets on Alzheimer's Disease. Experimental results demonstrate a considerable reduction in execution time while maintaining atrophy mapping performance comparable to state-of-the-art solutions. We believe that our pipeline could significantly benefit clinical applications for measuring brain atrophy, especially in scenarios requiring the evaluation of large cohorts, such as clinical trials. Our code is freely available at https://github.com/Efficient Atrophy Mapping: A Single-Step U-net Approach for Rapid Brain Change Estimation.