Understanding how our brain works and how it breaks down in disease requires studying its structure across multiple spatial scales.

This includes nanoscale synapses to macroscale brain circuits. The rise of nanoscopic and fast 3D imaging modalities with large volumetric coverage, such as Synchrotron X-ray imaging, now makes this possible. However, each scan produces terabytes of 3D data that would take years to analyze manually, if at all.

Through this grant, postdoc Mahsa Amirrashedi will explore memory-efficient, multi-scale deep learning models to segment brain microstructures of different sizes from these massive volumes in order of minutes, enabling large-scale analysis of one of the most detailed brain images to date.