Increasing Urban Tree Canopy Associated With Reduced Mortality: A Longitudinal Analysis of Chicago Neighborhoods.

Magnetic Resonance Imaging (MRI) is a non-invasive technique widely used in clinical and research protocols for studying human brains in-vivo. In research settings, MRI scans are commonly acquired at voxel sizes of approximately 1 mm isotropic and analyzed with automatic image processing pipelines (; ; ). However, early detection of fine brain atrophy patterns in neurodegenerative disorders (e.g., Alzheimer’s disease) requires accurate morphometry of small brain regions for which standard resolutions are suboptimal (). Acquisition voxel size is a hindering factor when performing brain morphometry, since it limits the minimum size of brain regions that would be measurable on an MRI. Therefore, it is desirable to acquire MRIs with the highest resolution possible. However, increasing acquisition resolution substantially increases noise level as well as time, noise, discomfort of scanned individuals, and makes the scan more prone to movement artifacts (). Therefore, in clinical studies and practical applications, it is uncommon to use high-resolution full brain MRI acquisitions. Furthermore, numerous legacy datasets have been scanned using standard-resolution sequences, but contain a great amount of non-imaging biomarker information. As an example. the the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (), one of the largest databases of individuals on the continuum of Alzheimer’s Disease containing longitudinal demographic, genetic, and clinical information, as well as positron-emission tomography (PET) scans, provides longitudinal MRI data acquired at ~1 mm isotropic voxel sizes. Single-Image Super-Resolution (SISR) is a technique to artificially increase the resolution of images after their acquisition, generating a high-resolution (HR) image from a single low-resolution (LR) input. SISR has gained popularity in naturalistic image processing (; ; ; ; ; ; ; ; ), as well as in medical image processing (; ; ; ; ; ). The latter implies potential valuable gains in accuracy for medical diagnostic and prognostic purposes, and could even make it possible to measure anatomical structures that would be too small to capture in the LR regime (). SISR is an ill-posed problem since a single LR sample can be upsampled into many different HR images. Therefore, caution should be taken when utilizing image-reconstruction techniques in the medical field, and thorough validations should be performed before publicly releasing an SISR method.