To date, no stem cell therapy has been directed to specific recipients—and, conversely, withheld from others—based on a clinical or molecular profile congruent with that cell's therapeutic mechanism-of-action (MOA) for that condition. We address this challenge preclinically with a prototypical scenario: human neural stem cells (hNSCs) against perinatal/neonatal cerebral hypoxic-ischemic injury (HII). We demonstrate that a clinically translatable magnetic resonance imaging (MRI) algorithm, hierarchical region splitting, provides a rigorous, expeditious, prospective, noninvasive "biomarker" for identifying subjects with lesions bearing a molecular profile indicative of responsiveness to hNSCs' neuroprotective MOA. Implanted hNSCs improve lesional, motor, and/or cognitive outcomes only when there is an MRI-measurable penumbra that can be forestalled from evolving into necrotic core; the core never improves. Unlike the core, a penumbra is characterized by a molecular profile associated with salvageability. Hence, only lesions characterized by penumbral > core volumes should be treated with cells, making such measurements arguably a regenerative medicine selection biomarker.