We automatically segmented the hippocampus in 400 brain MRI scans from the Alzheimer's Disease (AD) Neuroimaging Initiative, combining AdaBoost with a novel model, the Auto Context Model (ACM). Trained on 21 hand-labeled segmentations, ACM created binary hippocampus maps in 100 controls, 200 with mild cognitive impairment (MCI), and 100 AD subjects, (age: 75.8+/-6.6SD). A radial atrophy mapping technique computed average parametric surface models and local statistics of atrophy. We visualized correlations between regional atrophy and diagnosis (MCI v. controls: p = 0.008; MCI v. AD: p = 0.001), mini-mental state exam scores, and clinical dementia rating scores (CDR; all p < 0.0001, corrected). Based on false discovery rate curves in gradually reduced samples, 40 subjects were sufficient to correlate atrophy and CDR scores; MCI and AD were distinguishable with N = 304.