IC-P-102

(manual segmentation of baseline hippocampi only). Template-derived regions were not devised to be accurate but rather aimed to identify the region approximately. Results: Atrophy rates (mm loss year) are shown in the table and figure. The automated method was a significant predictor of disease (p 0.001) and gave similar group discrimination compared with semi-automated and manual methods, although there was a suggestion that a combination of the manual and semi-automated or fully-automated methods may improve group discrimination (p 0.03). Manual and semiautomated rates were also calculated on a relative scale, as a percentage of baseline hippocampal volume (see table). Automated atrophy rates were not calculated relatively owing to the low accuracy of the template-based segmentation method; mean (SD) similarity of overlap (intersection/union) of template segmentation with manual, was 0.49 (0.05) in controls and 0.43 (0.07) in AD. Adjusting for the absolute rate, there was no evidence that the manual relative rate added predictive information (p 0.12). There was borderline evidence that relative rates added predictive information, adjusting for absolute rates, using semi-automated methods (p 0.051). Automated hippocampal analysis in this study took 20 minutes per hippocampal pair on a 3.4 GHz Intel Xeon server, whereas manual delineation of each hippocampal pair took 90 minutes of operator-intensive labor. Conclusions: The automated method we describe may provide reliable atrophy measures of the hippocampus with reduced operator time and as a result may be useful in studies where analysis of many scan-pairs is required.