P3–085: Hippocampal texture predicts conversion from MCI to Alzheimer's disease

The purpose of this study was to investigate whether baseline hippocampal MRI texture predicts conversion from MCI to AD after one year. A standardized subset of the ADNI database recently released by ADNI, comprising 169 normal controls (CTRL), 233 MCI, and 101 AD, was considered. The MCIs were further subdivided into AD converters (MCI-C, 41) and non-converts (MCI-NC, 192) after one year. Segmentations of the hippocampi obtained using cross-sectional FreeSurfer (v5.1.0) were used to define the region of interest (ROI) in each baseline 1.5T T1-weighted MRI scan. A texture-based marker that has demonstrated good diagnostic capabilities in a previous study was trained to separate CTRL from AD, and it was subsequently applied to score the MCI-Cs and the MCI-NCs. The hippocampal fraction (HF) defined as hippocampal volume divided by intracranial volume (ICV) was also computed based on the same ROI and on FreeSurfers estimate of ICV. Two markers were evaluated, hippocampal texture in isolation and a logistic regression model combining texture, HF, and age. Results were reported by ROC-analysis of MCI-C vs MCI-NC, testing for significance using DeLong, DeLong and Clark-Pearson's test (P<0.05 was regarded as significant). ROC-curves for prognosis of conversion for the two markers are shown in the Figure, and the corresponding AUCs were for texture in isolation 0.731 (P<0.001) and for the combined marker 0.754 (P<0.001). Texture, HF, and age were all significant in the logistic regression model with the following P-values: 0.00001, 0.00166, and 0.00011.