ICOBRAIN DM: A NEW METHOD FOR HIPPOCAMPUS AND BRAIN LOBE SEGMENTATION IN ALZHEIMER’S DISEASE

Brain atrophy, especially of the hippocampi, is an important diagnostic feature of Alzheimer's disease (AD). However, for diagnosis and follow-up of AD patients, automatic measurements need to be reliable and sensitive. We propose icobrain dm, an automated brain segmentation method that quantifies total gray matter, gray matter of the frontal, parietal, and temporal lobes, and hippocampal (HC) volume based on T1-weighted MR images, normalized to intracranial volume. First, the accuracy of HC segmentation of 46 subjects of a memory clinic-based research population (mean age (±SD) 72.0 (±7.8) years, 50.0% females) is evaluated against manual segmentation by an expert. Next, reproducibility of all measures is assessed in a test-retest setting, by using longitudinal scans (up to 10 days apart) from 42 cognitively healthy subjects from the OASIS3 dataset (mean age (±SD) 61.4 (±8.6) years, 59.5% females) (www.oasis-brains.org). Finally, sensitivity is determined on the ability to distinguish 46 AD patients (mean age (±SD) 71.5 (±7.2) years, 60.9% females) from 46 age-matched controls (mean age (±SD) 70.4 (±7.1) years, 47.8% females), using the publicly available MIRIAD dataset (miriad.drc.ion.ucl.ac.uk). The diagnostic criteria of McKhann 1984 (1) were used to diagnose AD, while controls did not have cognitive impairment. Results of icobrain dm (v.4.2) are compared to FreeSurfer (v.6.0), a widely used and fully automated algorithm for measuring brain volumes. HC segmentation is highly accurate for icobrain dm when compared to manual segmentation and more accurate than FreeSurfer (Table 1). All segmentations presented with good reproducibility levels, with icobrain dm performing better than FreeSurfer (Table 2). All measures were significantly different (P<.001) between AD patients and controls and distinguished AD from controls with high accuracy, for both icobrain dm and FreeSurfer. The highest area under the curve value (AUC) was found for temporal lobe volume measured by icobrain dm (AUC=0.9896), which was significantly higher than FreeSurfer's AUC (AUC=0.9565) (Table 3).