Comparative reliability of total intracranial volume estimation methods and the influence of atrophy in a longitudinal semantic dementia cohort

BACKGROUND AND PURPOSE: Total intracranial volume (TIV) as a measure of premorbid brain size is often used to correct volumes of interest for interindividual differences in magnetic resonance imaging (MRI) studies. We directly compared the reliability of different TIV estimation methods to address whether such methods are influenced by brain atrophy in the neurodegenerative disease, semantic dementia. METHODS: We contrasted several manual approaches using T1-weighted, T2-weighted, and proton density (PD) acquisitions with 2 automated methods (statistical parametric mapping 5 [SPM5] and FreeSurfer [FS]) in a cohort of semantic dementia subjects (n = 11) that had been imaged longitudinally. RESULTS: Novel mid-cranial sampling of either PD or T2-weighted images were least susceptible to atrophy: of these, the PD method was both more precise and more user-friendly. SPM5 also produced good results, providing automation for only a small loss in precision compared to the best manual methods. The T1 method that underestimated TIV as atrophy progressed was the least reproducible and the most labor-intensive. Fully automated FS overestimated TIV with progressive atrophy, and the results were even worse after optimizing the transformation. CONCLUSION: The mid-cranial sampling of PD images achieved the best combination of precision, reliability, and user-friendliness. SPM5 is an attractive alternative if the highest level of precision is not required.