Estimation of the Intracranial Volume Is Crucial in Multi-Site Studies: Reliability for Longitudinal Investigations...

Accurate estimation of the total intracranial volume (TIV) is essential in brain magnetic resonance imaging (MRI) studies, particularly for multi-site longitudinal investigations. This study assessed the validity and reliability of segmentation-based TIV (sbTIV) implemented in FreeSurfer version 7.2 for large-scale multi-site MRI data, by comparing it with the widely used estimated TIV (eTIV). We analyzed 6524 structural MRI scans from two multi-site projects in Japan, consisting of 30 procedures across 21 sites, 13 MRI machine types, 3 vendors, and 4 protocol categories. We tested the intraclass correlation coefficients (ICCs) between eTIV and sbTIV for each procedure and identified procedural factors affecting these ICCs using a general linear model. Machine- and protocol-specific biases were considered by a traveling subject harmonization approach. To specifically examine the reliability and validity of the longitudinal scans, we employed a general linear mixed model (GLMM). Overall agreement between eTIV and sbTIV was good (ICC = 0.78) but varied across procedures (0.62-0.94). The 1.0 mm isotropic protocol showed the highest reliability. Notably, there was poor consistency in participants with eTIV values of 120,000 mm³ or smaller (ICC = 0.053). sbTIV demonstrated superior cross-procedural consistency in adolescent and adult longitudinal scans compared to eTIV. In longitudinal scans, sbTIV showed greater sex difference and sex-specific increase for adolescents, and greater consistency for adults, compared to eTIV. sbTIV offers more robust and reliable estimation compared to eTIV, particularly for multi-site longitudinal studies. These findings highlight the need for careful consideration when interpreting previous multi-site studies using eTIV.