Correction for head size in brain-imaging measurements

Abstract

Structural brain-imaging measurements based on computed tomography (CT) or magnetic resonance imaging (MRI) are often corrected or adjusted for normal variation in head size. Some methods of head-size correction, such as the ventricle-brain ratio (VBR), are based on taking the brain structure size as a proportion of the estimated head size, while other methods have used a regression model to obtain head-size residualized structure measures. Recently, head-size correction was shown to result in less reliable volumetric measures of brain structures (Arndt et al., 1991). In the present study, MRI was used to examine the effects of head-size correction on the interrater reliability of volumetric measures of gray matter, white matter, and cerebrospinal fluid. Four raters independently scored MRI brain images from 26 subjects, generating separate estimates of head size and region of interest (ROI) size. Two methods were used to correct MRI values for differences in head size, one based on proportions and the other based on linear regression. Results confirmed that head-size correction did produce measures with lower reliability; however, further analysis based on classical measurement theory showed that the lower reliability was attributable not only to increased measurement error variance, but also to reduced true score variance. Subsequent analyses of criterion validity compared the raw (uncorrected) and head-size-corrected ROI measures in terms of their correlations with age in a sample of 43 normal control subjects, and in terms of their ability to differentiate schizophrenic patients (n = 22) from normal control subjects (n = 20). Results indicated that head-size correction often improved criterion validity, producing higher correlations with age and with diagnostic status than those produced by the raw measures. These findings suggest that head-size correction removes irrelevant true-score variance which reduces reliability yet improves the correlations with validity criteria such as age and diagnostic status.