Effects of different intracranial volume correction methods on univariate sex differences in grey matter volume and multivariate sex prediction

Abstract

Sex differences in 116 local gray matter volumes (GMVOL) were assessed in 444 males and 444 females without correcting for total intracranial volume (TIV) or after adjusting the data with the scaling, proportions, power-corrected proportions (PCP), and residuals methods. The results confirmed that only the residuals and PCP methods completely eliminate TIV-variation and result in sex-differences that are “small” (∣d∣ < 0.3). Moreover, as assessed using a totally independent sample, sex differences in PCP and residuals adjusted-data showed higher replicability ($$\approx $$ ≈ 93%) than scaling and proportions adjusted-data $$( \approx $$ ( ≈ 68%) or raw data ($$\approx $$ ≈ 45%). The replicated effects were meta-analyzed together and confirmed that, when TIV-variation is adequately controlled, volumetric sex differences become “small” (∣d∣ < 0.3 in all cases). Finally, we assessed the utility of TIV-corrected/ TIV-uncorrected GMVOL features in predicting individuals’ sex with 12 different machine learning classifiers. Sex could be reliably predicted (> 80%) when using raw local GMVOL, but also when using scaling or proportions adjusted-data or TIV as a single predictor. Conversely, after properly controlling TIV variation with the PCP and residuals’ methods, prediction accuracy dropped to $$\approx $$ ≈ 60%. It is concluded that gross morphological differences account for most of the univariate and multivariate sex differences in GMVOL

A comprehensive atlas of white matter tracts in the chimpanzee

“…However, reported sex differences in fasciculi between adult male and female humans are small and limited to the left SLF [55]. Group sex differences found in human brain morphometry have reproducibility problems [56][57][58], and to the degree to which they are reproducible, are not categorical differences but rather differences of mean values, with high degrees of population overlap (e.g., [59]; for reviews, see [60,61]). Crucially, sex differences are often a matter of total volume and not anatomical organization of fascicular projections, and therefore we would not expect our anatomical descriptions to change if male chimpanzee scans were added to the dataset.…”

Analyzing brain data by sex: Are we asking the right question?

“…We would like to stress that the conclusion that sex category is not a major predictor of variability in human brain structure does not contradict evidence that sex-related genes and hormones affect specific brain measures (for a recent review see 1 ), nor evidence that supervised machine learning algorithms may use sex-related variability in brain structure to predict the sex category of a brain’s owner (e.g., 23-26 ). Indeed, one such approach (logistic regression, as in 23 ) over the 289 brain measures analyzed in the present study, accurately predicted the sex category of brains’ owners in 75% of cases (see Supplemental Materials; The lower classification rate compared to previous studies (e.g., 23-25 ) is expected given that we used data “corrected” for total brain size ( 26 )).…”

A comprehensive atlas of white matter tracts in the chimpanzee

“…However, reported sex differences in fasciculi between adult male and female humans are small and limited to the left SLF [54]. Group sex differences found in human brain morphometry have reproducibility problems [55][56][57], and to the degree to which they are reproducible, are not categorical differences but rather differences of mean values, with high degrees of population overlap (e.g., [58]; for reviews, see [59,60]). Crucially, sex differences are often a matter of total volume and not anatomical organization of fascicular projections, and therefore, we would not expect our anatomical descriptions to change if male chimpanzee scans were added to the dataset.…”