female was observed in the lingual gyrus, precuneus, left hippocampus, bilateral parahippocampal cortex, olfactory cortex, left putamen, and left insula (Fig. 3a). No common regions for female > male were observed among the four pipelines. The two FSL pipelines shared only 13.16% overlap (Table 1), with overlapping higher GMV for females being located in the bilateral postcentral cortex, right angular, right inferior parietal lobule, and cerebellum (Fig. 3a). In contrast to the comparably small overlap between the pipelines, wide variations in the location and extent of the identified GMV sex-differences were observed specifically in medial prefrontal and occipital regions. For instance, whereas CAT revealed higher GMV in widespread cerebellar and limbic regions in men, FSLANAT and FSLVBM revealed higher GMV in widespread posterior/superior parietal regions in women (Fig. 3a)./p>female contrast. The right panels of a correspond to the female > male contrast. For a and d the pipelines are coded as: red = CAT, green = FSLVBM, blue = FSLANAT, light blue = sMRIPrep, additional colors visualize the overlap between the results, e.g., CAT ∩ FSLVBM = yellow, CAT ∩ FSLANAT = purple, FSLVBM ∩ FSLANAT = light blue, CAT ∩ FSLVBM ∩ FSLANAT = white and etc. b The variability of unthresholded statistical maps. The correlation values between whole-brain unthresholded statistical maps of four pipelines were computed respectively for sex differences. Only positive values are visualized for display purpose. c Decoding the functional properties of the identified brain regions of male > female (a, red = CAT, green = FSLVBM, blue = FSLANAT, no difference in sMRIPrep) using NeuroSynth. Only the top 20 functional terms are visualized. The font size reflects the size of the correlation. d Reliable brain patterns to distinguish sex differences via bootstrapping test (5000 permutations, pFDR < 0.05), and e cross-predicted accuracy of four pipelines in independent samples. The color from cold to warm indicates increasing classification accuracy (from 0.5 to 1)./p> 5, Fig. 3d), and there were wide differences in the location of predictive voxels. For instance, predictions based on CAT strongly weighted voxels in the putamen, hippocampus, middle cingulate cortex, and angular gyrus, while FSLANAT identified strongly predictive voxels in a widespread network including the superior frontal cortex, orbitofrontal cortex, pre- and postcentral cortex, insula, temporal pole, angular gyrus, and cerebellum. FSLVBM and sMRIPrep revealed generally similar findings to FSLANAT./p>86%), as well as relatively high accuracy for predicting data processed by sMRIPrep (FSLANAT: 80%, Cohen's d = 0.6437, and FSLVBM: 76%, Cohen's d = 0.7231) and CAT (FSLANAT: 72%, Cohen's d = 0.7810, and FSLVBM: 76%, Cohen's d = 0.7402). For further independent validation of the sex-predictive pattern in dataset 2 see Supplemental Results./p>5, Fig. 4c). Considerable spatial variations became apparent (Table S11), for instance, FSLANAT revealed high predictive weight for regions in the putamen, hippocampus, hypothalamus, brainstem, medial frontal cortex, middle temporal gyrus, middle frontal gyrus, and insula, while data processed with FSLVBM suggested that postcentral gyrus, superior frontal gyrus, superior temporal gyrus, and cerebellum strongly contributed to the prediction. Despite marked differences in the spatial distribution, an accurate prediction of age was possible based on data from all pipelines, as reflected by high correlations between the predicted and true age (all r values >0.8, Fig. 4d)./p>0.8, Fig. 4d, for sex classifiers, higher than chance level, Figs. 3e and S10). These results suggest that all pipelines retained biologically and functionally relevant information. However, further examination of the spatial distribution of the most predictive voxels revealed considerable variations across the four pipelines, similar to the mass-univariate analyses (Figs. 3d and 4c, Table S11). For instance, the application of CAT processed data to develop sex classifiers would have emphasized the region-specific contribution of the putamen, hippocampus, middle cingulate cortex, and angular gyrus, while FSLANAT would have indicated that a widely distributed pattern allowed successful sex classification. Finally, the preprocessing pipeline had a significant effect on prediction accuracy and prediction effect sizes, such that, depending on the pipeline, our sex classifiers reached 70–94% classification accuracy in an independent dataset. This indicates that the processing pipeline can have a considerable effect on the sensitivity and specificity of multivariate predictive signatures./p>70 indicates satisfactory to excellent image quality). Thus, all data passed the quality control procedure./p>0.74 excellent64,65,66./p>