Extended Data Fig. 3: Comparable performance in brain tissue segmentation of DeepPrep and fMRIPrep.
DeepPrep’s performance in brain tissue segmentation was assessed and compared with that of fMRIPrep. To demonstrate segmentation accuracy, both DeepPrep and fMRIPrep were applied to the Mindboggle-101 dataset. This dataset includes anatomical images and manually segmented brain regions by experts as the ‘ground truth’ of brain tissue segmentation. The accuracy of automatic segmentations produced by both pipelines was evaluated by calculating Dice coefficients, which measure the similarities of brain regions between the automatic and manual segmentations. Comparison of segmentation accuracy between DeepPrep and fMRIPrep was conducted by evaluating the percentage difference in Dice coefficients for each brain region. Positive percentages indicate higher accuracy achieved by DeepPrep, while negative percentages indicate higher accuracy for fMRIPrep. a) Fifty-two out of 62 cortical regions display positive percentages. b) Similar comparisons were performed for 39 subcortical regions, where 31 regions show positive percentages. All statistics are reported in Supplementary Table 4 and Table 5.
