Fig. 2: Quantitative results of segmentation performance.

a, b Box plots of Dice scores per organ; each box extends from lower to upper quartile values of data, with a black line at the median; the whiskers extend to the outmost data point within 1.5× the interquartile range; the outliers beyond the whiskers are shown as diamonds; the blue lines represent human segmentation performance. ^$The human annotators segmented bones with a semi-automatic thresholding function, which may overestimate human consistency as compared to purely manual segmentation. a Each box plot represents n = 140 independent scans from 20 biologically independent animals. b Each box plot represents n = 81 independent scans from eight biologically independent animals. c Processing time (in log scale) to segment one 3D whole-body scan of a mouse if done manually, with current state-of-the-art methods (atlas registration^36,37) or with machine learning approaches (here: AIMOS). d Median dice scores of AIMOS compared with those of the state-of-the-art from literature for native and contrast-enhanced microCT (scores are not based on the same dataset). Studies marked with asterisks exploited multimodal inputs of native microCT and MRI; as MRI provides high contrasts for soft abdominal organs, these studies are listed for both tables. Source data are provided as a Source Data file.