Deep learning segmentation of the choroid plexus from structural magnetic resonance imaging (MRI): validation and normative ranges across the adult lifespan

Table 1 Performance metrics for each machine learning method and FreeSurfer using manual segmentations as the ground truth

Method	Sørensen–Dice Coefficient	95% Hausdorff Distance (mm)	AUC
Deep Learning from T₁-weighted MRI	0.72 (0.55–0.78)***	1.97 (1.00–6.71)***	0.87 (0.75–0.96)***
Deep Learning from T₂-weighted MRI	0.72 (0.57–0.78)***	2.22 (1.00–16.6)***	0.87 (0.75–0.96)***
Deep Learning from T₂-weighted FLAIR MRI	0.74 (0.61–0.80)***	1.69 (1.00–3.74)***	0.87 (0.74–0.96)***
FreeSurfer from T₁-weighted MRI	0.19 (0.02–0.37)	10.4 (4.12–17.2)	0.56 (0.50–0.62)

Values are shown as mean (range). Metrics for the machine learning-based methods were calculated from ten testing participants across five cross-validation iterations, whereas metrics for FreeSurfer were calculated from all 50 participants included in the algorithm development. *** indicates two-tailed Wilcoxon test revealed a significant difference between the machine learning method and FreeSurfer (p-value < 0.001)

ISSN: 2045-8118