Skip to main content
. Author manuscript; available in PMC: 2023 Sep 1.
Published in final edited form as: J Neuroimaging. 2022 Apr 17;32(5):968–976. doi: 10.1111/jon.12997

Table 3.

Performance metrics for DNN segmentation of ICH and IVH versus semi-automated planimetry.

Clinical Trial Subtype Dice Volume Correlation (95% CI) RMSE (ml) Abs. Vol Diff. (ml) < 5ml
MISTIE II ICH
(n=135)
0.946 0.994
(0.99, 0.996)
2.15 0.566
(0.264, 1.43)
96.3%
IVH
(n=48)
0.776 0.980
(0.96, 0.99)
2.02 1.08
(0.463, 1.698)
96.1%
ICH + IVH
(n=135)
0.940 0.995
(0.99, 0.996)
2.22 0.655
(0.339, 1.69)
95.6%
CLEAR IVH ICH
(n=40)
0.881 0.931
(0.87, 0.96)
2.95 0.857
(0.388, 1.94)
92.6%
IVH
(n=51)
0.863 0.985
(0.97, 0.99)
6.67 2.91
(1.52, 6.27)
72.0%
ICH + IVH
(n=51)
0.905 0.995
(0.99, 0.997)
4.93 2.74
(1.19,5.80)
70.0%

Total subjects (n) are listed for each study, then Dice score and volume correlation with 95% confidence interval (95% CI). RMSE is the root mean squared error in ml. Abs. Vol. Diff. is the absolute volume difference in ml by median and interquartile range.. %< 5ml is the percentage of predictions that were within 5ml.