Skip to main content
. Author manuscript; available in PMC: 2021 May 25.
Published in final edited form as: Proc SPIE Int Soc Opt Eng. 2020 Mar 10;11313:113130L.

Figure 3. Quantitative results of augmented transfer learning.

Figure 3.

A) The best performance DSC (pDSC) on the pediatric dataset was achieved when using the pediatric cohort exclusively during transfer learning showing an increase of 0.08 in pDSC compared to the original SLANT. Statistical significance (*) is achieved at p < 0.0167 (Wilcoxon Sign-Rank Test with Bonferroni corrections). B) The best reproducibility DSC (rDSC) is achieved when using a mixed cohort of contrast-enhanced images and manually traced non-contrast images, for a mean increase in rDSC of 0.05 compared to original SLANT. Statistical significance (*) is achieved at p < 0.0167 (Wilcoxon Sign-Rank Test with Bonferroni corrections). C) There is a decrease in performance of the manual labels in the adult dataset when new data are introduced, although this decrease is mitigated when using a mixed training cohort for both pediatric SLANT and contrast SLANT. Statistical significance is achieved at p < 0.083 (Wilcoxon Sign-Rank Test with Bonferroni corrections).