Fig. 7: Expert and external validation of computational cytomorphology.
From: Computational analysis of peripheral blood smears detects disease-associated cytomorphologies
a Correspondence between WCMs and expert annotated WBC types (left) and number of annotated WBC types. b Correspondence between RCMs and expert-annotated RBC types (left) and number of annotated RBC types. c External validation for glmnet. d External validation performance for the Morphotype analysis with all morphotypes (small white diamonds) and using only stable morphotypes (green diamonds). Enrichment values in a and b marked with an asterisk “*” are significant for a chi-squared test. Consensus morphotypes are highlighted in bold and labeled according to Fig. 6 and preceded by “CM”, whereas uncertain CM are labeled arbitrarily with letters. The whiskers in c, d represent the range described by \(\left[{maximum}\left({AUC}-{se},0\right),{minimum}\left({AUC}+{se},1\right)\right]\), where \({se}\) is the standard error calculated as \(\frac{1}{\sqrt{n}}\), where \(n\) is the number of samples used for this estimate (n = 63, 52, 30 and 22 independent PBS for the disease detection disease classification, MDS genetic subtyping and anemia classification tasks, respectively).
