Human Error and Quality Assurance in Diatom Analysis

Quantitative comparison of taxa and taxon concepts in the diatom genus Fragilariopsis: a case study on using slide scanning, multiexpert image annotation, and image analysis in taxonomy1

“…Full consensus in taxonomic labeling of individual diatom specimens (valve images) may not be given for all taxa, even among experts who in principle largely agree on their discriminative criteria. This is not unique to the present target group but has been observed in other studies as well, both with diatoms and other organisms (Culverhouse et al 2003, Kelly et al 2011, Schoening et al 2016. The taxa targeted in this study, and some of the samples analyzed, were FIG.…”

“…However, a potentially preferable alternative, acknowledging that even the best expert might be wrong occasionally (and that this could be recognized by her/his deviation from the majority of other experts), would be to simply say that the gold standard is defined by how the majority of experts identified a specimen (Kelly et al. , Schoening et al. ).…”

“…Based on this, one could designate the identifications by expert E3 to be the gold standard for identifying the three species. However, a potentially preferable alternative, acknowledging that even the best expert might be wrong occasionally (and that this could be recognized by her/his deviation from the majority of other experts), would be to simply say that the gold standard is defined by how the majority of experts identified a specimen (Kelly et al 2011, Schoening et al 2016. For the following analyses, we took this latter approach and grouped specimens into one of the three taxa based on majority votes.…”

“…This is, however, a situation that often needs to be dealt with, i.e., whenever independent information for ground-truthing taxonomic identifications (for example, from molecular markers), is not available, as is the case for most recent and all fossil taxa. The availability of multiple taxonomic opinions still enables the generation of useful reference image sets and corresponding training data for computational classification even in the face of, and acknowledging, taxonomic disagreement, as done here and in other studies (Culverhouse et al 2003, Kelly et al 2011. Updated differential diagnoses.…”

Identification automatique des diatomées de la Merja fouarate : Une alternative à la détermination et à la classification manuelle

“…La problématique de l'identification et de la classification des diatomées s'avère délicate, même pour les spécialistes. Le nombre d'espèces connues est très grand (11 000 environ) et l'identification à l'oeil nu est une source d'erreur (Kelly et al, 2002). Ces difficultés entraînent inévitablement un processus d'analyse long, fastidieux et peu sûr (Guo, 2004 ;Loke, 2002).…”