100 Outliers for Self-explaining Artificial Intelligence for the Classification of B cell Non-Hodgkin Lymphoma

A set of 100 randomly drawn atypical cases was reviewed by immunophenotyping experts. By manual analysis in bivariate plots, the selected data cases were indeed judged as altered or hardly evaluable by the human experts. This manual investigation of the cases confirmed that only four percent of the atypical cases were considered usable for a trustworthy manual diagnostic workflow across all tubes. All other cases (96%) were atypical because of alterations such as incomplete erythrocyte lysis, irregular antibody staining patterns or other aberrations, most likely acquisition artifacts. The manual validation confirms that TM reliably distinguishes inlier cases from atypical ones.