Evaluation of machine learning-driven automated Kleihauer-Betke counting: A method comparison study

The Kleihauer-Betke (KB) test is the diagnostic standard for the quantification of fetomaternal hemorrhage (FMH). Manual analysis of KB slides suffers from inter-observer and inter-laboratory variability and low efficiency. Flow cytometry provides accurate quantification of FMH with high efficiency but is not available in all hospitals or at all times. We have developed an automated KB counting system that uses machine learning to identify and distinguish fetal and maternal red blood cells (RBCs). In this study, we aimed to evaluate and compare the accuracy, precision, and efficiency of the automated KB counting system with manual KB counting and flow cytometry.

The ratio of fetal RBCs of the same blood sample was quantified by manual KB counting, automated KB counting, and flow cytometry, respectively. Forty patients were enrolled in this comparison study.

Comparing the automated KB counting system with flow cytometry, the mean bias in measuring the ratio of fetal RBCs was 0.0048%, with limits of agreement ranging from -0.22% to 0.23%. Using flow cytometry results as a benchmark, results of automated KB counting were more accurate than those from manual counting, with a lower mean bias and narrower limits of agreement. The precision of automated KB counting was higher than that of manual KB counting (intraclass correlation coefficient 0.996 vs 0.79). The efficiency of automated KB counting was 200 times that of manual counting by the certified technologists.

Automated KB counting provides accurate and precise FMH quantification results with high efficiency.