Clinical Validation of the MC-80 Digital Morphology Analyzer in Leukocyte Classification: A Large-Scale Comparison With Manual Microscopy.

INTRODUCTION

Although AI-powered digital morphology analyzers are widely used for leukocyte classification, their reliability in detecting clinically significant abnormal cells requires rigorous validation. We evaluated the Mindray MC-80 system by comparing its results with those of conventional hematology analyzers and manual microscopy to assess accuracy, sensitivity, and diagnostic limitations.

METHODS

We analyzed 1188 routine blood samples (881 normal, 307 abnormal). Preclassification results from the MC-80 were compared with hematology analyzer results for normal samples and with manual microscopy by senior morphologists for abnormal samples. Inter-method agreement was assessed by Passing-Bablok regression, and sensitivity and specificity were calculated using expert-reviewed findings as the reference standard.

RESULTS

The MC-80 preclassification demonstrated excellent correlation with reference assays, with r values greater than 0.9 for the majority of leukocyte subsets. Overall specificity reached over 99% for all cell populations. Excluding blasts and reactive lymphocytes, the sensitivity for all remaining cell types was higher than 90%. For clinically critical abnormal cells, the system attained 75% sensitivity for plasma cells (≥ 1%) and over 90% for the other five abnormal cell groups. Its specificity was above 90% for all abnormal cells except blasts, which had a specificity of 86.73%. Expert review further increased the correlation coefficients by 0.31% to 5.71%.

CONCLUSION

The MC-80 provides robust performance for routine leukocyte classification, with a screening capability for most abnormal cells comparable to that of manual microscopy. However, its reduced sensitivity for reactive lymphocytes, blasts, and plasma cells indicates that it cannot fully replace manual review. Targeted optimization is therefore warranted before full integration into diagnostic workflows.