Computational Flow Cytometry Accurately Identifies Sezary Cells Based on Simplified Aberrancy and Clonality Features

Abstract

Flow cytometric identification of circulating neoplastic cells (Sezary cells) in patients with mycosis fungoides (MF) and Sezary syndrome (SS) is essential for diagnosis, staging and prognosis. While recent advances have improved the performance of this laboratory assay, the complex immunophenotype of Sezary cells and overlap with reactive T cells demand a high level of analytic expertise. We utilized machine learning to simplify this analysis using only 2 pre-defined Sezary cell-gating plots. We studied 114 samples from 59 patients with SS/MF, and 66 samples from unique patients with inflammatory dermatoses. A single dimensionality reduction plot highlighted all T-cell receptor constant β chain-restricted (clonal) CD3+/CD4+ T-cells detected by expert analysis. On receiver operator curve analysis, an aberrancy scale feature computed by comparison with controls (area under the curve = 0.98) outperformed loss of CD2 (0.76), CD3 (0.83), CD7 (0.77) and CD26 (0.82) in discriminating Sezary cells from reactive CD4+ T cells. Our results closely mirrored those obtained by exhaustive expert analysis for event classification (positive percent agreement = 100%, negative percent agreement = 99%) and Sezary cell quantitation (regression slope = 1.003, R squared = 0.9996). We demonstrate the potential of machine learning to simplify the accurate identification of Sezary cells.