Detection of circulating tumor cells with image cytometry is limited by the sensitivity and specificity of the biomarker panel. We collected confocal images of ~100,000 cells labeled for DNA, lipids, CD45, and Cytokeratin on a model system of MCF7 and WBCs representing disease positive, D+ and disease negative, D- populations. We computed spatial image metrics and performed multivariable regression and feature selection, increasing the separation of the D+ and D- populations to 7 standard deviations with detection limit of ~1 in 480. In conclusion, simple regression analysis holds promise to improve the separation of rare cells in cytometry applications.
|