Automation and improvement of WBC mechanical profiling in deformability cytometry

Deformability cytometry (DC) is a powerful biophysical technique that enables cost-effective, high-throughput characterization of disease-associated changes in blood cell mechanics. Mechanical profiling of living white blood cells (WBCs) is particularly valuable due to their critical role in the immune response. However, reliably identifying and classifying WBC subtypes in a label-free manner remains a significant challenge. Until now, the analysis pipeline has relied on manual gating by trained experts, limiting scalability and reproducibility. In this study, we present a fully automated and generalizable framework for WBC classification in shear flow DC experiments, based on box filters and unsupervised clustering of cell populations. Both box filters and unsupervised clustering rely on cell shape features derived from high-accuracy segmentation and on cell texture features derived from bright-field images. This unsupervised approach not only improves reproducibility and reduces processing time but also overcomes key limitations of supervised models that require extensive training data and often suffer from reduced performance under varying imaging conditions. We validated our method by comparing cell features obtained through manual gating and automated classification across six experimental sets. These sets incorporated variations in blood donors, anticoagulants (EDTA and citrate), blood collection sources (capillary and venous), and device brightness settings. Each set included five repeated measurements. The results consistently confirmed the reliability and robustness of the method across all tested conditions and WBC types. Importantly, this automated pipeline enables the inclusion of WBCs with membrane protrusions-typically excluded from standard analyses-allowing for morphological characterization of potentially activated cells. Moreover, by using shape features derived from the original contour rather than the convex hull, we improve morphological accuracy and reduce measurement variability. This approach thus enhances the accuracy, consistency, and scalability of WBC mechanophenotyping and enables high-throughput analysis across large cohorts.