Open-Set Based Single Cell Identification in Microfluidics

Neural networks are commonly used for image classification in life sciences. However, classifying images of unknown cells poses challenges as existing knowledge cannot guide classification, leading to misclassification. To address this issue, open-set recognition was introduced but has not been extensively tested in single-cell applications. In this work we applied open-set recognition to scattering snapshots of living cells to distinguish known from unknown cells. We examined the impact of neural network parameters to improve unknown cell detection accuracy. Our open-set neural network approach highlights measurement uncertainty in cell prediction, offering potential for diverse single-cell classifications.