(157bn) Deep Learning-Aided Intelligent Flow Cytometry for Label-Free Cell Detection

Recent efforts are being directed to developing various label-free methods for cell detection and analysis. Besides offering simplicity and low-cost operation, a label-free method avoids any adverse effects of the staining reagents on cellular viability and functions. Nevertheless, the scope of the label-free techniques developed to date remains limited. Moreover, most of these techniques lack adequate throughput or precision. This work introduces a deep learning-leveraged label-free approach that could potentially enable cell detection and sorting at precision and throughput comparable to conventional flow cytometry. Unlike other existing label-free methods, this approach detects a cell by its autofluorescence measured using a multicolor flow cytometer. Analyses provided here indicate that autofluorescence may obscure sufficient information to discriminate different cell types. Supervised deep learning could be employed to analyze features hidden in the single-cell autofluorescence and infer cell types by mapping those features. This approach is demonstrated by classifying heterogeneous mixtures of human endothelial and breast carcinoma cells. The integration of the approach with conventional flow cytometry may enable intelligent systems for data-driven cell detection based on single-cell autofluorescence.