Label-free viability detection of T-cells based on 2D bright-field microscopic images and deep learning

Abstract

The measurement of cell viability is critical in the biomedical field. It is currently accomplished by staining cells with various stains and then manually or with instruments such as counters counting dead or live cells. However, the cell staining step is relatively time-consuming, and the stain is toxic. The internal structure of the cells is destroyed after staining, resulting in valuable cells that cannot be reused later. We proposed a label-free cell detection algorithm based on 2D bright-field images of T-cells and deep learning in this work. When used, this method eliminates the need for staining operations on cells, and cell viability is determined directly from the detection of bright-field cell images. The method based on YOLOX deep learning analysis has an excellent detection performance on bright-field images of T-cells, and the framework achieves the mAP (mean average precision) of more than 96.31% after cell detection. Experimental results show that combining 2D cell bright-field images with deep neural networks can yield a new label-free method for cell analysis.