Microfluidics-based cell recognition through optimizing suspended cell staining techniques and artificial intelligence.

Abstract

PURPOSE: Blood staining is necessary to differentiate between its various cells. It is typically performed on a microscopic slide that contains a blood film. However, in some applications, there is a need to stain blood in suspension. METHODS: Three techniques were optimized (Romanowsky stains, Hematoxylin stain, and our proposed modified Romanowsky stain) for cell staining before injection into a microfluidic device. Images of stained cells in the microfluidic device were captured under an optical microscope. The effectiveness of staining techniques in cell differentiation was evaluated using color-based k-means clustering. Then, four pre-trained models (Alexnet, VGG16, Resnet, and Densnet) were applied to extract features and classify RBCs, WBCs, and CTCs. RESULTS: Only images showing clear cell borders were retained. The models achieved high testing accuracy, reaching 100%, 98%, 98%, and 95.34%, respectively. CONCLUSIONS: This high accuracy was achieved without the use of fluorescence labeling or expensive materials, thereby preserving the cells alive. Based on these results, the proposed staining technique was the most effective in differentiating between WBCs, RBCs, and CTCs, where all WBCs and RBCs were classified correctly.