Chemical cytometry image analysis has been performed through various computational methods to classify numerous cells or to figure out the physical characteristics (e.g. size, and shape) of cells. Recently, new methods in chemical cytometry to quantify NO efflux of macrophages by incorporating fluorescent NO nanosensor arrays and microfluidic collagen channels, called nanosensor-collagen-integrated microfluidic channels (NCIM), were reported. In particular, near-infrared (nIR) images obtained through NCIM can show chemical properties in a photonic manner, such as NO efflux rate, as well as biophysical characteristics. However, it is difficult to recognize whether a cell is activated in a single cell unit when cells have diverse shapes and sizes and exist attached to each other. Herein, we distinguished activated states and nonactivated states of macrophages in nIR images by integrating deep learning, which was based on object detection in images. A total of 91 nIR images were prepared for model training: 28 nIR images with lipopolysaccharide (LPS+), which stimulates macrophages to be activated, and 63 nIR images without lipopolysaccharide (LPS-). These images were cropped to 5,096 images for the effect of data augmentation and clear cell labeling, of which 280 were used for model training and the remaining 4,816 were used for detection. The highest performance of mean average precision(mAP) for all classes was 0.84, averaging 0.94 for nonactivated and 0.74 for activated cells. The cause of the low performance of activated cells was considered that the activated cell boundaries were more ambiguous because activated cells were adhered and spread. Next, the activation state of macrophages was analyzed through the detection process for the remaining nIR images to verify the model. The activated ratio of each nIR image was higher with LPS activation (+LPS) than without LPS activation (-LPS), which is consistent with previous results. Furthermore, the activation ratio of LPS+/LPS- increased as the detection confidence threshold increased. In addition, we investigated that activated cells and nonactivated cells have an average intensity of 0.36 (scaled) and 0.40 (scaled) and an average size of 235 μm<sup>2</sup> and 103 μm<sup>2</sup> in a single cell unit analysis. Our approach provides a simple method to indirectly profile the activation status of the same cell within nIR images of cytometry.

Immune cell activation profiling with nanosensor integrated deep learning

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.