Live cell imaging and classification via microscopic ghost imaging

This study focuses on the application of microscopic ghost imaging (MGI) in live cell imaging and classification. High-resolution imaging is performed on human red blood cells, 293T cells, and three types of cancer cell line (Caov3, Molm13, and Ishikawa) with the use of MGI. Subsequently, cell classification is conducted with the random forest classifier based on image features and bucket detection data, respectively. The imaging results reveal the characteristic biconcave disk structure of red blood cells. Moreover, significant morphological and aggregation differences are observed between the similarly sized 293T cells and the three cancer cell lines. In terms of classification, the maximum accuracy achieved on the basis of image features is 73%. Remarkably, classification using bucket data achieves an accuracy of 88% even at an extremely low sampling rate (0.03%), with accuracy stabilizing at around 94% as the sampling rate increases to 0.12%. Principal component analysis of the bucket data further demonstrates significant feature differences among the cell lines. This study not only demonstrates the advantages of MGI in live cell imaging but also uncovers the potential of bucket data for efficient cell classification. These findings provide valuable insights and methods for applications in biomedical imaging, drug screening, and clinical diagnostics.