Label-free 8-part leukocyte differential enabled by imaging flow cytometry of acoustic focusing and deep neural network of ResNet-18

Leukocyte differential analysis represents a critical first-line diagnostic tool in clinical examinations. However, conventional methodologies exhibit significant limitations: Automated slide stainers demonstrate low throughput, while standard hematology analyzers lack the capability to accurately classify abnormal leukocytes. To address this, an imaging flow cytometry system was developed based on acoustic focusing to capture high-quality, label-free images of individual leukocytes, including 5 subtypes of normal cells and 3 subtypes of abnormal cells. A convolutional neural network (CNN) was employed for feature extraction and cell-type classification, enabling an 8-part leukocyte differential. Based on the developed system, high-quality label-free images of 4515 normal leukocytes and 3289 abnormal leukemia cells were obtained. Using ResNet-18, classification accuracies for 5-part normal leukocytes, 3-part abnormal leukocytes, and 8-part normal/abnormal leukocytes were quantified as 98.5 %, 97.0 %, and 97.9 %, respectively. To summarize, this label-free imaging flow cytometry enables an 8-part leukocyte differential through a combination of acoustic focusing and a deep neural network.