In this study, we used spatial light interference microscopy (SLIM), an ultrasensitive QPI method, and deep learning, to first generate a virtually-stained micrograph of a blood smear. This approach of combining label-free QPI data with deep learning to infer chemical specificity has been recently developed in our laboratory and is referred to as PICS [Nat. Comm., in press]. Next, we applied a computational semantic segmentation to identify and delineate the white blood cells. Lastly, we ran a classification model on the leukocytes to identify their type and condition.
PICS renders synthetically stained blood smears rapidly, at a reduced cost of sample preparation, and provides quantitative clinical information. We validated this approach by successfully creating computationally stained micrographs and classified the leukocytes into five cell classes, with 92% accuracy.
|