|
Flow cytometry is the standard tool for blood analysis which generates information gathered from the interaction of lasers with cell flowing in a stream to classify them based on their size, granularity and fluorescent emission from biomarkers used as labels. However, for many emerging applications the use of labels is undesirable because they alter the cell behavior through activation or inhibition of cellular functions and hinder downstream genetic studies.
We have previously described an ultrahigh-throughput label-free imaging flow cytometer that analyzes cells using their biophysical features. Label-free ultrafast imaging in flow is implemented by photonic time stretch and the trade-off between sensitivity and speed is mitigated by using amplified time-stretch dispersive Fourier transform, a technique that was originally developed to enable realtime analog-to-digital data conversion with femtoseconds sampling resolution. In the application to time stretch to imaging, cells are illuminated by spatially dispersed broadband pulses, and the spatial features of the target are encoded into the short pulse spectrum. Both phase and intensity images are simultaneously captured, and this provides ample features such as the concentration of proteins, optical loss, and cell morphology, which are then used by a neural network to classify cells. However, image processing needed to extract these features from label-free images takes time and renders this technique unsuitable for cell sorting where decisions must be made in realtime before cells exist the fluidic stream.
To eliminate this predicament we have a deep convolutional neural network that directly processes the raw time stretch waveforms from the imaging flow cytometer. Eliminating the requirement of an image processing pipeline prior to the classifier, the running time of cell analysis can be reduced significantly, and cell sorting decisions can be made in less than a millisecond, orders of magnitude faster than the previous state of the art.
|
