Flow cytometry is the gold-standard technique for enumerating and sorting cellular populations. Although modern-day flow cytometers are adept at processing thousands of cells per second, almost all require excessively large sample/reagent volumes and do not provide spatially resolved information. To address these issues, imaging flow cytometry (IFC) marries the advantages of optical microscopy and flow cytometry to allow for high-throughput imaging of cells within flowing environments. Such developments combined with recent advances in machine learning, suggest significant utility in disease diagnostics, where large numbers of cells within bodily fluids must be investigated at high speed.
To this end, Otesteanu and colleagues recently used ICellCnn, a weakly supervised deep learning approach, in conjunction with label-free IFC for Sézary syndrome blood diagnostics. Sézary syndrome is an aggressive form of blood cancer characterized by circulating tumor T cells that possess cerebriform nuclei and altered cell morphology. Using a simple microfluidic platform, the authors assayed thousands of cells from 4 healthy donors and 5 Sézary Syndrome patients and successfully identified diseased cells. Interestingly, when using a strong supervision approach with naïve labels, diseased cells were corrected detected in Sézary syndrome patients, but also incorrectly detected “diseased” cells in healthy donors. On the other hand, when using strong supervision with expert labels, healthy patients were correctly diagnosed, but some Sézary syndrome patients were misdiagnosed.
A unique feature of ICellCnn is its representation of each cell image as a feature vector (computed by the encoder block of a convolutional autoencoder). Multiple cell image representations from the same patient specimen were concatenated as a “bag of cells” (BoC) in a two-dimensional feature vector that was then used to train a random forest classifier. Using such an approach, all healthy and Sézary syndrome patients were correctly identified with clear differences in the amounts of Sézary syndrome cells between healthy and Sézary syndrome patients. Since this IFC-based method is able to recognize variations in cell morphology, it can also be applied to the diagnosis of any disease that is accompanied by morphological aberrations in blood cells.
Written by Sarah Duclos Ivetich
Link for the full article here