Professor Kevin Tsia

Decoding cell with microscopy - from petabyte-scale data generation to intelligent morphological profiling

Abstract

Pairing up optical microscopy and computer vision becomes a common strategy adopted in a broad spectrum of biological and biomedical screening applications. The common rationale is to generate the characteristic "fingerprint" profiles of cell morphology that could underpin the cell states/functions but obscured through visual inspection or even in the molecular assay. In this talk, I will introduce how the synergism between ultrafast imaging, microfluidics, and deep learning allows us to overcome some of these current limitations. Specifically, I will present a few high-throughput, deep-learning-powered imaging techniques and imaging cytometry pipelines developed in our laboratory over the past few years. These platforms allow us to significantly scale up the single-cell phenotyping throughput (beyond millions of cells per run within ten minutes) - approaching to the petacyte-scale imaging capability; to enrich the phenotyping content by integrating with the biochemical cell-based assay in a single platform. These techniques have achieved biophysical/mechanical phenotyping specificity and sensitivity that were once inconceivable. Combined with self-supervised/unsupervised deep learning, these methods are now successfully employed in many biological research and clinical applications, including delineating immune-cell sub-types and their activations, predicting targeted-drug sensitivity, and more emerging applications, including genetic screens with image‐based profiling.