|dc.description.abstract||The field of microfluidics has been solving problems on the micro-scale for
decades, but many in-flow analysis techniques only take single dimensional
measurements. In this thesis, multi-dimensional, real-time image analysis has been used
to improve and expand upon current microfluidic techniques in several microfluidic areas.
Microdroplets within microfluidics are a promising technique for creating microscopic
vessels for chemical and biochemical experiments, however accurately controlling such
tiny objects can be difficult. The use of real-time image feedback has dramatically
improved the monodispersity (coefficient of variation of 0.32%) and accurate loading of
the contents of droplets. Beyond this, using these techniques, real-time analysis on the
morphology of living cells can be carried out and used to isolate cells of interest. Machine
learning algorithms have provided a rapid method to identify the cell populations based
on quantitative parameters extracted from transmission or fluorescent images of the cells.
By integrating fast piezo-based fluid manipulation, highly selective and accurate cell
sorting can be performed within a lab-on-a-chip device for the isolation of subpopulations
of cells based on their morphological features. Using this method, K562 cells have been
sorted from a mixed population with an efficiency of 91.3% and a purity of 99.4%.||en