Quantification of microfluidic dye mixing using front line tracking in curvature scale space

Abstract

Microfluidic mixing or mixing at low Reynolds number is dominated by viscous forces that prevent turbulent flow. It therefore differs from conventional mixing (e.g., stirring milk into coffee), as it is driven primarily by diffusion. Diffusion is in turn dependent on (i) the concentration gradient along the interface between two fluids (dye front line) and (ii) the extent of the interface itself. Previously, we proposed an in vivo method to microscopically monitor the mixing interface using Shannon information entropy as mixing indicator and explored the use of length of dye front line as an indirect measure of mixing efficiency. In this work, we present a robust image processing chain supporting quantitative measurements. Based on data from ciliated surfaces mixing dye and water, the dye-water interface front line is extracted automatically using the following processing steps: (i) noise reduction (average filtering) and down sampling in time to reduce compression artifacts; (ii) subtraction imaging with key reference frames in RGB color space to remove background; (iii) segmentation of dye based on color saturation in HSV color space; (iv) extraction of front line; (v) curve smoothing in curvature scale space (CSS) with an improved Gaussian filter adaptive to the local concentration gradient; and (vi) extraction of length. Evaluation is based on repeated measurements. Reproducibility in unaltered animals is shown using intra- and inter-animal comparison. Future work will include a more comprehensive evaluation and the application to datasets with multiple classes.