Compressive sampling with prior information in remotely detected MRI of microfluidic devices

• Published in: Journal of magnetic resonance (1997) 2012-03, Vol.216, p.13-20
• Main Authors: Teisseyre, Thomas Z., Paulsen, Jeffrey L., Bajaj, Vikram S., Halpern-Manners, Nicholas W., Pines, Alexander
• Format: Article
• Language: English
• Subjects: Chips; Compressed sensing; Construction; Detectors; Devices; Microfluidics; MRI; Reconstruction; Remote detection; Sampling
• ISSN: 1090-7807; 1096-0856
• DOI: 10.1016/j.jmr.2011.10.001

[Display omitted] ► We incorporate known geometrical information into CS MRI image reconstruction. ► We demonstrate multidimensional velocimetric imaging of microfluidic chips. ► Reconstruction errors are quantified with synthetic data reconstruction. The design and operation of microfluidic analytical devices depends critically on tools to probe microscale chemistry and flow dynamics. Magnetic resonance imaging (MRI) seems ideally suited to this task, but its sensitivity is compromised because the fluid-containing channels in “lab on a chip” devices occupy only a small fraction of the enclosing detector’s volume; as a result, the few microfluidic applications of NMR have required custom-designed chips harboring many detectors at specific points of interest. To overcome this limitation, we have developed remotely detected microfluidic MRI, in which an MR image is stored in the phase and intensity of each analyte’s NMR signal and sensitively detected by a single, volume-matched detector at the device outflow, and combined it with compressed sensing for rapid image acquisition. Here, we build upon our previous work and introduce a method that incorporates our prior knowledge of the microfluidic device geometry to further decrease acquisition times. We demonstrate its use in multidimensional velocimetric imaging of a microfluidic mixer, acquiring microscopically detailed images 128 times faster than is possible with conventional sampling. This prior information also informs our choice of sampling schedule, resulting in a scheme that is optimized for a specific flow geometry. Finally, we test our approach in synthetic data and explore potential reconstruction errors as a function of optimization and reconstruction parameters.