Multi-Objective GA-Optimized Interpolation Kernels for Reconstruction of High Resolution EMR Images from Low-sampled k-Space Data

The low-frequency instrumentation and imaging capabilities facilitate electron magnetic resonance imaging (EMRI) as an emerging non-invasive imaging technology for mapping free radicals in biological systems. Unlike MRI, EMRI is implemented as a pure phase–phase encoding technique. The fast bio-clearance of the imaging agent and the requirement to reduce radio frequency power deposition dictate collection of reduced k-space samples, compromising the quality and resolution of the EMR images. The present work evaluates various interpolation kernels to generate larger k-space samples for image reconstruction, from the acquired reduced k-space samples. Using k-space EMR data sets, acquired for phantom as well as live mice, the proposed technique is critically evaluated by computing quality metrics viz. signal-to-noise ratio (SNR), standard deviation error (SDE), root mean square error (RMSE), peak signal-to-noise ratio (PSNR), contrast-to-noise ratio (CNR) and Lui's error function (F(I)). The quantitative evaluation of 24 different interpolation functions (including piecewise polynomial functions and many windowed sinc functions) to upsample the k-space data for the Fourier EMR image reconstruction shows that at the expense of a slight increase in computing time, the reconstructed images from upsampled data, produced using Spline-sinc, Welch-sinc, and Gaussian-sinc kernels, are closer to reference image with minimal distortion. Support of the interpolating kernel is a characteristic parameter deciding the quality of the reconstructed image and the time complexity. In this paper, a method to optimize the kernel support using genetic algorithm (GA) is also explored. Maximization of the fitness function has two conflicting objectives and it is approached as a multi-objective optimization problem.