In this paper, we focus on the reconstruction of iDQC MR spectroscopy data. Unlike standard 1-D MR spectroscopy schemes, iDQC acquires 2-D spectral data to decouple the magnetic field inhomogeneity effects from chemical shift information. This method suffers from a few limitations such as low intrinsic SNR, long acquisition time and the presence of a strong residual water peak. The standard IFFT based reconstruction schemes that rely on uniform sampling patterns have limited ability to overcome these problems. We propose a novel reconstruction scheme that employs a non-uniform sampling pattern, based on a flexible partitioning of the 2-D data-space. This approach enables the reconstruction of the metabolite region with a good resolution and signal to noise ratio, without being affected by aliases from the residual water signals. We pose the reconstruction problem as a conjugate gradient optimization. Using phantom data, we illustrate the significantly improved performance of the new method.