Improving SNR in deuterium metabolic imaging of cancer: spatial denoising, joint spectral/metabolic-kinetics processing, and beyond

As DMI is a promising cancer screening approach but is challenged by low sensitivity, this study assesses multiple approaches to increase its SNR. Some of these –apodization, Compressed Sensing Multiplicative (CoSeM), Block-matching/3D filtering (BM3D)– involve image denoising. Others take into account the metabolic kinetics, and include it as dimension to be denoised. This can be achieved by smoothing the kinetic data axis via regularization, or by using subspace-constrained representations to concurrently solve for the 4D spatial/spectral/kinetics set. These methods can be further denoised –e.g., by CoSeM– leading to much clearer observations of lactate generated by in vivo tumor models.