Denoising MRSI Data Using Atlas-Based Statistical Subspaces

Motivation: The ensemble statistics or population information of metabolic signals has not been effectively utilized for MRSI signal processing due to lack of high-quality training data. Goal(s): To leverage the large high-quality MRSI datasets we have created for denoising MRSI signals. Approach: A position-dependent statistical subspaces model was used to create a spectroscopic brain atlas that captured the population statistics of a large MRSI dataset. The statistical atlas was then used to denoise new MRSI data by solving a maximum-a-posterior estimation problem. Results: The method was validated using both simulated and in vivo data acquired from healthy subjects, demonstrating excellent denoising performance. Impact: This proposed method significantly improved the sensitivity of brain MRSI using atlas-based statistical subspaces. The method may further enhance the reliability and practical utility of high-resolution MRSI techniques.