Extracranial lipid contamination is a persistent challenge in proton magnetic resonance spectroscopic imaging (MRSI), particularly in short-acquisition-delay protocols where broad lipid resonances overlap metabolite and macromolecular signals. Retrospective lipid suppression approaches are commonly used in human MRSI workflows, but their effects have not been systematically examined in preclinical MRSI, where smaller anatomy and closer proximity of extracranial lipids make contamination more likely. This study assessed how retrospective lipid suppression and the choice of spectral fitting range influence spectral quality, spatial metabolite mapping, and quantification variability in preclinical proton MRSI acquired at 14.1 T in rat brains.
Retrospective lipid suppression was implemented using an orthogonal projection method. The approach was applied to two types of datasets: fully sampled MRSI and compressed sensing undersampled MRSI. For each sampling scheme, datasets representing two contamination conditions were used: one with minimal lipid contamination and one with pronounced lipid contamination. Spectral fitting and quantification were performed twice per dataset using two fitting ranges: a broad range spanning 4.1 to 0.2 ppm and a narrower range spanning 4.1 to 1.8 ppm. The experimental context was proton MRSI of rat brains at 14.1 T. The abstract did not report detailed acquisition parameters, exact preprocessing steps beyond the orthogonal projection, or numerical thresholds used to define minimal versus pronounced contamination.
When lipid contamination was minimal, the application of retrospective lipid suppression produced only modest changes in spectral appearance and spatial metabolite maps. Metabolite quantification remained consistent across conditions (pre- and post-suppression) and across the two spectral fitting ranges. In other words, datasets with low levels of extracranial lipid signal were robust to the orthogonal projection suppression procedure, yielding similar concentration estimates and spatial distributions regardless of suppression.
In datasets with pronounced lipid contamination, retrospective lipid suppression caused notable spectral alterations. These spectral changes propagated into differences in spatial metabolite maps and into estimated metabolite concentrations. The suppression altered signal components sufficiently that mapping and quantification were affected compared with the unsuppressed data. The abstract highlights that the impact was not uniform across metabolites: those with lower estimated concentrations were most susceptible to change after suppression. Exact magnitude of changes, numerical results, and statistical measures were not provided in the abstract.
The authors applied the same orthogonal projection suppression to compressed sensing datasets and observed similar effects to those in fully sampled data. Compressed sensing datasets with pronounced lipid contamination showed spectral and mapping changes after suppression, and low-concentration metabolites were again the most affected. This indicates that the interaction between retrospective lipid suppression and undersampled reconstruction can influence downstream metabolite quantification in preclinical MRSI.
Group analysis across datasets demonstrated a consistent pattern: metabolites with low concentration estimates were disproportionately affected by retrospective lipid suppression, especially in the presence of pronounced lipid contamination. Where contamination was minimal, group-level quantification remained comparable before and after suppression. The abstract does not enumerate which specific metabolites were most impacted nor provide effect sizes or variability metrics; those details were not reported in the source abstract.
These findings provide practical guidance for preclinical MRSI workflows: retrospective lipid suppression via orthogonal projection can be applied with minimal perturbation when extracranial lipid contamination is low, but it may alter spectral content, spatial maps, and concentration estimates when lipid contamination is pronounced. Special caution is warranted for metabolites present at low concentrations, which appear most vulnerable to changes due to suppression. The similar behavior observed in compressed sensing data suggests that suppression effects should be evaluated when combining retrospective lipid removal with undersampled acquisition and reconstruction strategies. The results support targeted optimization of suppression methods and fitting ranges to balance lipid removal with preservation of metabolite signals in preclinical MRSI.
The study authors declared no competing interests. Funding was reported from the Swiss National Science Foundation. The source abstract did not provide full methodological or quantitative details such as specific acquisition parameters, numerical results, or statistical analyses; those details were not reported in the abstract and would require consultation of the full preprint for complete appraisal.