Chemical shift encoding based double bonds quantification in triglycerides using deep image prior

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Medicine
• Biochemistry

Background:

• Fatty acids are crucial biomarkers for metabolic disorders and inflammation.
• Quantifying double bonds in fatty acids is essential for understanding their role.
• Current methods for fatty acid analysis can be complex and time-consuming.

Purpose of the Study:

• To assess a deep learning approach, Deep Image Prior (DIP), for quantifying double bonds and methylene-interrupted double bonds in triglycerides.
• To evaluate the efficacy of DIP using chemical-shift encoded multi-echo gradient echo images without requiring network training.
• To validate the method's performance through phantom experiments and in-vivo scans.

Main Methods:

• Implemented a Deep Image Prior (DIP) model for image reconstruction and parameter refinement.
• Utilized a cost function based on signal constraints for iterative optimization on single image slices.
• Performed validation using water-oil phantoms and in-vivo subcutaneous fat imaging.

Main Results:

• Achieved high concordance between quantified values and reference standards, with a Pearson correlation coefficient of 0.96 in phantom experiments.
• Demonstrated reliable quantification even with low-fat signals in water-oil phantoms.
• Showcased consistent in-vivo quantification results comparable to established findings for subcutaneous fat.

Conclusions:

• Deep Image Prior (DIP) effectively quantifies double bonds and methylene-interrupted double bonds from MRI data.
• The method offers a promising, training-free approach for fatty acid analysis.
• Potential for future research and clinical applications in metabolic and inflammatory disease assessment.