Non-alcoholic fatty liver disease (NAFLD) is a common consequence of metabolic disorders such as obesity and diabetes, and can lead to more serious complications like fibrosis, cirrhosis and hepatocellular carcinoma. The focus of this project is to 1) develop robust and efficient magnetic resonance image processing technology to study NAFLD, and 2) establish its utility in studies of hepatic fat accumulation by applying it in animal models of fasting.
First, we developed image analysis tools for small animal applications. Several imaging techniques were applied to extend the computational efficiency, robustness, and throughput of Dixon imaging. We enforced local smoothness on the field inhomogeneity map by adding a Markov Random Field (MRF) smoothness constraint to the three-point Dixon formulation, and were able to generate fat fraction maps that were generally robust to the field inhomogeneity seen at 7T. We solved this formulation efficiently by adding three implementation features to Iterated Conditional Modes (ICM): stability tracking, multiresolution image pyramids, and image masking.
Second, we made the estimation more robust by enforcing global smoothness of the field map. We used extrapolation-initialization to leverage 3-D smoothness, and combined this feature with error correction strategies, using hole-filling and k-means based error detection. The combination prevented all complete swaps, and almost all partial swaps in the absence of motion artifact.
Third, we attempted to increase the throughput of our methods by combining them with Compressed Sensing (CS) rapid data acquisition. A single-step formulation was developed that combined CS with the MRF chemical decomposition, but this formulation proved computationally expensive to solve. Using a two-step serial methed with CS followed by fat/water decomposition, we optimized the source image, and found that the decomposition quality can be improved by distributing the sampling time unequally amongst the source images.
Finally, we applied our imaging technology to monitor dynamic changes in hepatic fat during fasting and refeeding in mice, which elicit rapid changes in hepatic fat. We were able to measure these changes with MR imaging. With this project, we showed that our technology can reliably characterize and monitor hepatic fat for the study of dynamic fat accretion patterns.