The Local Motion of Food Reconstructed from MRI: Optimised k-means Clustering

The study of gastric motility is important to gain an understanding of disease and nutrition. Many studies focused on gastric motility require the use of synthetic tracers to reconstruct the motion of content. Our study instead exploits the properties of a specific magnetic resonance imaging (MRI) acquisition protocol, combined with multi-objective optimised clustering to map the motion of peas in a human stomach. We chose Non-dominated Sorting Genetic Algorithm II (NSGA-II) to optimise the starting positions for a modified k-means to create optimum clusters. We compared our optimisation approach with a pure random search (PRS) that took an equal amount of processing time. Since we have no ground truth available, we have created alternative methods of evaluation: if each pea’s velocity is within an expected range, and if each pea’s motion is correlated with neighbouring peas. We found that the optimised version was an improvement over the PRS for both the range accuracy objective (56.25%–61.67% optimised, 52.92%–59.17% for PRS) and the mean dot product objective (0.15–0.20 optimised, 0.12–0.15 for PRS) for each individual on the Pareto front from each dataset tested. Furthermore, we found many interesting food motion behaviours. For example, many peas appeared to collide, resulting in altered trajectories. Additionally, many peas had correlated motion, indicative of what would be expected as the stomach mixes its contents. Overall we found that the combined optimisation and clustering approach produced interesting findings relating to food dynamics in a human stomach.