Edward Herskovits
We were initially approached about this topic when one of our research collaborators in China told us about this dataset, which included DTI and clinical assessments of 65 patients with cirrhosis. Our collaborators were wondering whether we could apply spatial data-mining techniques to determine connectivity differences between subgroups with minimal hepatic encephalopathy (MHE) and those without encephalopathy. We had previously applied our Bayesian data-mining approach, Graphical-Model-based Multivariate Analysis (GAMMA), to brain MR imaging data to detect group differences in other contexts, and were therefore hopeful that it would yield interesting results for the MHE data.
Even with a small sample size, we found significant connectivity differences between the 2 groups in the frontal lobe, the parietal lobe, and the corpus callosum. Fifteen-month follow-up confirmed different mortality rates for those with and without these connectivity differences, independent of the Child-Pugh score. We look forward to confirming these preliminary results using data from other sites and larger sample sizes.
This research exemplifies our lab’s overarching goal of developing data-derived clinical practice guidelines. We have also designed data-driven approaches to brain atlas construction and to detecting patient subtypes, which is particularly salient to better characterization of complex disorders such as schizophrenia, Parkinson disease, and autism. We believe that a data-driven approach, in conjunction with clinical expertise, will allow physicians and researchers to best implement precision medicine.