Rapid Quantitative Pharmacodynamic Imaging with Bayesian Estimation

Jonathan M. Koller, M. Jonathan Vachon, G. Larry Bretthorst, Kevin J. Black
  • Frontiers in Neuroscience, April 2016, Frontiers Media SA
  • DOI: 10.3389/fnins.2016.00144

Return of QuanDyn(TM)

What is it about?

This is the second publication on our method for rapidly characterizing the sensitivity of a body part to a drug using imaging. It is a reanalysis, using an improved method, of simulated data and of a limited set of real-life data.

Why is it important?

Scientists often measure how sensitive a body part is to a medicine. The usual way is to get a bunch of people (or rats or yeast), and give a few of them (say) 1/10 of a pill, a few 2/10, a few 5/10, a few 1 pill, and so on up to maybe 10 pills. Plus you have to give some of them an empty pill with no medicine. Then you measure the response from each person (or rat or test tube), and turn the results into a graph called a dose-response curve. The problems come when you want to do that in live people. To calibrate the dose-response curve, you have to give some people higher doses than you will end up using. In my work there's an additional problem, because the brain is complex and I want to look at all of it. Besides, sometimes you want to know a single individual's response--if Fred needs the drug, he doesn't really need to know the right dose for Suzy. That's the question our method is trying to answer. This paper simply shows the results with our new, upgraded data analysis approach, using the same idea and data from our 2013 paper.


Dr Kevin J. Black
Washington University in St. Louis

Back in October, 2007, Jon Koller and I submitted a manuscript to Neuroimage (https://arxiv.org/abs/1304.5756v1). We were very excited about it. One reviewer asked us to change the analysis method from an iterative to a continuous approach. We'd tried that but the traditional numerical methods we were using were inefficient and tended to get stuck at local minima. Trying to solve that took long enough that we eventually gave up on the Neuroimage submission. A revised version was published in 2013 (https://peerj.com/articles/117/). In the meantime, the reviewer's request had led us to Larry Bretthorst and his Bayesian image analysis tools (thanks, Josh Shimony), which eventually produced much improved results, which now appear in this 2016 Frontiers in Neuroscience paper.

Read Publication


The following have contributed to this page: Dr Kevin J. Black