Projects and Projections

Now that the term is winding down I'm beginning to feel the crunch to finish my project. It will require a fair amount of effort, but I think it can be completed with time to spare. The parameters of my project are pretty simple. Dr. Schwartz is very interested in epilepsy, and specifically methods for treating cases of epilepsy that are refractory to drug therapy. Often in these cases, there is one localized portion of the brain responsible for the epileptic activity. These patients can opt to have this portion of the brain removed (which is where Dr. Schwartz comes into play), thereby eliminating the seizures. But removing a part of somebody's brain is obviously costly, dangerous, and potentially debilitating. Is it possible to do any better? Well maybe. However, first more knowledge needs to be attained about the way depolarization waves spread in the brain.

A medical student working in Dr. Schwartz's lab took massive amounts of data over the course of two years, stimulating rat brains at various frequencies and analyzing the hemodynamic response. One of the things Dr. Schwartz has asked that I do during my stay at Weill is analyze the data in a meaningful way, so that blood flow response can be correlated with the stimulus. As always in biological, medical, and engineering fields, this task is a little more convoluted than it superficially appears. Firstly, it requires sorting through volumes of data to determine which is "good" and which is "bad." Not to be confused with desirable versus undesirable data, there is actually an issue of noise corrupting some of the data to the point that it is totally unusable. In general, the data can be viewed by the equation: data=gain*truth + bias + noise. I'm trying to cleverly extract the truth from this equation so it can be analyzed. In a perfect world, gain would be 1, and bias and noise would be 0, but I don't expect to be so lucky. Fortunately there is a ton of information on estimation and detection methods, which I can use to hopefully solve this problem.