it didn't quite wrap but the lavash withstood the strain :)
katch wreck
@katchwreck
- 398 Followers
- 1,050 Following
- 9.7K Posts
partially-recovered workaholic ;) being myself and helping to build the world i want to see. contributing to human knowledge, and scientific knowledge in particular. problem-solving in biomedicine (e.g. noninvasive biomarker identification) is my career focus. i am in san francisco (cental richmond district) and i enjoy going to ocean beach and the presidio. i'm learning jazz bass and i also DJ once in a while, see https://mixcloud.com/katchwreck
corresponding training & test AUC-ROCs after reducing per-shapetype 1D Bayes classifier predictions to per-channel ROI predictions (reduction via a logical heuristic). each electrode position corresponds to a color code. red & green are contralateral to the trauma (brain lesion). yellow & black are ipsiperilesional. errorbars indicate 95% confidence intervals (jackknife estimate). holdout test partition came from the epileptogenic epoch (Day 1-7 post-TBI), training partition ~6 months post-TBI
according to facebook, this was a sandwich i made 17 years ago... thanks, i guess? :-P
while i enjoy algorithm development, data analysis, and programming, i am really glad that this study is almost over! so exciting to not feel like i'm chained to my desk for months at a time :) we currently compute two shape-feature distributions per electrographic waveform, yielding a total of 8 feature spaces for this 4-channel dataset. here is a montage showing the top-ranked ROI from each feature space! i fixed all hyperparameters except one, red outline shows top-ranked ROIs.
truth be told, we use two correlated but not identical "shape types" for each waveform we characterize. and, it turns out the other shapetype on the green also yielded a similar, but not identical, ROI! it was by combining these two ROIs, which were the top two ranked from the green channel based on training performance, that yielded the AUC-ROC > 0.95.
... imagine my surprise when i saw almost exactly the same ROI appear from yesterday's results using an automated system restricted to the green channel! guess who's baaaack :) but now i can show you a better plot, overlaid onto the heatmap used by the nonparametric ROI test i formulated to help us identify these subtle shifts in probability mass between the TBI+, PTE- and TBI+, PTE+ groups.
in essence, this formula allows us to infer topologic relations from topographic ones
i'm still fine-tuning the design of the test. there are a lot of adjustment you can make, but in the interest of minimizing parameter number, four will hopefully be enough! this is the most specific ROI we've found for post-traumatic epilepsy in this dataset.
