Week 7

It’s finally happened! It took all of the immersion term, but I was finally able to run some simple models on the data that I had received. It took a while to get to that point after collecting the data – (thanks to finding extra whitespaces in some of your spreadsheets that ends up throwing errors that take 30 minutes to debug), but there are preliminary results now.

Without getting too much in the way of scooping potential future work, preliminary results point to the importance of the perpendicularity of the angle relative to the pleural surface and the use of a collagen plug while finishing the procedure. Of interest is how we can use these results to now attempt to improve patient health prognosis, as both criteria are difficult to always meet – the collagen plug is an additional expenditure of ~500 dollars, and optimizing the angle is not always a feasible task, given the other constraints in place (some trajectories are not feasible) and the inability to discern the precise angle between the pleural surface and the needle along the trajectory to the lesion. Though it’s outside the scope of what I’m capable, we discussed potentially having someone comfortable with image processing design a program that can automatically provide those paths! It would certainly alleviate some of the downfalls of having the operator still decide.

With respect to how some of the analysis was performed – as mentioned previously, the data collected was fit to a logistic regression model. What was done to validate this was a k-fold stratified cross-validation, ie, iteratively, k times, separating the data into a testing set and a training set that still maintains the same ratio of pneumothorax positive samples, ensuring the model is being trained on a representative set of data. As a spoiler to how positive things seem, our predictive accuracy was ~85%!

With these results, it means that I’ll be able to keep having a bit of immersion contact, as we move towards publishing/presenting. It’s a satisfying feeling to have been able to come in and find a novel result in such a short time. Between the research output and the clinical work, this immersion experience has definitely been something I won’t soon forget.