Kidney Segmentation and other Adventures
The past few weeks have come to a head. I previously spoke about what disease my project focused on, but not how the disease is addressed. Kidney volume is a pretty important in determining polycystic kidney disease (PKD) outcomes in patients. Unfortunately, total kidney volume (TKV) measurements come from manually contouring kidney. It can be time consuming to contour an unremarkable kidney, but the complicated geoemetries and partial voluming effects found in advanced cases makes the process even more difficult.
So this is where we set off: why not try to automate the segmentation? Luckily, a research fellow developed and trained the model, and we set off to evaluating inferences and further debugging the code. For many cases, the inference performed very well, but we found that we need to train it with some harder stuff. In other words, a harder case was more of a coin toss with the algorithm, but after much thought we are pretty optimistic that re-training with a more diverse dataset will go a long way.
In other news, I found myself in an impromptu shadowing experience. I wanted to watch some CT guided ablations for my upcoming research project, and I had to head over to the hospital to talk over the details with one of the interventional radiologists. I got there and after talking with him for 10 minutes, he realized "We have one going on right now. Let's watch!" The flexibility was amazing, and I got a small taste of the procedure before I get the full introduction next week!

Comments
Post a Comment