Automated IVC Filter Detection

There is so much talk in medicine, and radiology in particular, about the coming AI boom that will perform everyone’s job for them. Ask anyone who has used the available tools, however, and it quickly becomes aparent that we are a long way from a single “do it all” piece of software. Instead, many algorithms are written to perform small (cognitively simple) tasks with the goal of eventually chaining many of them together.

In a recent quest, Howard showed off some cool image-processing capabilities built into Python. Using a combination of segmentation and masking, he demonstrated the start of a processing pipeline for pulmonary nodule detection.

This quest builds off many of the techniques Howard used. We are going to combine these skills with techniques common to facial detection algorithms (think Facebook) to search for the presence of an inferior vena cava (IVC) filter within abdominal CTs. I originally developed this algorithm using and internal dataset and it is surprisingly difficult to find publicly-available datasets with IVC filters. Fortunately, the Cancer Imaging Archive is an amazing image repository. However, I admit that I did have to do a lot of manual searching to find the example scan used below.

Joe Wildenberg
Joe Wildenberg, MD/PhD is an interventional radiology fellow at University of Pennsylvania. His interests include applying data science to radiology and medical data, interventional radiology, and traveling the world.

Joe will finish training in June 2018 and is moving back to Wisconsin to practice with the Mayo Clinic Health System.

2 Responses to “Automated IVC Filter Detection

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.