A team including researchers from UCSF's Center for Intelligent Imaging (Ci2) created and tested an algorithm to detect the presence of inferior vena cava (IVC) filters on radiographs. This automated detection will screen radiographs without human assistance to help identify which patients need IVC filter retrieval.
The investigators, led by first author John Mongan, MD, PhD, and senior author Andrew Taylor, MD, PhD, share their conclusions in "Automated detection of IVC filters on radiographs with deep convolutional neural networks." The work was published in Abdominal Radiology.
The authors relied on a dataset of 5,225 images to train and test a Cascade R-CNN (Region Based Convolutional Neural Network) object detection network. Of the images, 30% included IVC filters. The investigators used 85% of the images to train the algorithm and the remaining 15% to test it. The team also tested the model with an independently constructed dataset of 1,424 images.
The tests revealed a high degree of sensitivity with a result of 96.2% on the first test and 97.9% on the second. The specificity of the algorithm was even higher with a scores of 98.0% and 99.6%, respectively.
"Fully automated detection of IVC filters on radiographs with high sensitivity and excellent specificity required for an automated screening system can be achieved using object detection neural networks," concluded the investigators. "Further work will develop a system for identifying patients for IVC filter retrieval based on this algorithm."
The investigators also looked closely at the images that resulted in false positives and negatives to find any commonalities among the failures.
Dr. Mongan is the associate chair of translational informatics for the UCSF Department of Radiology and a Director of UCSF Ci2. Additional investigators from Ci2 and UCSF Radiology include senior author and Associate Professor Dr. Taylor and Professor Marc Kohli, MD. Co-authors Roozbeh Houshyar, MD, Peter Chang, MD, and Justin Glavis-Bloom, MD, are current and former members of the Center for Artificial Intelligence in Diagnostic Medicine at the University of California Irvine.
Read more about research and news at UCSF Ci2.
