To train and test the CAD algorithm, it's important to establish a reference standard by using manual marking of PE by radiologists on CTPA PE cases. However, previous studies, such as the Lung Imaging Database Consortium (LIDC), have found wide variability in radiologists' detection and segmentation of lung nodule boundaries on CT images, said presenter Chuan Zhou, PhD.
As a result, the researchers sought to evaluate the variability of PE identification by radiologists in 40 cases. They also applied their prototype CAD system to the same dataset to investigate the potential benefit of CAD, Zhou said.
The researchers found substantial agreement between the independent markings of pulmonary embolism by radiologists. However, consensus with multiple radiologists improved the reference standard, he said.
"The results also showed that our CAD system detected 23 PEs that were missed by radiologists, which demonstrated the feasibility of the CAD system for assisting radiologists in PE detection," Zhou told AuntMinnie.com.