AJR: Image quality linked to pulmonary embolism CAD false positives
Pulmonary embolism (PE) exclusion may be an ideal candidate for CAD. Multi-slice CT has become the modality of choice for diagnosis, but it produces large datasets, resulting in a time-consuming diagnostic process that may still overlook small embolisms.
Previous studies of PE CAD algorithms have examined their impact on radiologists’ performance or tested the stand-alone performance of the algorithm. A group of Dutch researchers undertook the current study to attempt to quantify the relationship between image quality and the number of false positive marks on PE CAD results.
The retrospective study comprised 278 pulmonary CT angiographies acquired during night and weekend shifts between January 2007 and April 2008 at an academic medical center. The prototype CAD algorithm performed multiple steps: lung and airway segmentation, vessel segmentation, analysis of contrast differences in the vessels, candidate generation and filtering to reduce false positive findings, wrote lead author Rianne Wittenberg, MD, of the department of radiology at University Medical Center in Utrecht, the Netherlands.
Two independent readers established a reference standard. CAD findings were compared to the reference and scored as true positive, false positive or false negative. Researchers classified false positives according to the most likely underlying reason and identified anatomic and scanning technique causes.
The team scored image quality parameters (overall image quality, vascular enhancement, motion artifacts and image noise) according to a five-point scale, graded accompanying lung disease and completed a statistical analysis to assess the relationship between CAD performance and image quality parameters, explained Wittenberg and colleagues.
The CAD algorithm detected 258 true positive findings and 1,298 false positive findings and marked a mean 4.7 false positives per exam. Although 30 percent of false positive findings were located in veins and 22 percent in areas with airspace consolidations, the remaining false positives were related to image quality, including 16 percent caused by motion artifacts and 14 percent caused by inadequate arterial enhancement.
The researchers found “a strong association between the number of false positive CAD findings and the overall quality, motion artifacts, vascular enhancement, noise and accompanying lung disease.
“It is very likely that the quality of pulmonary CT angiography influences the number of false positive findings found by CAD algorithms. Poorly timed contrast injection, motion artifacts and noise all directly influence the intravascular homogeneity and, thus, the detection of a CAD algorithm that looks for intravascular contrast differences,” wrote Wittenberg.
Wittenberg and colleagues shared several conclusions. Specifically, the investigators stated that increased CAD performance can be realized by making arterial venous separation an integral part of the algorithm. Alternatively, the contrast injection protocol can be adapted to more homogenously contrast enhance arteries and veins; however, this method does not address motion artifacts or anatomic misinterpretation.
The researchers emphasized the importance of good image quality as a prerequisite to PE CAD application in clinical practice, noting that the 72 percent of studies with five or fewer false positives had image quality assessments of four and five for the various parameters. They identified a few limitations to the study including the lack of an absolute standard of truth for the presence of PE in pulmonary CT angiography exams. They also acknowledged that different scan protocols, scanner types or study populations might produce different results at other institutions.
Wittenberg pointed out that the link between image quality and CAD performance is likely to be universal and emphasized the importance of software improvements or contrast protocol changes.