Radiologists lose to computer when assessing anemia

Radiologists' subjective observations don't fare well compared with a CT scanner's objective assessment of anemia on thoracic CT, according to a study in the November issue of the American Journal of Roentgenology.

Studies of anemia in animals have shown that it is possible to distinguish the boundaries of the ventricular and atrial cavities, the papillary muscles, the major trabeculae, and the aorta. Similar studies in humans, however, have not been extensively performed.

There are no data comparing qualitative and quantitative assessment of the attenuation of blood in the left ventricle on noncontrast thoracic CT, according to Dr. Rachel S. Title and colleagues at Boston Medical Center.

"Despite the efficacy of Hounsfield unit measurements to assess anemia, we have observed reviewers diagnose anemia with only a visual inspection, seldom performing an objective measurement for confirmation. This practice sometimes causes consternation among our clinical colleagues," said Title, now with Weill Cornell Medical Center in New York City.

Researchers found that the quantitative assessment of anemia (measured in Hounsfield units*) was significantly better than subjective reviewer analyses for differentiation of an anemic from a nonanemic state (area under ROC curve = 0.85 versus 0.72, 0.70, and 0.69, respectively, for three reviewers for a p<0.05).

For the study, three radiologists retrospectively reviewed 102 noncontrast thoracic CT exams and visually assessed anemia. A fourth recorded HU measurements of the blood in the left ventricle. Twenty-five of the 102 patients had a hemoglobin level of less than 10 g/dL, the cutoff for anemia.

The mean density for anemic patients was 31.8 HU, compared with 42.8 HU for nonanemic patients.

Setting the CT density threshold at 35 HU, researchers found a 76% sensitivity and 81% specificity for anemia. The sensitivity of the three reviewers was between 40% and 72%, while the specificity was between 60% and 83%.

Interobserver agreement was poor. Seven of 25 anemic patients were correctly classified by the reviewers as definitely or probably anemic, and 24 of 77 nonanemic patients were correctly classified by the reviewers as definitely or probably not anemic.

The researchers concluded that the visual method is unreliable. They suggest that even if anemia is suspected by visual assessment, objective measurement of density values in Hounsfield units of the ventricular blood is warranted for more confident diagnosis.

For more information from the Diagnostic Imaging archives:

MSCT tackles acute chest pain in emergency room

Death of CT inventor Godfrey Hounsfield signals end of era

*Hounsfield units measure density on CT. A number (between -1000 and 1000 HU) is assigned by the computer to represent the difference in x-ray attenuation between a given material and water, where air is -1000 HU and pure water is 0 HU. This number is then used by the computer to assign a gray-scale shade to the represented image. Current CT scanners can detect differences in contrast of less than 5 HU (this information is from the study).