Variability of perceived defect size in virtual lung scintigraphy

The diagnosis of pulmonary embolism is based on the presence of mismatched segmental or subsegmental defects. An important axiom is the classification of defect sizes into small, moderate and large. Little information about the recognition and classification of such defects has been published. We undertook a study of the perception of defect size using a model of the virtual scintigraphic anatomy of the lungs to address this issue.

Segmental anatomy of the lungs was modeled with CT, cadaveric lungs and standard anatomical tests. The emission, scatter and attenuation of photons were modeled within these virtual lungs and the surrounding tissues. Single segmental lesions, each 100% of a segment, were created in eight projections and submitted for blinded reporting by four experienced nuclear medicine physicians to obtain their assessment of the size of each defect on two occasions.

Of the 144 defects submitted for reporting, 15% were reported as <25% of a segment, 35% were reported as 25%-75% and 50% were reported as 75%-100%. The accuracy of each reporter and the intraobserver agreement were calculated; the weighted kappa value ranged from 0.34 to 0.60. The segmental defects that were most likely to be underestimated in size were in the right lower lobe.

It is clear that segmental defect sizes were underestimated, particularly in the right lower lobe. Although the intraobserver agreement in reporting was fair, the accuracy of estimation was only 50%. The variability and inaccuracy might be reduced by the use of a guide to segmental anatomy.