The diagnostic probability of pulmonary embolic disease is based on the recognition of unmatched segmental perfusion defects. Although interobserver and intraobserver reproducibility have been studied, accuracy has been an elusive goal due to the lack of a gold standard. We investigated the accuracy and reproducibility of reporting in a virtual scintigraphic model of the lungs, with and without the use of a lung segmental reference chart.
Methods: A Monte Carlo package was used to model lung scintigraphy from a digital phantom of the human lungs. An ideal lung segmental reference chart was created from the phantom. Five experienced nuclear medicine physicians reported a set of all possible defects involving 100% of a segment, without and with the chart. A further set of defects involving 45%-55% of a segment in the lower lobes was investigated using the chart.
Results: There was a significant improvement in accuracy (from 48% to 72%) and intraobserver agreement (from 61% to 77%) with the chart. The accuracy of reporting defects in the upper and middle lobes was consistently better than that in the lower lobes. There was no significant difference between the accuracy of reporting large defects and that of reporting moderate defects in the lower lobes.
Conclusion: The lung segmental reference chart significantly improves both the accuracy and reproducibility of reporting lung scintigrams; however, although reporting in the lung bases is improved, absolute accuracy is substantially less than that in the upper and middle lobes. This emphasizes the need for caution because the lung bases are the most common site of embolic disease.