SPECT imaging in the diagnosis of pulmonary embolism: automated detection of match and mismatch defects by means of image-processing techniques.

UNLABELLED

SPECT of ventilation/perfusion (V/Q) lung scans not only improves the diagnostic accuracy of the method but also facilitates the application of advanced image-processing techniques. On the basis of such techniques, our study aimed at developing a procedure that automatically analyzes V/Q lung scans with regard to match and mismatch defects.

METHODS

Fifty-three patients with suspected pulmonary embolism had lung scans using the SPECT technique as well as 16-slice multidetector-row spiral CT within an interval of 48 h. After iterative image reconstruction and computerized linear registration of the V/Q scans, the ventilation was normalized to the perfusion. For the automated detection of mismatch defects, the perfusion was subtracted from the ventilation, whereas for the detection of match defects, the perfusion was subtracted from the inverted ventilation. Two experienced referees assessed all images. The final diagnosis was made at a consensus meeting while taking into account all of the imaging modalities, laboratory tests, clinical data, and evaluation of a follow-up period.

RESULTS

The sensitivity, specificity, and accuracy of the conventional visual assessment were 0.91, 0.97, and 0.94, respectively, compared with 0.95, 0.84, and 0.89, respectively, for the automated algorithm. Artifacts imitating mismatch defects in the pulmonary recesses accounted for the relatively low specificity of the automated analysis. Artifacts of that kind were found in 15 patients and led to a false-positive diagnosis in 5 patients. However, by combining the visual and the automated approach, all artifacts could be easily identified leading to a sensitivity, specificity, and accuracy of 0.95, 1.0, and 0.98, respectively. Additionally, in all 12 patients of the cohort with highly heterogeneous ventilation and perfusion, the automated analysis made correct diagnoses.

CONCLUSION

Because of the 3-dimensional properties of the SPECT data, the analysis of lung scans can be automated and objectified. The algorithm produces images that are easy to read and well suited for demonstration. Because of artifacts in the pulmonary recesses introduced by the automated approach, its diagnostic accuracy does not reach the level of the conventional analysis yet. Could these artifacts be overcome, the efficiency of the automated algorithm would be at least equivalent to that of conventional image interpretation. At present, best results can be achieved by combining both approaches.