An Automated Method for the Detection of Pulmonary Embolism in V/Q-scans

Attila Frigyesi


Centre for Mathematical Sciences
Mathematical Statistics
Lund Institute of Technology,
Lund University,
2001

ISSN 1403-9338
Abstract:
In this paper a fully automatic method for the diagnosis of pulmonary embolism from V/Q-scans is presented. Image analysis is applied to the ventilation and the perfusion images obtained in the V/Q-scan. The difference of the ventilation and the perfusion is calculated after warping and hot-spot reduction of the images. From this difference image the integral of the underperfused areas are used as features. With the aid of these features we define a simple test for pulmonary embolism.
Our method is evaluated on two sets of patients. One set comprises 102 patients who have undergone both V/Q-scanning and angiography. The performance given as the area under the ROC curve is 0.85. Another set is made up by the 507 consecutive patients examined with V/Q-scanning at Lund University hospital in Sweden.In this case, were the reference was the consensus opinion of two radiologists, the ROC-area is 0.67.
Thus we have developed a fully automatic and robust expert system to aid the radiologist in the interpretation of V/Q-scans for pulmonary embolism.
Key words:
Pulmonary Embolism, V/Q-scan, Image Analysis, Hot-spots, erosion, opening, ROC