Ces radiol. 2008, 62(3):270-276

Automated detection (CAD) in CT diagnostics of pulmonary embolism: Possible benefit for junior radiologist?Original article

Jan Baxa¹, Jiří Ferda¹, Alena Bednářová¹, Renata Vondráková¹, Hynek Mírka¹, Lucie Baxová², Boris Kreuzberg¹

¹ Radiodiagnostická klinika LF UK a FN, Plzeň

² I. interní klinika LF UK a FN, Plzeň

Aim: Of our study was to evaluate the accuracy of CAD tool for automated detection of segmental and sub-segmental pulmonary embolism and the capability of this software to help junior radiologist in evaluation of CT pulmonary angiography (CTPA).

Method: We selected 18 patients (9 women aged 17-79 years, men aged 58 years) from our set of CTPA's with total of 78 emboli (41 segmental/37 subsegmental; 4.33 emboli per patient). As a control group we randomly picked 18 patients with negative CTPA. All 36 examinations were analysed by CAD tool (PE-CAD, Siemens Medical Solutions, Germany), by junior radiologist (2.5 years experience with CT) alone and in consensus with the CAD tool and finally by senior radiologist (14 years experience with CT). The findings of head senior radiologist (15 years practice of thorax CT) were considered as reference for presence of all of emboli.

Results: CAD correctly detected 44 emboli (30 segmental, 14 subsegmental), the average positive rate of the CAD was 3.1 per examination (overall 112). As for the analysis of segments, the overall (segmental + subsegmental) sensitivity of the CAD was 56.4% (segmental = 73.2%; subsegmental = 37.8%) and the positive predictive value (PPV) was 42.3%. Junior radiologist profited by the consensus with the CAD, their overall sensitivity increased from 83.3% to 87.2% (segmental - from 95.1% to 97.2%; subsegmental - from 70.2 to 75.2%) and the PPV from 93.2% to 95.6%. Senior radiologist achieved the overall sensitivity of 92.3% (segmental = 100%; subsegmental = 94.1%) and PPV of 96.1%.
As for the analyses of patients, the sensitivity of CAD was 83.3% and the negative predictive value (NPV) was 78.6%, the sensitivity of junior radiologist in consensus with the CAD was 94.5% and the NPV was 93.8%, the sensitivity and NPV of senior radiologist were 100%.

Conclusion: Our own results show the capability of the CAD tool to improve performance of junior radiologist in detecting of segmental and subsegmental pulmonary embolism at CTPA. CAD is feasible as a "second reader" in the evaluation of CTPA, but low sensitivity in detecting subsegmental embolism and high positive rate demand further improvement.

Keywords: automated detection, computer - aided diagnosis, CT pulmonary angiography, pulmonary embolism

Grants and funding:

Práce byla podpořena výzkumným projektem MSM 0021620819.

Accepted: August 5, 2008; Published: September 1, 2008  Show citation