Computer-aided detection of pulmonary embolism on CT angiography: initial experience

Abstract

Objective: To evaluate the performance of a computer-aided detection (CAD) system for diagnosis of pulmonary embolism on computed tomography (CT) pulmonary angiography.

Materials and methods: One hundred and four pulmonary CT angiograms for pulmonary emboli (PE) were reviewed both by radiologists and a CAD detection system (ImageChecker CT V2.0, R2 Technology Inc, Sunnyvale, CA). CT scans, read and reported by radiologists in a routine daily clinical setting, were later processed by the CAD system. The performance of the CAD system was analyzed.

Results: Forty-five PE were identified by the radiologists in 15 patients. The CAD system revealed 123 findings, interpreted by the system as PE. Twenty-six of them, detected in 8 patients, represented true-positive results. Ninety-seven (78.9%) CAD findings were not true PE and were defined as false-positive. Nineteen true PE in 7 patients were missed by the CAD system constituting 42% false-negative rate. Sensitivity of the CAD system was 53.3% and the specificity was 77.5%. The positive predictive value of CAD system was 28.5% and the negative predictive value was 90.7%.

Conclusions: With the evaluated CAD system, it is relatively simple and fast to check all detected findings and decide if they represent true PE. However, high false-negative results demand technologic improvement, to increase the sensitivity of the system. It is anticipated to become a promising supplement to the work and eyes of the radiologist in detecting PE on pulmonary CT angiography.