clinical study

Man and Machine: Implications of an Artificial Intelligence Based Algorithm in the Detection of Pulmonary Embolism on Computed Tomography Angiography

Materials & Methods

A retrospective study evaluated the diagnostic accuracy of AI-based detection of PE on CTA chest exams compared with diagnostic radiologists. CTA chest exams were collected using the AI algorithm. NLP assigned positive/negative results on each case per radiologist interpretation. Discordant cases were reviewed by three independent radiologists. 


1,281 consecutive CTPA chest exams over two weeks from July 2020. The performance of the AI was: sensitivity: 99% (CI: 94.6% – 100.0%); specificity: 99.3% (CI: 98.7% – 99.7%) and miss rate 0.1%. The performance of the unaided radiologist (no-AI) was: sensitivity: 86% (CI: 77.6% – 92.1%) specificity: 99.7% (CI: 99.3% – 99.9%) and miss rate 1.1%. The AI demonstrated a 14% enhanced detection rate compared to the un-aided (by AI) radiologist. The miss rate of the un-aided radiologist resulted in about one in seven PEs missed, of which 57% (8/14) were segmental.


The study demonstrates the benefit of the complementary performance of the radiologists plus AI is better than either party alone; standalone AI was superior to the standalone radiologist for sensitivity and how the opposite was true for specificity.