Exploratory validation of an AI-based pulmonary embolism detection tool: Bridging technology and clinical practice

Background: The survival of patients with acute pulmonary embolism (PE) is dependent on timely and accurate diagnosis. Recently, artificial intelligence (AI) has emerged as a promising tool to enhance diagnostic performance. This study aimed to evaluate the diagnostic accuracy of a novel AI algorithm developed to detect acute PE on computed tomographic pulmonary angiography (CTPA). Methods: In this single-center observational study, we assessed the diagnostic performance of the AI software using 100 consecutive PE-positive and 100 consecutive PE-negative CTPA cases from patients with suspected PE. The ground truth was established by consensus among three board-certified experts. The AI results and radiology reports were compared against this ground truth. Results: The sensitivity and specificity of the AI were 95.0% (95% confidence interval (CI): 88.7–98.4%) and 99.0% (95% CI: 94.6–100.0%), respectively. Overall, 194 out of 200 cases (95 PE-positive and 99 PE-negative) were correctly identified by the AI, yielding an overall accuracy of 97.0% (95% CI: 93.6–98.9%). Interestingly, the AI software and radiology reports showed discordant patterns of misclassification. Conclusions: This AI software demonstrated excellent diagnostic performance for detecting PE on CTPA. Our research suggest a potential complementary role for AI as an adjunctive diagnostic tool.