How artificial intelligence improves radiological interpretation in suspected pulmonary embolism.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Pulmonary embolism (PE) diagnosis relies heavily on CT pulmonary angiography (CTPA).
• Evaluating AI diagnostic tools against human expertise is crucial for clinical integration.
• Emergency radiologists face challenges in interpreting CTPAs, particularly with suboptimal image quality.

Purpose of the Study:

• To compare the diagnostic performance of a commercial AI algorithm for PE detection against emergency radiologists.
• To assess the impact of CTPA quality on the diagnostic accuracy of both AI and radiologists.
• To evaluate radiologists' perception of AI assistance in PE diagnosis.

Main Methods:

• Retrospective, multicentric study of 1202 patients with suspected PE.
• Comparison of AI algorithm outputs with radiologist interpretations from reports.
• Gold standard established through comprehensive review of imaging, reports, and outcomes.
• Performance metrics analyzed across the entire cohort and stratified by CTPA quality.

Main Results:

• AI demonstrated high sensitivity (92.6%) and NPV (98.6%), outperforming radiologists in these metrics.
• Radiologists achieved higher specificity (99.1%) and PPV (95%).
• AI's performance was particularly beneficial in subcohorts with poor-to-average CTPA injection quality.
• A significant majority of radiologists (69.6%) reported improved diagnostic comfort using AI.

Conclusions:

• AI serves as a valuable safety net in emergency radiology for PE detection, enhancing sensitivity and NPV.
• AI integration can increase radiologists' confidence, especially in cases with compromised CTPA quality.
• Emergency radiologists expressed positive feedback and recommended AI use to augment their diagnostic capabilities.