Deep learning-based algorithm for automatic detection of incidental pulmonary embolism on contrast-enhanced CT: a...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Incidental pulmonary embolism (iPE) is increasingly detected on contrast-enhanced computed tomography (CECT) scans performed for other indications.
• AI offers potential to improve detection accuracy and timeliness of iPE.
• Underreporting of iPE may occur due to the primary focus of diagnostic imaging studies.

Purpose of the Study:

• To evaluate the diagnostic performance and triage effectiveness of a standalone AI solution for detecting iPE.
• To assess the AI's ability to identify iPE in CECT exams not originally performed for PE evaluation.

Main Methods:

• A deep learning-based software (CINA-iPE) was used to analyze CECT images for suspected iPE.
• Retrospective CECTs from 5 clinical centers were collected, forming a balanced dataset.
• A reference standard was established by three U.S. board-certified radiologists.

Main Results:

• The AI achieved 87.8% sensitivity and 92.0% specificity for iPE detection, with an overall accuracy of 90.0%.
• The system processed results within 1.5 minutes, enabling rapid notification.
• Disagreements among radiologists occurred in 50% of false positive cases and 45.5% of missed PE cases, often involving complex findings.

Conclusions:

• The CINA-iPE application demonstrated high accuracy in identifying incidental PE on non-PE CECT studies.
• Rapid availability of AI-processed results can help prioritize interpretation and potentially enhance iPE detection.
• The AI tool shows promise for improving the accuracy and speed of incidental pulmonary embolism detection.