Deep Learning-based Automated Detection of Pulmonary Embolism: Is It Reliable?

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Pulmonary embolism (PE) diagnosis relies heavily on chest computed tomography pulmonary angiograms (CTPAs).
• Automated detection of PE using artificial intelligence (AI) can potentially enhance diagnostic efficiency and accuracy.
• The Canon Automation Platform is an AI tool designed for automated PE detection in CTPAs.

Purpose of the Study:

• To evaluate the diagnostic accuracy and clinical utility of the Canon Automation Platform for automated PE detection and localization in CTPAs.
• To assess the performance of the AI program across all levels of pulmonary arteries.

Main Methods:

• Retrospective analysis of 1474 CTPAs with suspected PEs.
• Evaluation by senior radiology residents and verification by experienced thoracic radiologists.
• Integration and assessment of the Canon Automation Platform's diagnostic performance with PACS.

Main Results:

• The AI platform demonstrated high diagnostic accuracy, with AUC-ROC scores ranging from 0.945 to 0.996.
• Overall accuracy was high (95.4%–99.7%), with excellent specificity (98.7%–100%).
• Sensitivity was slightly lower in subsegmental branches (81.4%–84.7%) but remained high overall.

Conclusions:

• The Canon Automation Platform effectively detects PE in CTPAs using deep learning, streamlining clinical workflows.
• AI offers robust diagnostic support for acute PE, particularly beneficial for institutions lacking 24/7 radiology specialist access.
• The AI tool shows significant promise in improving the management of patients with suspected pulmonary embolism.