Artificial intelligence in computed tomography imaging for pulmonary embolism: a narrative review from computed tomography pulmonary angiography to non-contrast computed tomography.
Authors
Affiliations (2)
Affiliations (2)
- Department of Radiology, Chinese PLA General Hospital, Beijing, China.
- College of Medical Technology, Beijing Institute of Technology, Beijing, China.
Abstract
Pulmonary embolism (PE) is a common life-threatening cardiovascular disorder. Computed tomography pulmonary angiography (CTPA) is the diagnostic reference standard, but iodinated contrast use may be limited by renal insufficiency, allergy, or other patient factors. Non-contrast computed tomography (NCCT) avoids contrast administration but has lower intrinsic diagnostic accuracy. This narrative review summarizes artificial intelligence (AI) applications for PE assessment across CTPA and NCCT, compares their clinical roles, and identifies evidence gaps for translation. A structured literature search was conducted across PubMed, Google Scholar, Web of Science, and IEEE (Institute of Electrical and Electronics Engineers) Xplore from database inception to June 2026. Search terms covered PE, CTPA, NCCT, AI, deep learning, machine learning, and related concepts. Eligible evidence included peer-reviewed original studies, validation studies, and peer-reviewed conference proceedings reporting AI methods, quantitative findings, or clinically relevant methodological insights. AI for CTPA has progressed from computer-aided detection toward segmentation, clot burden quantification, workflow triage, and risk stratification, with encouraging performance in selected retrospective and real-world studies. AI for NCCT remains exploratory and includes indirect sign analysis, direct detection, synthetic contrast generation, and cross-modality learning. Major limitations include small heterogeneous datasets, limited multicenter validation, suboptimal subsegmental PE detection, and uncertain lesion-level safety of synthetic imaging. AI may improve diagnostic efficiency and support radiologist interpretation, but it should remain an augmentative tool rather than an autonomous replacement for clinical expertise. CTPA-based AI has more mature evidence for central PE detection and workflow triage, whereas NCCT-based approaches require rigorous validation before broad clinical implementation. Priorities include multicenter datasets with vessel-level annotations, direct comparison of NCCT strategies under common reference standards, and lesion-level validation of synthetic imaging safety.