Pulmonary embolism ranks among the leading causes of preventable cardiovascular death in Europe and the United States [1], accounting for more than 100,000 deaths annually in the US alone. It is treatable when caught in time. Yet despite decades of clinical awareness and refined imaging protocols, PE remains one of the most frequently missed diagnoses in radiology.
Thank you for reading this post, don't forget to subscribe!AI pulmonary embolism detection is now providing measurable data on how that diagnostic gap can be narrowed. Six peer-reviewed studies document the clinical impact of CINA-PE, Avicenna.AI’s CE-marked and FDA-cleared chest AI tool, across four dimensions: miss rate reduction, time to notification, patient outcomes, and performance consistency across sites and scanner vendors.
The Structural Challenge of PE Detection
PE detection is not simply a matter of radiologist skill. High volumes of CT pulmonary angiography (CTPA) examinations, time pressure, variability in injection protocols, and the inherent subtlety of certain emboli, especially on scans not originally ordered for PE assessment, all create conditions where even experienced readers face a difficult task. Non-dedicated CT protocols produce images with suboptimal contrast timing; in busy departments, the cognitive load of reviewing hundreds of scans in a shift is a real and documented factor. AI does not address a skills gap. It adds a systematic, continuous layer to a workflow that was never designed to catch every finding under every condition.
AI Closes the Detection Gap - More PE Cases Identified at First Read
A real-world study of 1,204 CTPA examinations found that after AI implementation, the rate of PE-positive cases not reported at first read fell from 15.6% to 3.8%, with detection improving from 84.4% to 96.2%. The algorithm flagged 76% of cases that had not been identified at first read. Per-case sensitivity reached 93.9% (95% CI: 89.3%-96.9%) and specificity 94.8% (95% CI: 93.3%-96.1%). Known limitations include reduced sensitivity for emboli at the limit of subsegmental arterial resolution and for chronic PE, cases that represent a minority of the residual undetected rate and are documented transparently in the publication. The clinical implication: AI operating as a systematic second-pass review ensures more PE cases are identified before leaving the radiology department.
AI Reduces Time to Notification - From 318 to 5.47 Minutes
One of the most striking findings in the CINA-PE literature concerns the time from scan acquisition to specialist notification. A retrospective study at a US tertiary referral centre found that time-to-consult, from scan initiation to specialist consultation, fell from 241 minutes before AI implementation to 6.7 minutes after. A follow-up study at the same institution confirmed that scan-to-alert time averaged 5.47 minutes post-AI, compared to 318 minutes before implementation. The reduction was most pronounced in cases requiring PERT (Pulmonary Embolism Response Team) activation, precisely the highest-acuity patients, allowing multidisciplinary teams to mobilise earlier, compress treatment timelines, and make time-sensitive decisions with greater confidence.
AI Reduces In-Hospital Mortality - From 8.4% to 2.2%
A single-centre observational study comparing 113 pre-AI patients with 45 post-AI patients found that in-hospital mortality decreased from 8.4% before deployment to 2.2% after. This is a retrospective study conducted at a single centre, a limitation the authors acknowledge, and the cohorts differ in size. What the data demonstrates is a clinically meaningful shift in outcomes following the introduction of AI-assisted triage, in a real-world setting, with consecutive patients.
AI Improves Performance Across the Radiology Team
A reader study on 207 CTPA examinations compared the diagnostic performance of radiology residents with and without AI assistance. With AI support, residents achieved a sensitivity of 92.5%, compared to 81.7% without; overall accuracy improved from 95% to 98%. AI assistance helps less experienced readers achieve more consistent performance, providing an additional layer of assurance that benefits the full team, not only senior radiologists.
Consistent Performance Across Scanners and Sites
A consistent question when evaluating AI algorithms is generalisability: does performance hold across different scanner types, manufacturers, and imaging protocols?
Two independent validation studies address this directly.
- A multicentre validation on 387 real-world CTPA examinations reported sensitivity of 91.4% (95% CI: 86.4%-95.0%) and specificity of 91.5% (95% CI: 86.8%-95.0%).
- A second validation study, across 5 CT vendors and 42 different scanner models on 396 examinations, confirmed sensitivity of 91.1% (95% CI: 86.1%-94.7%) and specificity of 91.8% (95% CI: 87.1%-95.1%).
Consistent performance across scanner generations and manufacturers is essential for deployment in heterogeneous hospital environments and supports use of CINA-PE in enterprise settings where multiple CT platforms coexist.
AI and the Radiologist: A Validated Workflow
Every alert generated by CINA-PE is reviewed and validated by a radiologist before any clinical decision is made. The AI flags cases for priority attention; the radiologist interprets the finding, determines clinical significance, and issues the report. No alert reaches the clinical team without passing through that human review step.
CINA-PE is an FDA-cleared, CE-marked Class IIb decision-support tool. It contributes most effectively in clinical practice as a systematic, continuous first-pass layer that ensures no CT leaves the queue without a priority assessment, while keeping clinical judgement where it belongs.
How CINA-PE Integrates Into Existing Workflows
CINA-PE is delivered via the AVI Platform or third-party platform, Avicenna.AI’s integration layer, which connects directly to PACS and RIS systems without requiring additional viewers, worklists, or workflow modifications. Results are returned as DICOM secondary captures within the existing radiology system and/or worklist priozitization. No additional training is required for the radiology team. The algorithm processes chest and thoracoabdominal CTA acquisitions and is validated across multi-vendor environments.
CINA-PE holds CE Mark under MDR Class IIb for Europe and FDA clearance for the United States.
FAQ
What types of CT scans does CINA-PE analyse?
CINA-PE is designed for chest and thoracoabdominal CT angiography with contrast injection (dedicated CTPA protocol). For non-dedicated CT examinations, scans performed for another indication that include part of the thoracic region, Avicenna.AI offers CINA-iPE, a separate algorithm validated specifically for incidental PE detection.
How is CINA-PE regulated in Europe and the United States?
CINA-PE is CE-marked under MDR Class IIb, covering deployment across the European Economic Area. It holds FDA 510(k) clearance K210237 in the United States. Both clearances classify it as a Class II medical device intended to assist clinicians in workflow triage by flagging and communicating suspected positive PE findings.
Does CINA-PE generate the radiology report?
No. CINA-PE flags suspected positive examinations and delivers priority alerts through the existing PACS system. Every alert is reviewed and validated by a radiologist, who issues the clinical report. The AI supports the workflow; the radiologist drives the clinical decision.