In a new study involving over 11,700 chest computed tomography (CT) scans in oncology patients, adjunctive artificial intelligence software demonstrated a sensitivity rate of 91.6 percent for incidental pulmonary embolism (IPE) and reduced median detection and notification time for IPE-positive scans from multiple days to one hour for a radiology department at a comprehensive cancer center.
New research suggests adjunctive artificial intelligence (AI) software can significantly improve the detection of incidental pulmonary embolism (IPE) on chest computed tomography (CT) scans and shaves days off the time it takes to arrive at the diagnosis.
For the study, recently published in Radiology: Cardiothoracic Imaging, researchers reviewed a total of 11,736 chest CT scans for 6,447 oncology patients (mean age of 63) to assess the diagnostic utility of an AI software (Aidoc) for IPE as well as its impact on worklist triage. The study authors compared IPE detection across three separate 15-week time periods: an initial period of radiologist reporting with no AI assistance (April to July 2019); a second period with no AI assistance but instruction to review all new CT scans for IPE (November 2019 to February 2020); and the third period with AI assistance (June to September 2020).
The AI software had a sensitivity rate of 91.6 percent, a specificity rate of 99.7 percent, an accuracy rate of 99.6 percent and a positive predictive value of 80.9 percent, according to the study authors.
For the IPE-positive CT scans from the first two study periods (105), the researchers noted that radiologists missed the IPE finding in 47 cases (44.8 percent) but these were correctly identified with AI. However, when radiologists had AI assistance in the third study period, they missed one out of 38 cases positive for incidental PE (2.6 percent).
“This resulted in a 94% reduction (44.8% vs. 2.6%) in missed IPE with use of AI assistance,” noted study co-author Jacob J. Visser, M.D., who is affiliated with the Department of Radiology and Nuclear Medicine at the Erasmus University Medical Center in Rotterdam, the Netherlands, and colleagues.
Images courtesy of Radiology: Cardiothoracic Imaging.
(Editor’s note: For related articles, see “Can a New AI Tool Improve Detection of Incidental Pulmonary Embolism on CT?,” “Seven Takeaways from Best Practice Recommendations for Incidental Findings in the ER” and “Can AI Improve Triage of CT Pulmonary Angiography Exams for Acute PE?”)
The researchers also evaluated the impact of the AI tool upon detection and notification time (DNT), defined as the time interval between availability of the study in a radiology worklist and opening of the study by the radiologist, in a comprehensive cancer center’s radiology department. The study authors noted the department had a backlog of unreported chest CT exams, the majority of which were for post-treatment follow-up for outpatients with a known primary malignancy.
According to the study, the median DNTs for IPE-positive chest CTs were as follows:
• 7,714 minutes (during the first study period of routine radiology workflow without AI);
• 4,973 minutes (second study period of routine radiology workflow with direction to review for IPEs without AI)); and
• 87 minutes (AI-assisted worklist prioritization).
Noting that 37.8 percent of the IPE-positive scans in the study revealed emboli in main or lobal pulmonary arteries, Visser and colleagues said AI-based worklist prioritization tools can facilitate timely diagnosis and care for high-risk patients in busy tertiary care environments.
“Tools to prioritize the reading worklist would provide the most benefit in clinical settings with a high workload and a backlog of unreported examinations,” emphasized Visser and colleagues.
In regard to study limitations, the authors noted they did not review CT exams that were deemed negative by radiologists and the AI software, and this may have contributed to an underestimation of false-negative IPEs. The researchers acknowledged the study was not randomized and focused primarily on diagnostic efficacy. The study authors suggested that future research could examine patient outcomes related to early diagnosis of IPE.
Meta-Analysis Shows Merits of AI with CTA Detection of Coronary Artery Stenosis and Calcified Plaque
April 16th 2025Artificial intelligence demonstrated higher AUC, sensitivity, and specificity than radiologists for detecting coronary artery stenosis > 50 percent on computed tomography angiography (CTA), according to a new 17-study meta-analysis.
New bpMRI Study Suggests AI Offers Comparable Results to Radiologists for PCa Detection
April 15th 2025Demonstrating no significant difference with radiologist detection of clinically significant prostate cancer (csPCa), a biparametric MRI-based AI model provided an 88.4 percent sensitivity rate in a recent study.
Can CT-Based AI Radiomics Enhance Prediction of Recurrence-Free Survival for Non-Metastatic ccRCC?
April 14th 2025In comparison to a model based on clinicopathological risk factors, a CT radiomics-based machine learning model offered greater than a 10 percent higher AUC for predicting five-year recurrence-free survival in patients with non-metastatic clear cell renal cell carcinoma (ccRCC).
Could Lymph Node Distribution Patterns on CT Improve Staging for Colon Cancer?
April 11th 2025For patients with microsatellite instability-high colon cancer, distribution-based clinical lymph node staging (dCN) with computed tomography (CT) offered nearly double the accuracy rate of clinical lymph node staging in a recent study.