An artificial intelligence system identified more thoracic aortic aneurysms during routine breast magnetic resonance imaging exams compared with standard clinical readings, according to a multicenter analysis of over 5,000 examinations.
Investigators retrospectively evaluated 5,057 breast magnetic resonance imaging (MRI) scans from three independent data sets: 3,232 from Erlangen University Hospital, 922 from Duke University, and 903 from the EA1141 trial. The artificial neural network (ANN) analyzed these scans using fully automated segmentation and measurement techniques.
The ANN increased detection rates by 3.5 times compared with clinical radiologist reports. In the Erlangen data set, radiologists identified aneurysms in 0.09% of patients (n = 2/2,258), while the ANN identified aneurysms in 0.31% (n = 7/2,258). Across all patients, the ANN detected aneurysms in 0.30% (n = 11/3,660).
Established diameter thresholds were used to classify aneurysms: 4.5 cm or greater for the ascending aorta and 3.5 cm or greater for the descending aorta. Among ANN-flagged cases, ascending aneurysms measured 45 mm to 59 mm, and descending aneurysms ranged from 35 mm to 37 mm.
The investigators compared the ANN's measurements with manual assessments by expert radiologists. The ANN’s diameter measurements deviated by an average of 0.524 mm from radiologists’ readings, with a mean inter-reader variation of 0.3 mm. Across test sites, the ANN achieved Dice coefficients ranging from 0.88 to 0.91 and centerline Dice values from 0.97 to 0.99.
The system demonstrated a 100% sensitivity compared with routine reporting, correctly identifying all cases previously detected by radiologists.
Risk stratification applied to patients with ascending aortic diameters ≥ 3.5 cm identified 38 cases as elevated risk by Aortic Height Index, predicting a 7% annual risk of complications. Just 2 cases met elevated risk thresholds using Aortic Size Index, with an 8% predicted annual risk.
Female patients with breast cancer or a history of breast cancer had slightly larger ascending aortic diameters and a higher risk of aneurysm presence. In the Erlangen data set, average ascending aortic diameter was 3.04 cm among female patients with breast cancer and 2.96 cm among those without the disease. The odds ratio for aneurysm presence was higher among female patients with a history of breast cancer, though confidence intervals were wide as a result of the small sample size.
The ANN performed consistently across MRI vendors, field strengths from 0.55 to 3 Tesla, and multiple postcontrast imaging timepoints. Diameter measurements remained stable across follow-up scans, with 85.5% of paired examinations showing differences within 2 mm.
The artificial intelligence system analyzed existing breast MRI data without requiring changes to scanning protocols or additional imaging sessions, offering potential for secondary aneurysm screening during routine breast imaging.
The authors declared no competing interests.
Source: Nature Communications