Swiss Congress of Radiology (2022)
Klail T., Radojewski P.
Small incidental aneurysms represent a clinical challenge, since it is often difficult to distinguish them other structures (e.g. infundibulary branch) using standard field strengths (1.5/3T MRI). The diagnostic gold standard in form of digital subtraction angiography (DSA) is associated with radiation exposure and potential co-morbidities. On the other hand, tools based on artificial intelligence (AI) are becoming more accessible to support the radiologist in the assessment of MRI images. How such assistants perform in the most challenging cases remains unclear. Our first goal was to compare the evaluation of standard (1.5/3T MRI) images performed by the AI-assistant vs. radiological assessment and then compare these findings with a consensus reading performed with 7T-MRI.
Materials & Methods
Fifty patients with a suspected small-sized aneurysm in the standard 1.5/3T MRI examination (n=9/41) received 7T MRI examination for clarification based on the decision of the aneurysm-board. The 1.5/3T TOF-images were analyzed with the mdbrain software (mediaire, CE-certified and authorized for routine diagnostic use) for aneurysm detection. The sensitivity, specificity and accuracy of the software vs. radiological report on the 1.5/3T MRI were compared. Subsequently a comparison with the 7T MRI evaluation done by the consensus and the aneurysm-board was performed.
Twenty-one (42%) patients had an aneurysm confirmed by the 7T MRI (mean age = 53 years; m/f = 17/33). The sensitivity for aneurysm detection was higher in radiological assessment than in AI-based assessment (86% vs. 62%). The specificity was lower in radiological assessment than in AI (14% vs. 86%). Half of the evaluations done by physicians were false negative in this highly selected population. The accuracy was higher for the AI-based than for the radiological assessment (76% vs. 44%).
The use of AI in the diagnosis of small aneurysms as a support tool for physicians is promising, as it provides valuable support in image evaluation.