Künstliche Intelligenz zur Unterstützung der Detektion inzidenteller Aneurysmen in routine-MRT Untersuchungen

Intracranial aneurysms pose a significant risk of rupture, leading to severe disability, and non-invasive identification can facilitate prompt treatment. This study aimed to evaluate the diagnostic efficacy of integrating commercially available AI tools with conventional radiological techniques for detecting intracranial aneurysms by an expert radiologist, and to determine whether AI assistance improves the radiologist's precision in spotting aneurysms while decreasing the time spent on image analysis.

A total of 500 Time-of-Flight (TOF) Magnetic Resonance Angiography (MRA) brain scans were reviewed using commercially available AI software to detect intracranial aneurysms. These findings were contrasted with a reference standard, which is a consensus review by two seasoned neuroradiologists, to assess the sensitivity and specificity of the AI system. Additionally, we analyzed the duration needed for a radiologist to evaluate the TOF MRA images, with and without the AI tool.

Among the 500 TOF MRA brain scans examined, 106 aneurysms were identified in 85 scans. The radiologist detected 98 aneurysms, resulting in a sensitivity of 92.5%, while the AI tool found 77 aneurysms, with a sensitivity of 72.6%. The incorporation of AI into the image review process markedly decreased the average evaluation time for each TOF MRA scan, saving an average of about 19 seconds per scan.

The results suggest that AI-assisted software can aid radiologists in interpreting brain TOF MRAs. A collaborative approach, using both the AI software and expert radiologist assessments, showed greater accuracy in identifying intracranial aneurysms compared to evaluations done solely by radiologists or the software. The utilization of deep learning algorithms significantly reduced the time needed to analyze MRI data for detecting intracranial aneurysms, thereby enhancing the radiologists' efficiency in time management.