Introduction AI is increasingly used in clinical practice to support radiologists when reading imaging studies.

Aim of Study To evaluate the impact of an, AI based software trained to detect cerebral aneurysms on TOF-MRA on the diagnostic performance of multiple readers with different amounts of experience in diagnostic neuroimaging.

Methods 186 MRI studies were evaluated by six readers (three medical students, one radiology resident, one radiologist and one neuroradiologist) for the presence of cerebral aneurysms. First, the reading was done with the support of the software. After six weeks, the reading was repeated without the support of the software. The results were compared to the consensus reading of two neuroradiological specialists. Sensitivity (patient level and aneurysm level), specificity (patient level), and false positives/case were calculated.

Results Sensitivities (aneurysm level) ranged from 66.7%-87.0% with and 57.7%-87.0% without AI, sensitivities (patient level) were 63.4%-81.8% with and 52.3%-75.0% without AI. Specificities ranged from 93.7%-97.2% with and 89.4%-98.6% without AI. False positive findings/case ranged from 0.03–0.12 with and 0.02–0.17 without AI (differences not statistically significant, p-values 0.05–1). Four readers showed a significant decrease of reading times with the software, the remaining two readers showed a significant increase of reading times.

Conclusion We found equivocal results for the diagnostic performance of six different readers for the detection of cerebral aneurysms with and without the use of an AI software. Although we found a tendency towards better diagnostic performances, these differences were not statistically significant. The majority of readers showed a significant decrease of reading times.

Disclosure of Interest Nothing to disclose