Document resource
Introduction Detection of intracranial aneurysms (IAs) remains a time-consuming and error-prone process, necessitating solutions with high sensitivity for IA localization. Recently, artificial intelligence (AI)-based automated diagnostic systems have been proposed for medical imaging.Aim of Study This study aimed to develop an automated system capable of identifying the location and measuring the maximum diameter of IAs on MRI.Method We retrospectively analyzed MRI scans from 1310 patients with or without IAs, using 937 cases for training and 373 for testing. A correct localization was defined as the AI-predicted center falling within the ground truth IA area. A deep learning model based on nnU-Net was employed, and performance was evaluated via five-fold cross-validation on the test cohort. The maximum diameter of each IA was automatically measured based on segmented dome regions.Results Among the 937 training cases, 778 patients (83%) had IAs, including 146 patients (19%) with multiple aneurysms, while 159 patients (17%) had no IA. A total of 1213 aneurysms were identified, 78% of which measured 2–5 mm. In the 373 test cases, 17 patients (4.5%) harbored IAs, reflecting a real-world prevalence. The AI system demonstrated excellent internal validation results with an AUC of 0.92, sensitivity of 88% (15/17), and a false negative rate of 0.44 aneurysms per person. The diameter prediction model achieved a mean absolute error of 1.2 mm.Conclusion We successfully developed a high-performance AI-based diagnostic model for IA localization and measurement. This system has the potential to substantially reduce the burden of manual IA detection and improve diagnostic efficiency.Conflict of Interest No