Diagnostic image analysis for unruptured cerebral aneurysms using AI is highly sensitive but requires improvement due to false positives. This study validates fine-tuning an AI algorithm for aneurysm diagnosis using 10,000 Brain Dock MRI scans. Initial diagnosis provided feedback to the algorithm. In the primary analysis, sensitivity decreased from 96.5% to 90%, while false positives improved from 2.06 to 0.99 (P < 0.001). In the secondary analysis, sensitivity dropped from 98.8% to 94.6%, and false positives improved from 1.99 to 1.03 (P < 0.001). Tuning reduced false positives with minimal sensitivity decline.
Introduction: Asymptomatic unruptured cerebral aneurysms affect 2-6% of adults, with a 0.95% annual rupture rate in Japan. Japanese individuals face a 2.8 times higher rupture risk than Westerners, emphasizing the need for early detection through Brain Dock’s MRI and MRA scans. The Brain Dock system also aids in addressing risk factors and improving lifestyle. Currently, diagnostic imaging physicians manually review images, but cloud-based AI diagnosis offers a quicker alternative.
The AI-based UCA imaging software, despite high sensitivity, yields false positives, necessitating further clinical improvement.
This study assesses the impact of fine-tuning the AI algorithm using a dataset of 10,000 Brain Dock scans.
Results: Among 10,000 images, 5,000 pre-tuning and 5,000 post-tuning were compared. Sensitivity decreased from 96.5% to 90%, while false positives improved from 2.06 to 0.99 (P < 0.001). In the secondary analysis, sensitivity decreased from 98.8% to 94.6%, and false positives improved from 1.99 to 1.03 (P < 0.001). Tuning reduced false positives with minimal sensitivity decline.
