Validation of prediction algorithm for risk estimation of intracranial aneurysm development using real-world data

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Intracranial aneurysms (IA) are challenging to detect, leading to many undiagnosed cases due to risky and costly screening methods.
• Developing a risk assessment system is crucial for efficient and safe IA screening strategies.
• A previously developed AI model predicted IA incidence risk using cohort data.

Purpose of the Study:

• To validate the clinical performance of an existing AI-driven incidence risk prediction model for estimating IA prevalence risk.
• To assess the model's utility in a real-world health screening setting using cross-sectional data.
• To explore the potential for early IA detection through risk stratification.

Main Methods:

• Utilized cross-sectional data from 5942 individuals undergoing voluntary health checkups and brain CTA at a Korean hospital (2007-2019).
• Applied a pre-existing AI model to calculate IA risk scores (0-100) for participants without prior cerebrovascular disease history.
• Performed age-sex standardization using national data to compare IA prevalence across different risk groups.

Main Results:

• The overall age-sex standardized IA prevalence was 3.20%.
• Prevalence increased significantly with risk score: 0.18% (lowest risk) to 6.44% (highest risk), with an odds ratio of 38.50 between extreme groups.
• The AI model demonstrated viable performance with an optimal cut-off risk score of 60.5 and an AUC of 0.70.

Conclusions:

• The machine learning-based incidence risk prediction model effectively predicts IA prevalence risk in a health screening context.
• This AI system can stratify individuals by risk, facilitating targeted screening and potentially enabling earlier detection of intracranial aneurysms.
• The findings support the integration of AI risk assessment tools into routine health examinations for improved IA management.