AI-Based Dementia Prediction From Low-Dose Brain CT in Mild Cognitive Impairments.

BACKGROUND AND PURPOSE: Magnetic resonance imaging (MRI) is often limited by safety concerns and long acquisition times in elderly patients. We evaluated the potential clinical utility of an artificial intelligence (AI) model, Morph Computed Tomography (CT) Plus, to predict three-year dementia conversion in mild cognitive impairment (MCI) using rapid low-dose CT scans, prioritizing its role as a safe and efficient triage tool.

METHODS

We analyzed 896 MCI (Clinical Dementia Rating [CDR]=0.5) patients from the Korea-Registries to Overcome Dementia and Accelerate Dementia Research cohort. A deep learning model quantified regional brain atrophy. We implemented a dual-cutoff risk-stratification strategy to optimize clinical workflow, comparing 90% and 95% sensitivity/specificity operating targets.

RESULTS

The model achieved an area under the curve of 0.870. The 90% strategy was superior for clinical triage, limiting the intermediate-risk group to 27.3% (n=99) compared to 49.6% (n=180) in the 95% strategy. For the combined high- and low-risk groups, the 90% strategy yielded an overall accuracy of 87.5%, with a positive predictive value of 83.3% and a negative predictive value of 90.4%. High-risk patients demonstrated a 14.13-fold higher hazard of conversion (p<0.001).

CONCLUSIONS

AI-driven CT analysis provides a high-precision, inclusive triage tool. The 90% strategy offers a scalable solution for early dementia management, especially for patients with MRI contraindications, demonstrating significant potential clinical utility.