Artificial Intelligence Assistive Software Tool for Automated Detection and Quantification of Amyloid-Related Imaging...

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Amyloid-related imaging abnormalities (ARIA) are MRI findings in Alzheimer's disease (AD) patients treated with amyloid-β-directed monoclonal antibodies.
• Monitoring ARIA is crucial for adjusting treatment dosage and ensuring patient safety.
• Assistive software may enhance the accuracy of ARIA detection and interpretation.

Purpose of the Study:

• To evaluate the clinical performance of an AI-based software tool designed to assist radiologists in interpreting brain MRIs for ARIA.
• To compare the diagnostic accuracy of radiologists with and without AI software assistance in identifying ARIA.

Main Methods:

• A multiple-reader, multiple-case diagnostic study design was employed.
• 16 radiologists interpreted 199 retrospective MRI scans (baseline and follow-up) from aducanumab clinical trials.
• Radiologists' performance was assessed both with and without the AI assistive software (icobrain aria).

Main Results:

• AI-assisted reading significantly improved diagnostic accuracy for both ARIA with edema/effusion (ARIA-E) and ARIA with microhemorrhage/siderosis (ARIA-H).
• The area under the receiver operating characteristic curve (AUC) for ARIA-E detection improved by 0.05 (assisted AUC 0.87) and for ARIA-H by 0.04 (assisted AUC 0.83).
• Sensitivity for ARIA detection was significantly higher with AI assistance (87% for ARIA-E, 79% for ARIA-H) while maintaining high specificity (>80%).

Conclusions:

• The AI-based assistive software significantly enhances radiological reading performance for ARIA detection and diagnosis.
• This AI tool shows potential as a valuable clinical aid for monitoring patients with Alzheimer's disease on amyloid-β-directed therapies.
• Improved ARIA detection can contribute to safer and more effective management of Alzheimer's disease treatment.