Machine Learning-Derived MRI-Based Neurodegeneration Biomarker for Alzheimer's Disease: A Multi-Database Validation...

Area of Science:

• Neuroimaging
• Machine Learning in Medicine
• Biomarker Development

Background:

• Alzheimer's disease (AD) diagnosis relies on identifying neurodegeneration.
• Machine learning-derived biomarkers offer novel approaches to AD detection.
• The Alzheimer's disease resemblance atrophy index (AD-RAI) is an MRI-based neuroimaging biomarker indicating neurodegeneration.

Purpose of the Study:

• To validate and compare the diagnostic performance of AD-RAI against conventional volumetric hippocampal measures (hippocampal volume and fraction) for diagnosing mild AD dementia.
• To investigate the association of imaging biomarkers with age and gender in cognitively unimpaired individuals.
• To assess the impact of age and gender adjustment on the diagnostic performance of these biomarkers.

Main Methods:

• Retrospective analysis of participants with mild AD dementia (n=218) and cognitively unimpaired (CU) individuals (n=1,060) from four databases.
• Comparison of Area Under the Curve (AUC), sensitivity, specificity, and balanced accuracy for AD-RAI, hippocampal volume (HV), and hippocampal fraction (HF).
• Subgroup analyses including amyloid-negative CU participants and assessment of age/gender adjustment effects.

Main Results:

• AD-RAI demonstrated a significantly higher AUC (0.93) compared to HV (0.89) and HF (0.89) in differentiating mild AD dementia from CU.
• In subgroup analysis of amyloid-positive AD and amyloid-negative CU participants, AD-RAI's AUC (0.97) remained superior to HV (0.94) and HF (0.93).
• The diagnostic performance of AD-RAI and HF was not affected by age or gender, while HV performance improved after age adjustment.

Conclusions:

• AD-RAI exhibits excellent clinical validity for diagnosing mild Alzheimer's disease dementia.
• AD-RAI outperforms conventional volumetric hippocampal measures in diagnostic accuracy.
• AD-RAI does not require age adjustment, simplifying its clinical application for early AD detection.