Subtyping and staging of Alzheimer's disease from routine structural MRI with PHASE-AD

Area of Science:

• Neuroimaging
• Neurology
• Biostatistics

Background:

• Structural MRI is standard in Alzheimer's disease (AD) clinical assessment.
• Quantitative morphometric indices from MRI are underutilized in clinical settings.
• Existing methods lack robust classification of atrophy subtypes and progression.

Purpose of the Study:

• To present PHASE-AD, a framework for classifying AD atrophy subtypes and stages from structural MRI.
• To translate quantitative MRI indices into clinically interpretable classifications.
• To improve AD diagnosis, prognosis, and clinical trial stratification.

Main Methods:

• Developed PHASE-AD, a machine learning framework trained on 8,415 participants' MRI scans.
• Utilized cross-validation across diverse scanner configurations and field strengths.
• Validated atrophy classifications against clinical profiles, tau PET, and longitudinal cognitive data.

Main Results:

• PHASE-AD robustly identified limbic-predominant and hippocampal-sparing atrophy subtypes across datasets.
• Classifications showed high robustness across different MRI acquisition parameters.
• Atrophy subtypes correlated with distinct clinical and tau pathology profiles.
• PHASE-AD stratified longitudinal cognitive trajectories and outperformed visual MRI assessments in prospective trial simulations.

Conclusions:

• PHASE-AD effectively translates quantitative MRI into clinically relevant atrophy classifications.
• The framework aids in understanding AD heterogeneity and progression.
• PHASE-AD supports integration into clinical practice and AD clinical trials for improved patient stratification and outcome prediction.