Segmentation-Guided Development of Visual Classification Criteria for Alzheimer's Disease

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) causes brain changes years before symptoms, making early detection via MRI challenging.
• Visual MRI interpretation alone is insufficient for early AD detection.
• Automated segmentation offers precise measurements but is underutilized in clinical assessment.

Purpose of the Study:

• To investigate if automated semantic segmentation outputs can guide non-experts in creating diagnostic criteria for AD.
• To develop and test a segmentation-informed workflow for novice investigators.

Main Methods:

• A novice investigator used the MAPER segmentation model on Alzheimer's Disease Neuroimaging Initiative (ADNI) MRI data.
• Classification rules were developed based on visual assessment and volumetric readouts of the piriform cortex (PC).
• A blinded assessment on independent ADNI data was performed using the developed criteria.

Main Results:

• A classification rule based on PC volume (threshold < 430 mm³), shape, and global atrophy achieved 71% accuracy across four diagnostic groups.
• Accuracy improved to 77% when excluding mild cognitive impairment (MCI) cases, focusing on CN and AD.
• Segmentation-guided visual workflows enabled moderate accuracy in non-expert diagnosis.

Conclusions:

• Automated segmentation can empower non-experts to apply anatomically grounded diagnostic criteria for Alzheimer's disease.
• This framework supports interpretable models, explainable AI, and accelerates diagnostic skill acquisition.
• The approach shows promise for broader application in other brain regions and diagnostic tasks.