Biomarkers.

Area of Science:

• Neuroimaging analysis
• Machine learning in medicine
• Biostatistics

Background:

• Neurodegenerative diseases like Alzheimer's disease (AD) exhibit significant heterogeneity in progression.
• Accurate diagnosis and treatment are challenged by this variability.
• Identifying distinct patient subtypes is crucial for understanding disease mechanisms and improving precision medicine.

Purpose of the Study:

• To develop a simultaneous clustering and classification method for Alzheimer's disease.
• To classify subjects as cognitively unimpaired (CU) or AD.
• To identify distinct biological subtypes within the AD population.

Main Methods:

• Utilized T1-weighted MRI data from 2,374 participants in the Alzheimer's Disease Neuroimaging Initiative (ADNI).
• Processed images to extract 120 Region-of-Interest (ROI) gray matter density values using the Clinica toolbox.
• Applied a mixture-of-experts (MOE) framework with a clustering module for simultaneous classification and subtype identification.

Main Results:

• The MOE model achieved high classification accuracy, outperforming a single SVM classifier.
• Identified two distinct subtypes among AD subjects.
• Both subtypes showed significant hippocampal atrophy, with subtype 2 exhibiting more severe atrophy, indicative of later-stage AD.

Conclusions:

• The MOE model effectively classifies Alzheimer's disease and characterizes its heterogeneity.
• The method provides interpretable subtype information, enhancing diagnostic capabilities.
• This approach offers valuable insights for automatic and interpretable diagnosis of AD and other neurodegenerative diseases.