Deep learning for Alzheimer's disease: advances in classification, segmentation, subtyping, and explainability.

Area of Science:

• Neuroimaging and computational neuroscience
• Artificial intelligence in medicine
• Biomarker discovery for neurodegenerative diseases

Background:

• Alzheimer's disease (AD) diagnosis and prognosis remain challenging, necessitating advanced analytical tools.
• Deep learning (DL) shows promise in identifying subtle imaging and non-imaging biomarkers for AD.
• Translating DL advancements into clinical practice requires a structured framework.

Purpose of the Study:

• To review current deep learning (DL) methodologies for Alzheimer's disease (AD) analysis.
• To categorize DL architectures and applications in AD detection, prognostication, and subtyping.
• To address the clinical translation gap by examining explainable AI and validation strategies.

Main Methods:

• Survey of input modalities (MRI, PET, genetics, cognitive tests) and public data cohorts.
• Categorization of DL architectures: classification, multimodal fusion, and segmentation.
• Examination of subtyping algorithms (clustering, decision trees) and explainable AI (XAI) methods.

Main Results:

• DL architectures effectively perform AD diagnosis, integrate multimodal biomarkers, and delineate brain structures.
• Subtyping algorithms reveal latent AD phenotypes, enhancing personalized medicine approaches.
• Explainable AI methods are crucial for transparent and trustworthy DL model deployment in clinical settings.

Conclusions:

• A comprehensive framework integrating diverse DL approaches is essential for clinical AD management.
• Addressing data heterogeneity, interpretability, and privacy is key for robust DL tool development.
• Interdisciplinary collaboration and technical innovation are vital for scalable DL solutions in Alzheimer's care.