Development and validation of an interpretable deep learning framework for Alzheimer's disease classification.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Dementia Research

Background:

• Alzheimer's disease (AD) is the leading cause of dementia globally, posing a significant public health challenge.
• Current diagnostic methods for AD, relying on patient history, neuropsychological tests, and MRI, often lack sensitivity and specificity.
• The aging population exacerbates the need for improved diagnostic capabilities for Alzheimer's disease.

Purpose of the Study:

• To develop and validate an interpretable deep learning strategy for accurate Alzheimer's disease diagnosis.
• To delineate unique Alzheimer's disease signatures using multimodal data inputs.
• To create a clinically adaptable framework for nuanced neuroimaging-based diagnosis of Alzheimer's disease.

Main Methods:

• A deep learning framework combining a fully convolutional network and a multilayer perceptron was employed.
• The model integrated multimodal data: MRI, age, gender, and Mini-Mental State Examination scores.
• Training and validation were performed on diverse cohorts including ADNI, AIBL, Framingham Heart Study, and NACC datasets.

Main Results:

• The deep learning model demonstrated high and consistent diagnostic performance across independent validation cohorts (AUCs ranging from 0.876 to 0.996).
• The model's diagnostic accuracy exceeded that of a panel of practicing neurologists.
• Predicted high-risk cerebral regions aligned with post-mortem histopathological findings, indicating biological relevance.

Conclusions:

• The developed deep learning framework offers a clinically adaptable and accurate strategy for Alzheimer's disease diagnosis using routinely available MRI data.
• This approach provides a generalizable method for integrating deep learning with pathophysiological processes in human diseases.
• The study highlights the potential of AI in enhancing the sensitivity and specificity of Alzheimer's disease detection.