Explainable Machine Learning Models for Alzheimer's Diagnosis Using Routine and Low-Cost Clinical Data

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's Disease (AD) is a growing global health concern, characterized by progressive neurodegeneration leading to cognitive decline.
• Current diagnostic methods face challenges with long waiting times and high costs, necessitating faster, more accessible solutions.
• There is a clear need for cost-effective and rapid diagnostic techniques for Alzheimer's Disease.

Purpose of the Study:

• To develop an explainable deep learning model for diagnosing Alzheimer's Disease (AD).
• To utilize routine, low-cost clinical data for AD diagnosis, including demographics, patient history, and automatically acquirable neuropsychological test results.
• To create classification models capable of distinguishing Alzheimer's Disease as the cause of cognitive impairment.

Main Methods:

• Analysis of a large dataset (167,364 observations, 1,024 features) from the National Alzheimer's Coordinating Center.
• Development and application of an explainable deep learning model (deep neural network).
• Implementation of a Light Gradient Boosting Machine for comparative analysis.

Main Results:

• The deep neural network achieved 90% accuracy and a 0.96 ROC-AUC for AD diagnosis.
• The Light Gradient Boosting Machine model reached 90% accuracy with a 0.97 ROC-AUC.
• Diagnostic performance was comparable, and in some aspects superior, to that of clinicians under similar data constraints.

Conclusions:

• Explainable deep learning models can effectively diagnose Alzheimer's Disease using standard clinical data.
• The developed models offer a cost-effective and efficient alternative for early AD detection.
• This approach has the potential to significantly improve diagnostic accessibility and timeliness for Alzheimer's Disease.