Multi-knowledge informed deep learning model for multi-point prediction of Alzheimer's disease progression

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) diagnosis traditionally relies on visual features and clinical knowledge.
• Accurate prediction of AD progression is crucial for timely intervention and patient management.

Purpose of the Study:

• To introduce Mul-KMPP, a novel deep learning framework for multi-point prediction of Alzheimer's disease progression.
• To enhance diagnostic accuracy and prognostic capabilities in older adults through advanced AI.

Main Methods:

• Developed a dual-path deep learning model (Mul-KMPP) for extracting global and local brain features from MRIs.
• Integrated an Anatomical Automatic Labeling (AAL) knowledge-based diagnostic module prior to the prediction module.
• Utilized a composite loss function incorporating diagnosis, prediction, and consistency losses.

Main Results:

• Mul-KMPP achieved 86.8% accuracy, 86.1% sensitivity, 92.1% specificity, and 95.9% AUC on an 819-sample dataset.
• The model significantly outperformed existing diagnostic methods across all time points.
• Demonstrated robust performance in predicting AD progression.

Conclusions:

• Mul-KMPP offers a powerful, accurate, and reliable tool for predicting Alzheimer's disease progression.
• The integration of multi-knowledge and a dual-path approach enhances diagnostic and prognostic capabilities.
• This framework holds significant potential for clinical application in managing Alzheimer's disease.