Multi-modal sequence learning for Alzheimer's disease progression prediction with incomplete variable-length...

Area of Science:

• Neuroscience
• Computational Biology
• Medical Informatics

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder with a lengthy preclinical phase.
• Accurate prediction of AD progression is vital for timely diagnosis and patient management.
• Existing prediction models often fail to accommodate variable patient data, including missing information and irregular visit schedules.

Purpose of the Study:

• To develop a flexible framework for predicting Alzheimer's disease progression using longitudinal, multi-modal data.
• To address challenges posed by modality-missing data and variable-length patient histories.
• To improve the accuracy and applicability of Alzheimer's disease progression prediction models.

Main Methods:

• A novel multi-modal sequence learning framework integrating deep latent representation and collaborative sequence learning.
• A deep multi-modality fusion module to capture complementary information from incomplete datasets.
• Collaborative training of fusion and sequence learning modules for enhanced prediction performance.

Main Results:

• The proposed framework effectively handles incomplete, variable-length longitudinal multi-modal data.
• Demonstrated superiority over existing methods in predicting Alzheimer's disease progression.
• Validated on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.

Conclusions:

• The developed deep learning framework offers a flexible and robust approach for Alzheimer's disease progression prediction.
• This method can better manage real-world clinical data complexities, including missing modalities and irregular visits.
• The findings support improved diagnostic and prognostic capabilities for Alzheimer's disease.