RLAD: A Reliable Hippo-Guided Multi-Task Model for Alzheimer's Disease Diagnosis.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Early diagnosis of Alzheimer's disease (AD) is critical for effective intervention.
• Hippocampal atrophy is a key indicator in early AD diagnosis.
• Current deep learning (DL) methods for AD diagnosis often treat classification and hippocampus segmentation as separate tasks, missing crucial correlations and pathological interpretability.

Purpose of the Study:

• To propose a novel Reliable Hippo-guided Learning model for Alzheimer's Disease diagnosis (RLAD).
• To integrate AD classification and hippocampus segmentation using multi-task learning.
• To enhance pathological interpretability and diagnostic accuracy by correlating classification with hippocampal features.

Main Methods:

• Developed a hybrid shared features encoder to capture both local and global information from MRI scans.
• Utilized Task Specific Decoders for independent AD classification and hippocampus segmentation.
• Implemented a Task Coordination module to link the two tasks and guide classification towards the hippocampus region.

Main Results:

• Evaluated the RLAD model on 1631 subjects across three independent datasets (ADNI-1, ADNI-2, HarP).
• Demonstrated significant improvements in both AD classification and hippocampus segmentation performance.
• Showcased strong generalization capabilities of the proposed model.

Conclusions:

• The RLAD model effectively integrates multi-task learning for improved Alzheimer's disease diagnosis.
• The hippo-guided approach enhances diagnostic accuracy and provides better pathological interpretability.
• The model shows promise for reliable and generalizable early AD detection using MRI data.