Auto-weighted centralised multi-task learning via integrating functional and structural connectivity for subjective...

Area of Science:

• Neuroimaging
• Machine Learning
• Cognitive Neurology

Background:

• Early diagnosis of mild cognitive impairment (MCI) and subjective cognitive decline (SCD) is crucial for intervention.
• Accurate discrimination between SCD, MCI, and healthy individuals remains a significant challenge.

Purpose of the Study:

• To propose an auto-weighted centralised multi-task (AWCMT) learning framework for the differential diagnosis of SCD and MCI.
• To integrate structural and functional brain connectivity information for enhanced diagnostic accuracy.

Main Methods:

• Constructing separate functional brain networks using sparse, low-rank methods and structural brain networks via fibre bundle tracking from MRI data.
• Developing a novel multi-task learning algorithm to combine and identify features from both functional and structural connectivity.
• Implementing an auto-weighted mechanism to automatically learn the significance of each task in a balanced manner.

Main Results:

• The AWCMT framework demonstrated superior performance in classifying SCD, MCI, and healthy subjects compared to traditional algorithms.
• The method achieved high accuracy by integrating multi-modal neuroimaging data.
• The approach offers good interpretability by identifying disease-related brain regions and their connectivity.

Conclusions:

• The proposed AWCMT learning framework provides an effective approach for the differential diagnosis of early cognitive impairment.
• Integrating functional and structural brain connectivity offers valuable insights for early Alzheimer's disease detection.
• The method's interpretability aids in understanding the neurobiological underpinnings of cognitive decline.