MTC-MSFFNet: a multi-task classification model based on multi-source feature fusion for Alzheimer's disease

Area of Science:

• Neuroimaging
• Medical Diagnostics
• Artificial Intelligence

Background:

• Accurate prediction of Alzheimer's disease (AD) progression is vital for timely intervention.
• Distinguishing between stable mild cognitive impairment (sMCI) and progressive mild cognitive impairment (pMCI) is critical for personalized treatment.
• Current classification models often overlook the nuances within mild cognitive impairment (MCI).

Purpose of the Study:

• To develop a multi-task classification model for accurate diagnosis of Alzheimer's disease (AD) stages.
• To differentiate between cognitively normal (CN), mild cognitive impairment (MCI), and AD.
• To further classify MCI into stable (sMCI) and progressive (pMCI) subtypes.

Main Methods:

• Proposed MTC-MSFFNet, a multi-task classification model utilizing multi-source feature fusion.
• Integrated brain structure maps (hippocampus, entorhinal cortex, gray matter) with structural magnetic resonance imaging (sMRI) data.
• Employed task-specific weight learning and dedicated classification heads for each diagnostic objective.

Main Results:

• Achieved 98.09% average accuracy for classifying cognitively normal (CN), mild cognitive impairment (MCI), and Alzheimer's disease (AD).
• Attained 95.16% average accuracy for differentiating between stable mild cognitive impairment (sMCI) and progressive mild cognitive impairment (pMCI).
• Validated on a combined dataset from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and Open Access Series of Imaging Studies (OASIS).

Conclusions:

• The MTC-MSFFNet demonstrates high efficacy in diagnosing AD and its subtypes.
• The model shows significant potential for assisting clinicians in creating personalized treatment strategies.
• Advanced neuroimaging analysis can improve the accuracy of Alzheimer's disease staging.