Class activation attention transfer neural networks for MCI conversion prediction

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurology

Background:

• Early prediction of Alzheimer's disease (AD) progression is vital for timely intervention.
• Mild cognitive impairment (MCI) is an early stage that may precede AD.
• Distinguishing between stable MCI (sMCI) and progressive MCI (pMCI) is challenging.

Purpose of the Study:

• To develop a novel attention transfer method for predicting MCI to AD progression within 3 years.
• To improve the accuracy of early Alzheimer's disease diagnosis using neuroimaging data.
• To investigate the efficacy of attention transfer compared to traditional methods.

Main Methods:

• A 3D convolutional neural network was trained using an attention transfer approach.
• A source task was used to learn regions of interest (ROIs) automatically.
• The model was trained to simultaneously classify pMCI/sMCI and transfer attention maps from the source task.

Main Results:

• The attention transfer method significantly outperformed traditional transfer learning and expert-defined ROI methods.
• The transferred attention maps highlighted known Alzheimer's pathology regions.
• The model accurately predicted which MCI patients would progress to AD within 3 years.

Conclusions:

• Attention transfer is a promising technique for predicting Alzheimer's disease progression from MCI.
• This method offers a more effective approach than traditional transfer learning for neuroimaging analysis.
• The findings support the use of AI in early AD detection and patient stratification.