AD-NET: Age-adjust neural network for improved MCI to AD conversion prediction

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomarker Discovery

Background:

• Predicting Alzheimer's Disease (AD) progression in Mild Cognitive Impairment (MCI) patients is vital for timely interventions and clinical trial subject selection.
• Neuroimaging and deep learning are increasingly used for AD diagnosis and prognosis, but limited data poses a challenge.
• Transfer learning is explored to enhance deep learning models with limited medical imaging data.

Purpose of the Study:

• To propose AD-NET, a novel deep learning model for predicting MCI to AD conversion.
• To leverage transfer learning by extracting and transferring both features and age-related knowledge from pre-trained models.
• To improve the accuracy of identifying MCI patients at high risk of progressing to AD.

Main Methods:

• Developed AD-NET, an Age-adjust neural network, utilizing transfer learning.
• The pre-training model in AD-NET serves to extract/transfer features and transfer an age-related surrogate biomarker knowledge.
• Evaluated AD-NET against 8 existing classification models using a public neuroimaging dataset.

Main Results:

• AD-NET demonstrated superior performance in predicting MCI patients at risk of converting to AD.
• The proposed model outperformed all 8 competing classification models from the literature.
• Transferring age-related knowledge significantly enhanced predictive accuracy.

Conclusions:

• AD-NET offers a promising advancement in predicting Alzheimer's Disease progression from Mild Cognitive Impairment.
• The model's ability to transfer age-related knowledge represents a key innovation.
• This approach holds potential for improving early diagnosis and patient stratification in AD research.