Classification of Alzheimer's Disease Using Ensemble of Deep Neural Networks Trained Through Transfer Learning

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Alzheimer's disease (AD) is a leading cause of death in the elderly.
• Deep learning (DL) models show promise in detecting AD from MRI scans.
• Existing DL models often require specific architectures and extensive training.

Purpose of the Study:

• To develop a computationally efficient and DL-architecture agnostic ensemble for AD classification.
• To introduce the Deep Transfer Ensemble (DTE) method utilizing transfer learning.
• To evaluate DTE's performance on large and small datasets for AD detection.

Main Methods:

• An ensemble of deep neural networks (DTE) was trained using transfer learning.
• DTE leverages complementary feature views and diversity from randomized hyperparameters.
• The model was tested on cognitively normal (NC) vs AD and mild cognitive impairment (MCI) vs AD classification tasks.

Main Results:

• DTE achieved 99.05% and 85.27% accuracy for NC vs AD classification on two independent splits.
• DTE achieved 98.71% and 83.11% accuracy for MCI vs AD classification on two independent splits.
• The model demonstrated reasonable performance on a small dataset, reaching 85% accuracy for NC vs AD.

Conclusions:

• Deep Transfer Ensemble (DTE) offers a robust and efficient method for Alzheimer's disease detection.
• DTE outperforms other deep learning models and snapshot ensembles in AD classification.
• The proposed method shows potential for early and accurate diagnosis of neurodegenerative diseases.