Three-round learning strategy based on 3D deep convolutional GANs for Alzheimer's disease staging.

Area of Science:

• Neuroimaging and Machine Learning
• Artificial Intelligence in Medical Diagnosis

Background:

• Accurate diagnosis of Alzheimer's disease (AD) and its early stages is crucial for timely intervention.
• Convolutional Neural Networks (CNNs) show promise for structural MRI (sMRI)-based AD diagnosis but are limited by small labeled datasets.
• Overfitting is a significant challenge in developing diagnostic models due to insufficient training data.

Purpose of the Study:

• To develop an effective strategy for diagnosing Alzheimer's disease (AD) and mild cognitive impairment (MCI) using structural MRI (sMRI).
• To overcome the limitations of small labeled datasets in deep learning models for neuroimaging analysis.
• To enhance the interpretability of diagnostic models by highlighting relevant brain regions.

Main Methods:

• A three-round learning strategy combining transfer learning and generative adversarial learning was proposed.
• A 3D Deep Convolutional Generative Adversarial Network (DCGAN) was trained unsupervisedly on sMRI data.
• Transfer learning was employed to fine-tune the model for AD vs. Cognitively Normal (CN) and subsequently for MCI diagnosis, with interpretability enhanced by 3D Grad-CAM.

Main Results:

• The model achieved high accuracies: 92.8% for AD vs. CN, 78.1% for AD vs. MCI, and 76.4% for MCI vs. CN.
• The proposed method effectively mitigated overfitting issues caused by limited sMRI data.
• 3D Grad-CAM highlighted brain regions crucial for accurate predictions, enhancing model interpretability.

Conclusions:

• The novel three-round learning strategy successfully enables early detection of Alzheimer's disease (AD) and mild cognitive impairment (MCI).
• This approach demonstrates robustness against data scarcity in neuroimaging-based diagnostic models.
• The combination of generative adversarial learning and transfer learning offers a promising avenue for improving AI-driven medical diagnosis.