DAMNet: Dynamic mobile architectures for Alzheimer's disease

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Alzheimer's disease (AD) diagnosis requires precise and early detection methods.
• Current diagnostic tools face challenges in accuracy and efficiency.
• Advanced computational models are needed to analyze complex medical data.

Purpose of the Study:

• To develop and evaluate DAMNet, an efficient AI model for early Alzheimer's disease detection.
• To assess DAMNet's performance on multi-dimensional 2D and 3D MRI data.
• To introduce a novel parallel intelligent framework for enhanced medical diagnostics.

Main Methods:

• Developed DAMNet, a neural network model with 7.4 million parameters.
• Applied a 20% pruning rate to assess model robustness.
• Utilized 3D MRI data for validation and testing, including three-fold cross-validation.
• Implemented a parallel intelligent framework for feature extraction and data management.

Main Results:

• DAMNet achieved 98.3% accuracy in validation and 99.9% in testing.
• Model performance remained consistent with <0.2% accuracy loss after pruning.
• Achieved a 95.7% F1 score on 3D MRI data within 805s.
• Demonstrated effective handling of both 2D and 3D imaging data.

Conclusions:

• DAMNet offers a highly accurate and efficient AI solution for early Alzheimer's disease detection.
• The model's performance is robust to parameter pruning.
• The novel parallel intelligent framework advances intelligent medical diagnostics for neurodegenerative diseases.