CDR-Net: A computerized framework to detect Alzheimer's diseases and mild cognitive impairment

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) and mild cognitive impairment (MCI) are prevalent neurodegenerative disorders.
• Current diagnostic methods for AD and MCI, such as CT, PET, MRI, and MMSE, are often expensive and time-consuming.
• Traditional machine learning classifiers have limitations in accurately diagnosing AD and MCI due to shallow architectures.

Purpose of the Study:

• To propose an advanced framework utilizing electroencephalography (EEG) for improved multiclass diagnosis of MCI, AD, and healthy subjects (HSs).
• To develop a novel deep learning architecture, the Cognitive Decline Recognition Network (CDR-Net), for enhanced diagnostic performance.

Main Methods:

• Acquisition and preprocessing of EEG data, including down-sampling, noise cleaning, and segmentation.
• Feature extraction and classification using the proposed CDR-Net architecture.
• Performance assessment through cross-validation techniques like 10-fold and leave-one-out cross-validation to ensure stability and prevent overfitting.

Main Results:

• The CDR-Net architecture achieved high multiclass diagnostic performance.
• Achieved accuracy of 99.25%, sensitivity of 99.13%, and specificity of 99.32% for diagnosing MCI, AD, and HSs.
• Cross-validation confirmed the framework's stability, consistency, and robustness against overfitting and underfitting.

Conclusions:

• The proposed EEG-based framework and CDR-Net offer a highly accurate, efficient, and affordable method for diagnosing cognitive decline.
• This approach provides a strong foundation for developing future systems for detecting various brain disorders.
• EEG presents a viable, non-invasive alternative to traditional neuroimaging techniques for early detection of neurodegenerative diseases.