Fractional gradient optimized explainable convolutional neural network for Alzheimer's disease diagnosis

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Alzheimer's disease (AD) progressively impacts brain memory, with mild cognitive impairment (MCI) representing an early, uncertain stage.
• Early AD detection is vital for preserving cognitive function and preventing memory loss.
• Current deep learning models for AD diagnosis often lack efficiency, accuracy, and interpretability.

Purpose of the Study:

• To develop an accurate, efficient, and interpretable deep learning model for Alzheimer's disease classification.
• To address limitations in feature extraction and optimization within existing CNN architectures for AD diagnosis.
• To enhance the transparency of AI models used in medical diagnosis.

Main Methods:

• A generalized fractional order-based Convolutional Neural Network (CNN) classifier was developed.
• The model incorporates an enhanced feature extraction mechanism with an unexplored pooling technique.
• Explainable Artificial Intelligence (XAI) capabilities were integrated for model transparency.
• A fractional order-based optimization approach was employed for adaptive learning and rapid convergence.

Main Results:

• The proposed model achieved a high accuracy of 99% in classifying Alzheimer's disease using the ADNI dataset.
• It demonstrated superior performance compared to complex benchmark architectures.
• The model provides interpretable predictions, enhancing trust and understanding in diagnostic outcomes.

Conclusions:

• The fractional order-based CNN classifier offers a significant advancement in accurate and efficient Alzheimer's disease diagnosis.
• The integration of XAI ensures model transparency, crucial for clinical applications.
• This interpretable deep learning approach holds promise for early detection and management of AD.