A Tutorial on Explainable Image Classification for Dementia Stages Using Convolutional Neural Network and...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Neuroscience

Background:

• Dementia diagnosis relies on clinical assessment and neuroimaging.
• Accurate classification of dementia stages is crucial for timely intervention.
• Current deep learning models for medical image analysis often lack interpretability.

Purpose of the Study:

• To present a tutorial on an explainable AI approach for dementia stage classification.
• To utilize Convolutional Neural Networks (CNN) and Gradient-weighted Class Activation Mapping (Grad-CAM) for classifying four progressive dementia stages.
• To enhance the interpretability of AI models in neuroimaging.

Main Methods:

• Implementation of a Convolutional Neural Network (CNN) architecture for MRI brain image classification.
• Application of Gradient-weighted Class Activation Mapping (Grad-CAM) for visualizing CNN decision-making processes.
• Training and testing the model on open MRI brain image datasets for four dementia stages.

Main Results:

• The proposed CNN architecture achieved over 99% accuracy on the test dataset.
• Grad-CAM visualization provided insights into the CNN's high accuracy, highlighting relevant brain regions.
• The explainable AI approach demonstrated potential for clinical utility.

Conclusions:

• Explainable AI, combining CNN and Grad-CAM, can effectively classify dementia stages from MRI.
• Visualizations aid in understanding the 'black box' nature of CNNs in medical applications.
• This approach offers valuable information for physicians, potentially improving diagnostic confidence and patient care.