Constructing personalized characterizations of structural brain aberrations in patients with dementia using...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning models for neurological disorder prediction using MRI show promise but lack clinical impact due to opacity.
• Understanding the decision-making process of these AI models is crucial for clinical trust and application.

Purpose of the Study:

• To develop an explainable AI model for differentiating dementia patients from healthy controls using structural brain MRI.
• To validate the model's findings against known dementia-related brain abnormalities.
• To assess the utility of explainable AI in predicting dementia transition in mild cognitive impairment patients.

Main Methods:

• Convolutional neural networks (CNNs) were trained on structural brain MRI scans.
• Layerwise relevance propagation (LRP) was used to generate individual-level explanations for model predictions.
• The explainable model was applied to a longitudinal dataset of mild cognitive impairment patients.

Main Results:

• The AI model successfully differentiated dementia patients from healthy controls.
• Model explanations aligned with established knowledge of structural brain changes in dementia.
• Spatially rich explanations from the model aided in forecasting dementia transition and characterizing disease manifestation.

Conclusions:

• Explainable AI, through CNNs and LRP, enhances the clinical utility of MRI-based deep learning for neurological disorders.
• This approach provides biologically relevant insights into individual patient conditions.
• The study highlights the potential of explainable AI in advancing precision medicine for dementia and related disorders.