Basic Science and Pathogenesis.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroimaging Analysis
• Biomarker Discovery

Background:

• Neurodegenerative disorders (ND), including Alzheimer's Dementia (AD) and Fronto-Temporal Dementia (FTD), present diagnostic challenges due to complexity and traditional method limitations.
• The Cognitive and Neuroimaging in Neurodegenerative Disorders (CogNID) study collected data from 450 patients, including cognitive, MRI, and biomarker information.
• Developing advanced AI frameworks is crucial for early and accurate detection of dementia.

Purpose of the Study:

• To create an AI-driven framework for the early detection of dementia.
• To integrate multimodal data (cognitive, imaging, clinical biomarkers) for improved diagnostic capabilities.
• To enhance the accuracy and efficiency of diagnosing neurodegenerative disorders.

Main Methods:

• A multimodal approach combining clinical data, MRI scans, and radiology reports.
• Natural Language Processing (NLP) using PubMedBERT for report analysis and BART-based classifiers for risk scoring.
• Machine learning models (ensemble methods, neural networks) trained on fused features from imaging, reports, and cognitive/CSF data.
• Model interpretability ensured using SHAP values.

Main Results:

• Successful development of the Medical Risk Analyser (MRA) for processing textual data and generating risk scores.
• Demonstrated robust model training with decreasing training and validation loss as epochs increased.
• Ongoing work focuses on integrating MRI features and enhancing classification models.

Conclusions:

• The study advances AI for early ND diagnosis by addressing data imbalance and ensuring model interpretability and scalability.
• The developed AI framework has the potential to support precise dementia diagnosis.
• This research contributes to improving patient diagnosis and further research in neurodegenerative disorders.