Integrating Demographics and Imaging Features for Various Stages of Dementia Classification: Feed Forward Neural...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Magnetization-prepared rapid acquisition (MPRAGE) MRI is a key tool for dementia diagnosis.
• Volumetric analysis of brain MRI is crucial for classifying dementia stages.
• This study integrates volumetry, radiomics, and demographics for advanced dementia classification.

Purpose of the Study:

• To develop an artificial intelligence model for classifying dementia stages: Alzheimer's disease (AD), mild cognitive decline (MCI), and cognitive normal (CN).
• To evaluate the effectiveness of integrating brain regional volumes, radiomics, and demographic data for dementia classification.

Main Methods:

• Utilized the Alzheimer's Disease Neuroimaging Initiative (ADNI) and Open Access Series of Imaging Studies (OASIS) datasets.
• Employed Freesurfer for brain segmentation and 3D Slicer for radiomics feature extraction.
• Developed and compared artificial intelligence models, including feed-forward neural networks (FFNN), support vector machine (SVM), ensemble classifier (EC), and decision tree (DT).

Main Results:

• The integrated model achieved the highest overall accuracy: 76.57% (ADNI) and 73.14% (OASIS).
• Subclass accuracies for MCI, AD, and CN were high, with balanced sensitivity and specificity across both datasets.
• The FFNN model demonstrated strong performance in classifying different dementia stages.

Conclusions:

• The FFNN model effectively categorizes MCI, AD, and CN with good accuracy and balanced subclass performance.
• The proposed FFNN model is simple and can aid in the initial triage of patients for further diagnostic confirmation.
• This AI-driven approach offers a potential tool for supporting early dementia detection.