Identification of Conversion from Normal Elderly Cognition to Alzheimer's Disease using Multimodal Support Vector...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder affecting the elderly.
• Early identification of conversion from normal cognition to mild cognitive impairment (MCI) or AD is critical for timely intervention.
• Current diagnostic methods may not capture subtle early changes effectively.

Purpose of the Study:

• To develop and validate a multimodal machine learning model for early detection of cognitive decline.
• To differentiate between normal elderly controls (NC) who convert to MCI/AD and those who do not.
• To assess the efficacy of combining magnetic resonance imaging (MRI) and positron emission tomography (PET) data.

Main Methods:

• Utilized a multimodal support vector machine (SVM) approach.
• Employed data from two independent cohorts (training and testing sets) from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
• Included MRI and PET imaging data from 121 AD patients, 120 NC, 20 NC converters, and 20 NC non-converters.

Main Results:

• Multimodal classification achieved 67.5% accuracy, 73.33% sensitivity, and 64% specificity in distinguishing NC converters from non-converters.
• Feature selection improved classification performance to 70% accuracy, 75% sensitivity, and 66.67% specificity.
• Multimodal data significantly outperformed single-modality approaches for predicting NC to MCI/AD conversion.

Conclusions:

• A multimodal SVM model effectively identifies elderly individuals at risk of developing MCI or AD.
• Combining MRI and PET imaging enhances the accuracy of early cognitive decline detection.
• This approach holds promise for improving clinical diagnosis and guiding pathological research in Alzheimer's disease.