A Data Fusion Framework for Mild Cognitive Impairment Classification: Hippocampal Volume and GLCM Features Using Machine Learning

Alzheimer’s disease (AD) is a cumulative brain disorder that may initiate mild memory loss, possibly 15 to 20 years prior to the onset of overt cognitive symptoms. This underscores the importance of early detection for better outcomes. While much of the existing literature has focused on primarily AD and cognitively normal stages, the intermediate stage mild cognitive impairment has often been overlooked. To address this, we aim to classify data fusion vectors of hippocampal 2D Gray-Level Co-occurrence Matrix (2D-GLCM), along with hippocampal volume of 3D Magnetic Resonance Imaging (3D MRI) images into three primary yield classes (AD, MCI, and cognitive normal (CN) subjects). Features were extracted from the Medical Decathlon Hippocampus Dataset. Three classification algorithms were applied to evaluate the proposed fused feature model, namely, K-Nearest Neighbor (KNN), Probabilistic Neural Network (PNN), and Random Forest (RF). The proposed framework is used for multi-class (CN, AD, and MCI), and one-vs-rest/one-vs-one binary-class classifications (MCI, CN), (AD, MCI), and (MCI, (CN, AD)). Binary-class outperformed multi-class classification, achieving 97% accuracy for MCI vs. CN and 89.14% accuracy for multi-class (AD, MCI, and CN) using PNN. Our results further show that data fusion (2D-GLCM and volume) performs more accurately than single modality (2D-GLCM), highlighting its potential in enhancing MCI classification with added modalities and diagnostic biomarkers.