A Hybrid Feature Extraction Approach Using DenseNet and Local Binary Patterns for Alzheimer's Disease Classification

Alzheimer’s Disease (AD) is a prevalent neurodegenerative disorder that significantly impacts cognitive and functional abilities. Early and accurate diagnosis of AD and its associated cognitive impairments is critical for effective management and intervention. In this study, we propose a hybrid feature extraction method combining Local Binary Patterns (LBPs) and the DenseNet deep learning model to enhance the classification accuracy of AD and related cognitive conditions. The ADNI3 dataset, consisting of five distinct classes, Alzheimer's Disease (AD), Control, Early Mild Cognitive Impairment (EMCI), Late Mild Cognitive Impairment (LMCI), and Mild Cognitive Impairment (MCI), was employed in this analysis.

Images from the dataset were preprocessed by converting them to grayscale for LBP extraction and resized to 224x224 pixels for DenseNet processing. The extracted LBP and DenseNet features were concatenated to form a comprehensive feature set, which was then used to train a multi-class Support Vector Machine (SVM) classifier with Error-Correcting Output Codes (ECOCs).

The proposed method demonstrated a robust performance with an overall accuracy of 95.36%. The confusion matrix analysis revealed precision, recall, and F1 scores of 96.93%, 91.54%, and 93.96%, respectively, indicating high reliability in classifying the different stages of cognitive impairment. These findings suggest that the integration of LBP and DenseNet features provides a powerful approach for the early diagnosis and classification of Alzheimer's Disease, with potential applications in clinical settings for facilitating timely interventions and improving patient outcomes.