Early Detection and Classification of Alzheimer’s through Integration of 3D Local Binary Patterns and SVM classifier

The World Health Organization (WHO) statistics reveal that Alzheimer’s disease (AD) is one of the dominant neuro-degenerative diseases causing 33% disability adjusted life years (DALYs) and it is anticipated to a rise of 106 million cases globally over the next four decades. AD is characterized by the acute confusional state and memory loss that progresses gradually and leads to irreversible significant decline in socio cognitive behaviour. Hence, early detection and diagnosis of AD is crucial. Heterogeneous attempts have been made to detect AD through different imaging biomarkers to identify the brain atrophy. Among, several neuroimaging biomarkers structural MRI has been employed most frequently to distinguish populace with AD from healthy counterparts. Contemporary literature provides an extensive work on automated AD detection employing several machine and deep learning models. Despite the superior capabilities of machine or deep learning models, incorporation of non-discriminative features lead to inaccurate results. Incorporating discriminative features with machine or deep learning models enhances the diagnostic accuracy of AD classification. Although the MRI images are 3 dimensional, most of the existing literature focussed upon extraction of features from 2D slice of the MRI which eventually ignore the spatial relationship between the adjacent slices Examining features on three dimensional MRI contribute to a significant improvement in diagnostic accuracy. Hence, the present article explores the effectiveness of incorporation of LBP texture feature on 3D structural MRI for early AD detection and diagnosis through the SVM classifier. To evaluate the effectiveness of the proposed model we performed exhaustive experimental analysis on the proposed model and state-of-the-art models on ADNI1 dataset and evaluated performance in terms of precision, recall, accuracy, and confusion matrix. The experimental findings clearly show that the proposed model offers 99.01% accuracy over AD (Alzheimer’s disease) vs. CN (cognitively normal) classification. Hence, 3D LBP feature can be regarded as an effective and optimal attribute for early AD detection and classification.