Multi-class Alzheimer's disease classification using image and clinical features

Alzheimer's disease (AD) is the most common form of dementia, which results in memory related issues in subjects. An accurate detection and classification of AD alongside its prodromal stage i.e., mild cognitive impairment (MCI) is of great clinical importance. In this paper, an Alzheimer detection and classification algorithm is presented. The bag of visual word approach is used to improve the effectiveness of texture based features, such as gray level co-occurrence matrix (GLCM), scale invariant feature transform, local binary pattern and histogram of gradient. The importance of clinical data provided alongside the imaging data is highlighted by incorporating clinical features with texture based features to generate a hybrid feature vector. The features are extracted from whole as well as segmented regions of magnetic resonance (MR) brain images representing grey matter, white matter and cerebrospinal fluid. The proposed algorithm is validated using the Alzheimer's disease neuro-imaging initiative dataset (ADNI), where images are classified into one of the three classes namely, AD, normal, and MCI. The proposed algorithm outperforms state-of-the-art techniques in key evaluation parameters including accuracy, sensitivity, and specificity. An accuracy of 98.4% is achieved for binary classification of AD and normal class. For multi-class classification of AD, normal and MCI, an accuracy of 79.8% is achieved.