Machine learning-based tri-stage classification of Alzheimer's progressive neurodegenerative disease using PCA and mRMR administered textural, orientational, and spatial features

Alzheimer's disease is a subject of substantial concern with ample scope for novel discoveries in the field of modern computational medical study for its reputation of being one of the most exorbitant and life-threatening neurodegenerative diseases of the current age. The prime objective of this research is to develop a system that can automatically detect three stages of Alzheimer's disease-Alzheimer's dementia, mild cognitive impairment, and cognitively normal using the traditional machine learning approaches. The dataset collected from Alzheimer's Disease Neuroimaging Initiative containing three types of data as mentioned above with labeled images is used throughout the research. In the proposed method, contrast limited adaptive histogram equalization handles the qualitative visual distortion in advance of feature calculation. Three distinct types of features are identified from structural MR images such as textural, orientational, and spatial features as the gray-level co-occurrence matrix, histogram of oriented gradients, and vector of locally aggregated descriptors. Apart from this, principal component analysis and minimum redundancy maximum relevance operate on the generated feature set for dimensionality reduction and to confer a comparative perspective as well. Experiments conducted upon the availed dataset exhibit that the proposed methodology outperforms other noteworthy existing methods for multiclass detection of Alzheimer's disease achieving accuracy ranging from 94% to 97% with respect to the feature set and models in action. Moreover, a significant outcome is found after applying the findings to a new independent test dataset from the same data source.