Abstract:
Classification of patients of different categories of Alzheimer's disease (AD) is a challenging task with subsequent applications in the diagnosis of AD. This task requires careful examination of results by a panel of experts which is usually cumbersome and hard to obtain and is intricate in conventional MRI images due to similar intensities of background pixels and surrounding brain structures. Manual interpretation of results is also difficult and time consuming. There is a need for an accurate and robust method of classification of initial stages of AD that uses disease non-specific features and requires very little or no intervention of a medical domain expert.
In the present research effort, an automated method for the classification of initial stages of Alzheimer's disease has been presented that extract useful discriminating features from phase images using independent component analysis (ICA) technique. Comparative analysis has been performed between proximal support vector machine, K nearest neighbor and artificial neural network classifiers to obtain accurate results and to check the efficacy of ICA features. A multiclass classification architecture based on the principle of progressive two class decision classification has been tested to classify the examples in three initial categories of AD. Moreover, ICA based features are qualitatively assessed and their correlation with different socioeconomic parameters such as age, education and mini-mental status examination (MMSE) is critically presented. Lastly, visualization of different parts of the brain is performed by making use of the independent components matrix.
Obtained classification results indicate that features obtained from the phase images using ICA technique can be useful in diagnosing early categories of AD patients with accuracy rate as high as 97%. Moreover, ICA based features essentially preserve important information which correlates strongly with socioeconomic parameters such as age, education and MMSE and essentially the affected brain parts preserved by these features have been found to be the same that are usually affected during the course of progression of AD.
