Investigation on performance enhancement methods for swarm intelligence algorithms in transformation technique based classification of dementia from mri images

Abstract

Dementia is a commonly used term for denoting the progressive decline of mental skills like memory, reasoning, learning, etc. The number of demented peoples in the world is rapidly growing every year and it will reach around 65 million in 2030. Diagnosis of dementia is crucial to decide about the treatment possibilities. Magnetic Resonance Imaging (MRI) is very helpful in the diagnosis of many diseases and it is used as an efficient tool for examining the tissue pattern of the human brain. For each patient, MRI is taken at various angles for different time intervals. Hence there will be a large number of MRI images for each subject and analyzing these many images will be a tedious task for the clinician. Sometimes the minor changes in the human brain due to the abnormality may not be visible to human eyes and this may result in the erroneous prediction of disease. Therefore a computerized classification technique will be helpful for the clinicians during diagnosis. Based on learning type, classification techniques are broadly classified into three types namely, supervised classification, unsupervised classification, and reinforcement learning. Our research work concentrates on unsupervised classification. Due to the complexity involved in dementia classification, popular unsupervised clustering techniques like K-Means and Fuzzy C Means (FCM) algorithms provide poor classification accuracy in the range of 60-71%. Swarm Intelligence (SI) algorithms are widely used as clustering technique but the dementia classification performance metrics earned by these SI algorithms based clustering techniques are also not sufficient. Hence other approaches have to be developed since a good classifier should provide at least 85% accuracy.Therefore the main objective of this research work is to build an accurate classifier for categorizing the input MRI image as two classes: Nondemented and demented. MRI images of 65 non-demented and 52 demented subjects downloaded from Open Access Series of Imaging Studies (OASIS) website are con