Morphological and texture based classification of dementia from MR images

Abstract

With degradation in quality of life, threat of brain disorders has become a serious concern from past few decades. Dementia is one such abnormality which includes a group of symptoms that deteriorate cognitive functions of the brain. The use of neuroimaging techniques has revealed the abnormal anatomy of Corpus callosum (CC) for diagnosing various brain disorders. This paper focuses on classification of magnetic resonance (MR) images of dementia using CC features. CC is segmented from each mid-sagittal brain MR image using K-means clustering algorithm and then used for feature extraction. Significant features between demented and normal groups are identified using statistical analysis. Depending upon the statistical significance, hybrid feature vectors are designed for male and female dataset. Support vector machine (SVM) and Back propagation neural network (BPNN) classifiers are trained and tested using the designed feature vectors. Considering the sexual dimorphism of CC structure, feature classification is performed separately for male and female data. This paper reports the highest classification accuracy of 97% for male data and 95% for female data.