The classification of brain magnetic resonance imaging (MRI) images into normal and abnormal classes, has great potential to reduce the radiologists workload. Statistical analysis based approaches has been widely employed for this purpose which are comprised of four stages such as pre-processing, feature extraction, feature reduction and classification. The outcome of such approaches are highly dependent upon the image quality: better the image, higher the outcome. In this paper, we present a hypothesis that the quality of the image, which is enhanced at the pre-processing stage, can play a significant role in enhancing the classification performance of any statistical approach. To strengthen our theory we first employed an improved image enhancement technique, which consists of three different sub-stages: noise removal using median filter, contrast enhancement using histogram equalization technique and image conversion from gray-scale to RGB. After image enhancement, we extract features from an enhanced MR brain image using a discrete wavelet transform and these feature are further reduced by color moments i.e mean, standard deviation, and skewness. Finally, we trained an advanced deep neural network (DNN) to categorize the human brain MRI images as normal or pathological. The approach obtained 95.8% which is significantly higher than the previous state-of-the-art techniques. The result evident that our hypothesis about the role of image enhancement process in medical image classification, is realistic and also have potential to improve the performance of other medical image analysis technique.
Keywords: Artificial neural network; Color moments; Discrete wavelet transform; Histogram equalization; MRI classification.
Copyright © 2020 Elsevier Ltd. All rights reserved.