期刊名称：International Journal of Advances in Soft Computing and Its Applications
印刷版ISSN：2074-8523
出版年度：2018
卷号：10
期号：3
出版社：International Center for Scientific Research and Studies
摘要：Image intensity values which are extracted from magnetic resonance imaging (MRI) are not standardised and do not have tissue-specific interpretation due to the limitation of MRI instrumentation. The limitation poses many difficulties on data visualisation and texture feature analysis. Intensity and texture features extracted from MRI are not comparable for each inter-scan and intra-scan. Hence, they are not appropriate to be applied in supervised learning approaches to analyse the texture of white matter lesions. Consequently, this drawback often requires a standardisation method prior to further image analysis, which remains a common problem. In this study, a new automated method for image intensity standardisation is proposed to provide a standard intensity scale. In the proposed method, the landmarks in the intensity scale are automatically detected in the brain tissue intensity distribution using an adaptive outlier detection approach. Subsequently, landmarks are used to transform the brain tissues and lesion intensity into a standard scale by using the proposed transformation method. The method is validated using the cranial MRIs (FLAIR sequence) that contain the white matter lesions from 10 subjects during their 3-year follow-up study. A paired t-test: t(29) = 2.045 and P(29)=1.42x10-15 where P<0.0001 confirms the significant difference in the before and after intensity range. In addition, intensity and texture features between the output images from the proposed approach and a leading intensity standardisation algorithm are further compared using the coefficient of variation, Pearson's correlation coefficient, and Kullback-Leibler divergence. Finally, qualitative evaluation of the MRI intensity is presented using the fixed-window-level method.
关键词：Intensity normalisation; intensity standardization; MRI; outlier detections; white matter lesions
