Non-invasive Classification of Orbital Tissue Pathology Based on Texture Analysis Parameters from Magnetic Resonance Images

Abstract

Introduction:This proof-of-concept study aims to demonstrate that quantitative texture analysis of magnetic resonance imaging (MRI) of orbital tumours can produce a unique footprint as an adjunct to histology and as a reference guide.Methods:‘MaZda’ texture analysis software programme version 4.6 and MedCalc software (18.10.2) were used for data procuring and analysis. The following statistical analyses were performed: analysis of variance (ANOVA) to compare histogram means; Wilcoxon signed rank sum test to compare intra-lesion variability; Mann-Whitney U test to compare inter-lesion feature differences; area under curve to test sensitivity and specificity in differentiating abnormal from normal tissue; and Fisher’s coefficient and linear discriminant analysis to reduce data vector dimensions.Results:Thirteen cases were assessed. Eleven cases were imaged with similar protocols. The software produced characteristic histograms and other quantitative parameters for a variety of orbital pathologies. The mean of histograms differed significantly between pathologies as well as other texture features and there were significant minimal misclassifications on inter-tumour analysis. Though the test showed 100% sensitivity in detecting abnormal tissues, it was not specific in differentiating some of the adnexal normal tissues from certain types of orbital tumours.Conclusions:This proof-of-concept study confirms that the non-invasive classification of orbital tumours is achievable. Further studies are needed to create a larger reference framework.