Texture analysis for quantifying and visualising loss of corticomedullary differentiation

Background: Currently, no method is available for non invasive, early diagnosis of chronic kidney disease. Corticomedullary differentiation calculated on native T1 maps is a promising parameter. Purpose: This research aims to find an image-analysis technique that can quantify the loss of corticomedullary differentiation on native T1 maps. Method: Texture analysis, using a grey level co-occurrence matrix, was performed on segmented images from T1 maps of kidneys of 14 healthy volunteers and 15 patients with mild diabetic nephropathy. The first-order features tested include minimum, maximum, mean, median, range, standard deviation, skewness, kurtosis and entropy. The texture features tested include contrast, correlation, homogeneity, and energy. Results: First-order statistics indicate significant differences between groups in standard deviation (p = 0.000), kurtosis (p = 0.000), skewness (p = 0.005) and entropy (p = 0.025). Moreover, a significant difference between groups was found for texture features contrast (p = 0.005) and correlation (p = 0.012). Conclusion: Texture analysis is a promising method to quantify slight changes in corticomedullary differentiation. If explained and substantiated clearly to a nephrologist, the value of contrast and correlation could be used in a clinical setting. Future research should focus on visualizing texture features in feature maps to improve its clinical application and apply texture features to a larger population.