Texture Analysis of T2-Weighted Images as Reliable Biomarker of Chronic Kidney Disease Microstructural State

Objectives The chronic kidney disease (CKD) diagnostics consists of three most basic groups of examinations: laboratory tests, radiological imaging and histopathological examinations. In most severe cases when clinical decision is required to be fast and undisputed the histopathological tests are the only examinations capable to fulfill this need. Unfortunately, preforming histopathological test require undertaking kidney biopsy which is invasive procedure excluding numerous patients. The aim of this study is to create algorithm which will be able to divide CKD patients of active and non-active phase on the basis of MRI texture analysis compared with histopathological examinations. Methods The study is based on MRI examinations of healthy volunteers (group 1, N=14) and patients suffering from CKD who were subjects of kidney biopsy. On the basis of histopathological outcome, patients were divided to group with active faze of CKD (group 2, N=58) and non-active faze (group 3, N=22). The T2-weighted images of these examinations were the subject for analysis of Support Vector Machines model created with qMazDa software which was trained to classify images to appropriate group of CKD activity. Results The evaluation metrics calculated for final SVM models corresponding to confusion matrices are as follows: for texture analysis – balanced accuracy 81.6%, sensitivity 68.2-92.0%, specificity 82.5-97.5%, precision 62.5-95.8%; for texture and shape analysis - balanced accuracy 87.3%, sensitivity 77.3-100.0%, specificity 87.5-100.0%, precision 65.4-100.0%. Conclusion Texture analysis of T2 weighted images associated with kidney shape features seem to be reliable method of assessing state of ongoing CKD.