FC 120MAGNETIC RESONANCE IMAGING TEXTURE ANALYSIS PREDICTS INTERSTITIAL FIBROSIS / TUBULAR ATROPHY IN TRANSPLANTED KIDNEYS: A SINGLE CENTER CROSS-SECTIONAL STUDY

Abstract Background and Aims Interstitial fibrosis / tubular atrophy (IFTA) is a common, irreversible progressive form of chronic allograft injury, it considered critical predictor kidney outcomes. Inflammation, both microvascular interstitial, on the contrary regarded as reversible graft injury. Since treatments for rejection other causes dysfunction bear substantial toxicity could have limited efficacy, extent scarring crucial information clinician, to evaluate risks benefits specific therapies. The diagnosis pathology acquired through biopsy, which an invasive procedure can be subjected sampling bias. Magnetic resonance imaging (MRI), especially with functional techniques, has emerged possibility non-invasive estimation tissue fibrosis; nevertheless, MRI not widely available. Texture analysis (TA-MRI) radiomic technique that provides quantitative assessment heterogeneity from standard images, generating features fitted into machine-learning model assess their ability predict clinical or histological parameters. Method Single-center cross-sectional observational cohort study enrolling transplant recipients who underwent biopsy within 6 months indication, at “Azienda Ospedaliero-Universitaria di Modena”, Italy. was approved by local Ethical Committee (AOU0010167/20). primary outcome identify best TA-MRI subset IFTA > 50% in biopsy. Secondary outcomes were of: 25%, presence total inflammation (ti) (glomerulitis + peritubular capillaritis [g+ptc]). Graft reported according Banff 2017 system. Radiomic performed axial T2 pre-contrast T1 fat-suppressed post-contrast sequences. whole renal parenchyma (PAR) segmented labelled T1, cortex (COR) only T2. After preprocessing, PyRadiomics used extract features. removal shape features, 93 included reduced using LASSO regression produce signatures. These introduced Machine Learning (ML) models test association Results are AUC value sensitivity specificity. Sixty patients study, 67 – pairs available analysis. Demographic characteristics enrolled depicted table 1; main parameters biopsies 2. Among ML models, three showed acceptable performance. COR “firstorder_minimum/firstorder_range/glrlm_run_entropy” IFTA>50% (AUC=0.77, sensitivity=73%, specificity=71%), PAR “firstorder_energy” IFTA>25% (AUC=0.71, sensitivity=74%, specificity=51%), “firstorder_energy/gldm_small_dependence_low_gray_level_emphasis” g+ptc >0 (AUC=0.74, sensitivity= 78%, specificity=68%); see figures 1–3. No prediction detected ti >0. Conclusion Our shows feature signatures degree biopsies, diagnostic results suggest further investigating sequences potential tool parenchymal Moreover, since jeopardized sample size, we hypothesize evaluation provide more comprehensive regarding parenchyma. To this hypothesis, currently evaluating baseline eGFR variation over time.