Radiomics-based Prognostication in Primary Sclerosing Cholangitis: a Proof-of-concept Study

IntroductionMagnetic resonance imaging (MRI) in primary sclerosing cholangitis (PSC) risk assessment generally relies on qualitative analysis, leading to interpretation variability. Radiomics emerges as a promising field for developing quantitative radiological biomarkers for PSC monitoring and risk stratification.AimThis study aims to identify and validate radiomic features from MRI images to identify patients at higher-risk of developing poor outcome.MethodsThis is a prospective, observational study (Jan 2019 - Dec-2022) recruiting 100 PSC patients undergoing routine gadoxetate disodium-enhanced MRI with standardized protocol. From PyRadiomics implemented in Python both morphological and radiomics features were extracted by five selected MRI sequences. Patients were categorized into high-risk groups based on the Mayo risk score (MRS)>0 and the liver stiffness measurement (LSM)>9.6kPa. Predictive features from a training cohort of 58 patients were validated in 42 additional PSC patients, followed by survival analysis in the combined 100-patient cohort.ResultsOne-hundred patients were analysed. Among the 58 patients of the training cohort 15 (25.0%) and 17 (30.0%) were defined at high-risk by MRS and LSM. One-hundred and seven radiomic features were extracted from each of the 5 MRI sequences selected. GLRLM-Run Entropy in T2WI with fat saturation significantly correlates with estimates of clinical outcomes with an OR of 4.04 (CI 3.63-4.71, p=0.0002) for MRS and 2.93 (CI 1.71-3.43, p=0.009) for LSM (Table). Its prognostic potential was confirmed on observed clinical events by univariate Cox analysis (HR per 0.1 of increase 1.478 95% CI 1.175;1.860) showing an excellent predicting performance (C-index = 0.85).ConclusionsThis study highlights the potential of a unique, quantitative radiomic feature for monitoring and risk-stratify PSC patients. Its quantitative nature, and extraction using free, globally available software makes it a promising candidate in radiological biomarkers’ field in PSC. Additional research with wider cohorts and longer follow-up is required to confirm these findings.