Abstract
Patients with clinically significant portal hypertension (CSPH) are at high risk of developing severe complications of portal hypertension. The aim of our preliminary study was to assess multiparametric liver and spleen MRI and MR elastography for the estimation of CSPH. Seventy-eight patients with advanced chronic liver disease and hepatic venous pressure gradient measurements, who had liver and spleen MRI and MR elastography were assessed. Liver and spleen elasticity and viscosity maps with texture features were derived from the MR elastography acquisitions. Liver surface nodularity (LSN) and texture features were obtained from the T2-weighted MR images. Features were analyzed with regularized generalized canonical correlation analysis for dimensionality reduction. Models based on the combination of MRI and MR elastography features were trained and evaluated, using a support vector machine classifier. Forty-eight patients (62%) had CSPH. Most models based on hepatic and/or splenic MR elastography features, T2 texture features or LSN had fair performance (AUCs 0.7–0.79) for CSPH diagnosis. Multiparametric models that included MR elastography features and LSN had good diagnostic performance (AUCs 0.8–0.89), without improvement with the addition of T2 texture features. Our results suggest that combining liver and spleen MR elastography features with LSN may produce efficient models for diagnosing CSPH.