Non-invasive estimation of right atrial pressure using magnetic resonance imaging: a development and validation study

Right atrial pressure (RAP) is a key determinant of right-sided heart function and guides diagnosis and management across a wide range of cardiovascular diseases. However, reliable non-invasive estimation remains difficult. We aimed to develop and validate a magnetic resonance imaging (MRI)-based model for predicting invasive RAP.

This study included 181 adults who underwent clinically indicated MRI and right heart catheterization within 48 h between 2016 and 2025. The sample was split into a development (N = 125) and an independent held-out test-set cohort (N = 56) stratified by RAP category. Generalized additive models and systematic transformation screening were used to characterize non-linear predictor-outcome associations. Predictors were selected using pre-defined criteria (ΔAkaike information criterion ≥ 2, partial R2 ≥ 0.02, least absolute shrinkage and selection operator [LASSO] bootstrap inclusion ≥90%, physiological plausibility). The final model included inferior vena cava expiratory diameter (linear and quadratic terms), liver T1 relaxation time (linear and quadratic terms), right atrium area (linear), and body weight (linear). In the development cohort, performance was strong (adjusted R2 = 0.698; residual standard deviation 2.58 mmHg) and remained consistent across three internal validation approaches. Held-out test-set validation showed an R2 of 0.621, a root-mean-square error of 2.78 mmHg, a mean absolute error of 2.07 mmHg, and a concordance coefficient of 0.777. Calibration in the testing set was excellent (intercept + 0.18, slope = 0.92; integrated calibration index 0.65 mmHg). Discrimination for elevated RAP was high with area under the curves of 0.749 (>5 mmHg), 0.949 (>10 mmHg), and 0.965 (>15 mmHg).

An MRI model incorporating abdominal venous and hepatic tissue biomarkers provides a clinically useful, well-calibrated approach for estimating invasive RAP.