Background: Pulmonary arterial hypertension (PAH) is a progressive disease with high morbidity and mortality. Assessing PAH prognosis is paramount for guiding therapy and transthoracic echocardiography (TTE) remains the mainstay of regular assessment. We hypothesized that deep learning networks (DL) could provide essential prognostic information of comparable quality as a comprehensive expert assessment. Methods: All PAH pts. with routine TTE between 2005 and 2018 were included. A combination of U-Net DL-frameworks was developed to automatically segment cardiac chambers and extract geometric information throughout the cardiac cycle (see Figure ). The prognostic value of chamber dimensions and functional parameters for all-cause mortality was assessed using Cox proportional-hazard analyses. Results: Overall, 408 PAH patients (median age 59 years, 74% female; 34% idiopathic PAH, 66% associated PAH) were included. Over a median follow-up of 0.6 years , 196 patients died. On univariable Cox analysis automatically determined right atrial area, right ventricular (RV) area, RV fractional area change, RV inflow diameter, RV length, left ventricular (LV) eccentricity index (p<0.001 for all) were significantly related to mortality. On multivariable analysis DL-based RV fractional area change (HR 0.97/%, p=0.004), right atrial area (HR 1.03/cm 2 , p=0.01) and LV longitudinal diameter (HR 0.92/cm, p=0.02) emerged as independent predictors of outcome. The prognostic value of all EDL parameters was non-inferior to traditional measures obtained by expert echocardiographers (p>0.05 for all concordance-C comparisons). Conclusions: The current study demonstrates for the first-time the utility of ensemble-based DL algorithms to assess prognosis of PAH based on a large tertiary centre dataset. Due to the automated process, these DL algorithms can ultimately serve as online based decision-making tools and the performance is comparable to expert investigators.