Can Artificial Intelligence Substitute Right-Heart Catheterization With Chest X-Rays?

I read, with great interest, the article by Hirata et al., which reported on the prediction method for elevated pulmonary artery wedge pressure (PAWP) through chest x-rays, using deep learning (DL). Their model achieved significantly higher predictive performance over the traditional radiographic parameter—cardiothoracic ratio—for heart failure. The availability of noninvasive assessment of circulatory dynamics will be a preferred method as the use of machine intelligence becomes more widespread. I also found that the authors from the same institute reported that DL can predict elevated pulmonary artery pressure (PAP) using a similar method. Now that it is clear that both PAWP and PAP can be estimated by DL, the next goal is to differentiate between precapillary and postcapillary pulmonary hypertension, as the direction of the future, as the authors noted. I was wondering whether a single model can estimate these 2 parameters at the same time, and if the other remaining important indicator—right atrial pressure (or central venous pressure)—can be predicted by chest radiography. Finally, the most important limitation of the results of the study by Hirata et al. is probably that their DL model could not outperform the diagnostic accuracy of echocardiography. In the machine-learning era, I think the combination of chest radiography and electrocardiography, as attempted by Ostojic et al. several decades ago, will be re-evaluated.