Artificial intelligence-assisted point-of-care lung ultrasound to compare B-lines and congestion scores with NTproBNP levels prior to discharge patients with acute heart failure stage C

Abstract Background Regarding acute heart failure (AHF), it is still unclear whether conventional congestion scores in relation to NT-proBNP levels are more effective than artificial intelligence (AI)-assisted B-line quantification. This study aimed to determine whether AI-assisted lung ultrasound B-line assessment correlates more strongly with NT-proBNP levels than conventional congestion scores in Stage C AHF patients at discharge. Methods In this prospective cohort study, 21 Stage C AHF patients underwent AI-assisted lung ultrasound examination using the Philips Lumify S4-1 system prior to discharge. We compared correlations of both B-line quantification and congestion scores with NT-proBNP levels, and assessed their predictive value for 30-day rehospitalization. Results AI-assisted B-line scores showed modest correlation with NT-proBNP levels (r=0.090, p=0.697), compared to congestion scores (r=0.380, p=0.089). However, B-line assessment demonstrated superior predictive value for 30-day rehospitalization (AUC=0.9556) and independently predicted rehospitalization after multivariate adjustment (adjusted OR=4.567, 95% CI:1.248-16.721, p=0.022). Conclusions While conventional congestion scores demonstrated a stronger correlation with NT-proBNP levels, AI-assisted B-line quantification was superior at predicting rehospitalization. Therefore, it implies that the automated B-line measurement has the potential to identify features of residual congestion relevant to post-discharge outcomes in Stage C heart failure patients.Key characteristics and AI validation Primary and secondary outcomes