Ultrasound is a critical tool for detecting fluid in the lungs, particularly in the alveolar and interstitial spaces, by identifying artifacts known as B-lines. This study evaluated the correlation between a deep-learning-generated B-line severity score and pulmonary congestion severity, as determined by clinical assessments such as the Composite Congestion Score (CCS) and Rothman Index. It also explored the score’s responsiveness to treatment changes. Conducted at an academic medical center from 2018 to 2019, the study included 110 patients presenting with dyspnea or hypoxia and a diagnosis of heart failure or pulmonary edema. Daily lung ultrasounds were performed using an eight-zone scanning protocol and analyzed by a deep learning algorithm. Results demonstrated a significant association between the B-line severity score and the CCS, indicating a coefficient of 0.7 (95% CI 0.1–1.2, p=0.02), but no notable correlation with the Rothman Index.
B-lines, which reflect interstitial fluid and loss of lung aeration, are critical indicators of pulmonary congestion seen in conditions like acute heart failure, pneumonia, and COVID-19. However, their quantification often varies with user experience. Deep learning algorithms offer a solution by automating B-line assessment, enabling reproducibility and broader accessibility, even for clinicians with limited ultrasound expertise. This capability has significant implications for enhancing diagnostic precision, monitoring treatment progress, and improving patient outcomes, particularly in resource-limited settings. The findings suggest that AI-driven B-line scoring can serve as an objective, scalable tool for tracking disease severity and treatment response in hospitalized heart failure patients.
