Artificial intelligence-assisted focused cardiac and lung ultrasound by novice operators in acute dyspnoea admissions: diagnosis and 30-day prognosis
C Kane, M M Crestanello, D A Schonfeld, M M Killalea, M A Gulati, G Tsaban, E Lee, T J Poterucha, Z I Attia, G C Kane, J J Thaden, S V Pislaru, J G BirdAbstract
Background/Introduction
Acute heart failure admissions have high resource use and readmissions; objective bedside tools for diagnosis and discharge risk stratification are needed.
Purpose
To evaluate diagnostic accuracy and prognostic value of AI-guided, novice-performed focused cardiac and lung ultrasound (FoCUS) in adults admitted with dyspnoea.
Methods
Prospective observational study of 109 adults admitted with dyspnoea to a cardiac unit. Four ultrasound-naïve users performed handheld FoCUS with real-time AI guidance to assess LVEF, lung congestion and IVC. Diagnostic performance for acute left heart failure was compared with clinical lung examination, chest X-ray and AI-enabled 12-lead ECG. The following were classified as positive for the purposes of acute left heart failure classification on admission: Inspiratory crackles on lung auscultation (clinical); classic findings on chest X-ray (interstitial infiltrates compatible with pulmonary congestion); the presence of grade 2 or grade 3 diastolic dysfunction on the AI-ECG algorithm for diastolic function (grade 1 or normal classified as normal) or bilateral (>3) B-lines on admission focused lung ultrasound. The reference diagnosis of acute left heart failure was adjudicated by clinicians using all available clinical data (excluding research AI outputs). FoCUS acquisition and AI outputs were obtained blinded to clinical data and comparator tests. Discharge FoCUS findings were related to 30-day death or heart failure readmission. Research ultrasound results were not available to treating teams or adjudicators.
Results
Mean age 73 (SD 11) years; 61% male. AI-lung ultrasound demonstrated sensitivity 97.0%, specificity 97.1% and accuracy 97.0% for acute left heart failure, with higher accuracy than chest X-ray (accuracy 80.0%), AI-ECG (65.7%) and clinical lung examination (53.0%). Abnormal FoCUS (residual B-lines and/or dilated IVC on ultrasound) at the time of discharge was associated with increased rates of 30-day death or heart failure readmission (age and sex-adjusted HR 5.67 [2.18-14.74]; p<0.0001), despite clinician assessment of euvolaemia at discharge.
Conclusion(s)
AI-guided FoCUS performed by novice users provides high diagnostic accuracy for acute left heart failure and identifies residual abnormalities at discharge that are associated with high 30-day adverse outcome rates.Figure 1Figure 2