Deep Learning for Algorithmic Detection of Pulmonary Hypertension Using a Combined Digital Stethoscope and Single-lead Electrocardiogram

Pulmonary hypertension (PH) is a syndrome resulting from restricted flow through the pulmonary circulation causing increased pulmonary vascular resistance and ultimately right heart failure. There are several different subtypes of PH, however, all carry a poor prognosis and often result in or hasten death. Multiple pathogenic pathways have been implicated in the development of PH, including those at the molecular and genetic levels and in the smooth muscle and endothelial cells and adventitia.

Patients with PH are classified into five groups based on the etiology and mechanism of the disease group.1 Group 1, also called pulmonary arterial hypertension (PAH), is associated with several other systemic diseases (e.g., connective tissue disease), genetic syndromes, or drugs. Whereas, group 2 is associated with left-sided heart disease. Group 3 is due to chronic lung disorders and hypoxemia. Group 4 is due to pulmonary artery obstructions and is the subtype found in patients with chronic thromboembolic pulmonary hypertension. Lastly, Group 5 is idiopathic PH or PH with unidentified mechanism.

PH is a major pathophysiological disorder that can involve multiple clinical conditions and can complicate most cardiovascular and respiratory diseases. PH is defined as an increase in mean pulmonary artery pressure (mPAP) >20 mm Hg at rest, as assessed by right heart catheterization. Due to the invasive nature of right heart catheterization, echocardiography is an established non-invasive alternative diagnostic tool.

About 80% of all right heart catheterizations have evidence of elevated PA pressures (mPAP> 19 mm HG) and ~60% have a mean PA pressure > 25 mm Hg. Also, the prevalence of elevated PA pressure is ~ 50% on clinically indicated echocardiograms.5 Elevated PA pressure either by echocardiography or right heart catheterization is associated with increased mortality, hospitalizations and heart failure admissions.

However, since PH requires either echocardiogram or invasive catheterization, it remains underdiagnosed. Identification of a minimally invasive and rapid screening process for PH will help identify this at risk group in a primary care setting to target for further evaluation and aggressive risk factor modification. We hypothesize that combining phonocardiography (PCG) from heart auscultation with electrocardiography (ECG) may provide specific elements that correlate with PA pressures on echocardiogram and can help screen for the probability of pulmonary hypertension in a patient.