Development and Evaluation of a Deep Learning-Based Pulmonary Hypertension Screening Algorithm Using a Digital...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Pulmonary hypertension often goes undiagnosed due to low clinical suspicion and limited accessibility of screening tools like echocardiography.
• Early detection is crucial for timely treatment of underlying causes and improving patient prognosis.
• A readily available screening tool is needed to identify elevated pulmonary artery systolic pressure.

Purpose of the Study:

• To develop and validate a deep learning-based method for detecting elevated pulmonary artery systolic pressure using phonocardiograms (PCGs).
• To assess the feasibility of using digital stethoscopes and AI for noninvasive pulmonary hypertension screening.

Main Methods:

• A deep convolutional network was trained using approximately 6000 PCG recordings with known pulmonary artery systolic pressure values and ~169,000 unlabeled PCGs.
• The model was trained in a semisupervised manner to detect pulmonary artery systolic pressures ≥40 mmHg.
• PCG recordings were processed into mel-spectrograms, and GradCAM++ was used for result visualization.

Main Results:

• The model achieved an average area under the receiver operator characteristic curve of 0.79 across cross-validation.
• On the testing dataset, the model demonstrated a sensitivity of 0.71 and a specificity of 0.73.
• The GradCAM++ technique successfully highlighted physiologically relevant segments in PCG recordings indicative of pulmonary hypertension.

Conclusions:

• Digital stethoscopes combined with deep learning algorithms offer a feasible, low-cost, noninvasive, and accessible screening tool for early pulmonary hypertension detection.
• This approach has the potential to improve early diagnosis and management of pulmonary hypertension.