Development and validation of multimodal deep learning algorithms for detecting pulmonary hypertension.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Transthoracic echocardiography (TTE) is a standard screening tool for pulmonary hypertension (PH).
• TTE often exhibits insufficient accuracy for definitive PH diagnosis.
• There is a clinical need for more precise and reliable PH screening methods.

Purpose of the Study:

• To develop and validate a multimodal fusion model for pulmonary hypertension screening (MMF-PH).
• To enhance the diagnostic accuracy of PH detection compared to conventional methods.
• To assess the model's performance across diverse patient subgroups and clinical settings.

Main Methods:

• Development and validation of the MMF-PH model using large patient datasets (n=2451, prospective n=477, external).
• Comparative analysis of MMF-PH against transthoracic echocardiography (TTE).
• Ablation study to confirm the contribution of individual model components.

Main Results:

• The MMF-PH demonstrated robust performance across multiple datasets.
• MMF-PH significantly outperformed TTE in specificity and negative predictive value.
• The ablation study confirmed the critical role of each MMF-PH module.

Conclusions:

• The MMF-PH represents a significant advancement in pulmonary hypertension detection.
• The model offers improved diagnostic accuracy, reliability, and robustness.
• MMF-PH provides a promising tool for PH screening in various clinical scenarios.