Explanatory deep learning to predict elevated pulmonary artery pressure in children with ventricular septal defects...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Pulmonary arterial hypertension (PAH) risk assessment in congenital heart disease (CHD) is critical for early intervention.
• Ventricular septal defect (VSD) is a common form of CHD where PAH can develop.
• Timely identification of PAH risk in VSD patients is essential for effective management.

Purpose of the Study:

• To investigate the potential of artificial intelligence (AI) in predicting the risk of pulmonary arterial hypertension (PAH) in patients with ventricular septal defect (VSD) using chest x-rays (CXRs).
• To develop and validate an AI algorithm for early risk stratification of PAH in VSD patients.

Main Methods:

• A retrospective study included 831 VSD patients (161 with PAH-VSD, 670 without).
• A residual neural network (ResNet) model was trained on chest radiographs to classify VSD patients based on their risk of developing PAH.
• The study endpoint was the occurrence of PAH in VSD children, either pre- or post-surgery.

Main Results:

• The AI algorithm demonstrated strong performance with an area under the curve (AUC) of 0.82 in the validation set.
• In an independent test set, the AI algorithm's AUC was 0.81, significantly outperforming human observers (AUC 0.65).
• Class Activation Mapping (CAM) revealed the AI model focused on the pulmonary artery segment for its predictions.

Conclusions:

• Artificial intelligence applied to chest x-rays can effectively identify patients with VSD who are at risk of developing pulmonary arterial hypertension.
• AI-powered analysis of CXRs offers a promising non-invasive tool for early PAH risk assessment in VSD patients.
• These preliminary findings support the use of AI for improved early detection and management strategies for PAH in VSD.