Artificial intelligence-based detection of aortic stenosis from chest radiographs

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Aortic stenosis (AS) is a significant cardiovascular condition.
• Chest radiography is a fundamental and widely accessible diagnostic tool.
• Current methods for AS detection may have limitations in certain settings.

Purpose of the Study:

• To develop and evaluate artificial intelligence (AI) models for detecting aortic stenosis (AS) using chest radiographs.
• To assess the performance of deep learning models in differentiating AS from normal cases based on radiographic data.

Main Methods:

• Utilized a dataset of 10,433 retrospectively collected digital chest radiographs from 5,638 patients.
• Trained, validated, and tested three deep learning models for AS detection.
• Ensemble modeling (soft voting) was employed to combine the predictions of the three models.

Main Results:

• The ensemble AI model achieved an area under the receiver operating characteristic curve (AUC) of 0.83 in the test dataset.
• The model demonstrated a sensitivity of 0.83 and specificity of 0.69 for AS detection in the test set.
• The negative predictive value was high at 0.97, indicating strong performance in ruling out AS.

Conclusions:

• Deep learning models can effectively differentiate between chest radiographs of patients with and without aortic stenosis.
• AI-powered analysis of chest radiographs offers a potential supplementary tool for AS detection.
• The accessibility and cost-effectiveness of chest radiography make this AI approach a valuable resource for clinical practice.