Transfer learning with chest X-rays for ER patient classification.

Area of Science:

• Medical imaging
• Artificial intelligence in medicine
• Pulmonary medicine

Background:

• Differentiating cardiac and infectious etiologies in acute respiratory distress syndrome (ARDS) presents a diagnostic challenge in emergency settings.
• Accurate classification is crucial for timely and appropriate patient management.

Purpose of the Study:

• To improve the classification accuracy of cardiac versus infectious causes of pulmonary findings in emergency room (ER) patients with ARDS.
• To evaluate the utility of a deep-learning model trained on external data for feature extraction and classification.

Main Methods:

• Retrospective analysis of 171 ER patients diagnosed with ARDS.
• Classification of patients based on clinical data and chest X-ray (CXR) imaging.
• Application of a deep-learning model trained on an external image dataset to extract features and enhance classification accuracy.

Main Results:

• The deep-learning model, utilizing external image data, significantly improved classification accuracy for ER patients.
• Analysis identified key clinical features critical for differentiating cardiac from infectious etiologies.
• The developed model demonstrated enhanced performance even with limited internal image data.

Conclusions:

• Deep-learning models trained on external datasets can effectively augment classification capabilities when internal data is insufficient.
• Integrating clinical data with AI-driven image analysis offers a promising approach for urgent ARDS patient evaluation.
• The study provides a publicly accessible tool for improved ER patient classification.