XRAInet: AI-based decision support for pneumothorax and pleural effusion management

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pediatric Pulmonology

Background:

• Pleural effusion and pneumothorax are critical conditions in pediatric patients.
• Accurate and timely diagnosis is essential for effective treatment and improved outcomes.
• Current diagnostic methods can be challenging, necessitating advanced decision support tools.

Purpose of the Study:

• To develop and evaluate XRAInet, an AI-driven system for detecting pediatric pleural effusion and pneumothorax.
• To assess the system's capability in determining the need for tube thoracostomy intervention.
• To enhance diagnostic accuracy and clinical decision-making in pediatric respiratory emergencies.

Main Methods:

• Retrospective analysis of 510 X-ray images from 170 pediatric patients (2005-2022).
• Development and training of a deep learning algorithm (XRAInet) on categorized patient data (Tube vs. Conservative management).
• Performance evaluation using metrics such as mean Average Precision (mAP), sensitivity, and specificity.

Main Results:

• XRAInet achieved a high mean Average Precision (mAP) score of 0.918.
• The AI system demonstrated strong localization capabilities for regions requiring intervention.
• Independent testing showed a sensitivity of 64.00% and a specificity of 96.15%.

Conclusions:

• XRAInet shows significant potential as a decision support tool for pediatric pneumothorax and pleural effusion.
• The AI system can aid in improving the detection and management of these conditions using X-ray imaging.
• Future work should focus on interpretability, data quality, and regulatory alignment for clinical deployment.