Quantitative Measurement of Pneumothorax Using Artificial Intelligence Management Model and Clinical Application

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Thoracic Surgery

Background:

• Delayed or misdiagnosed pneumothorax can lead to adverse patient outcomes.
• Artificial intelligence (AI) offers potential solutions for improving diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a deep-learning-based AI model for quantifying pneumothorax on chest radiographs.
• To integrate the AI model into a clinical treatment algorithm for thoracic surgeons.

Main Methods:

• Utilized U-net architecture for semantic segmentation and classification of pneumothorax and non-pneumothorax regions.
• Quantified pneumothorax using chest computed tomography (gold standard) and chest radiographs (AI model prediction and true labels).
• Compared AI-predicted pneumothorax amounts with gold standard and true labels, analyzing clinical outcomes.

Main Results:

• The AI model achieved 97.8% accuracy, 69.2% sensitivity, and 99.1% negative predictive value.
• AI-predicted pneumothorax amount (16%) was not significantly different from the gold standard (15%, p=0.11).
• The AI model demonstrated comparable performance in predicting pneumothorax severity for thoracostomy patients.

Conclusions:

• A deep-learning AI model can effectively estimate pneumothorax amount from chest radiographs.
• The AI model shows potential for clinical application in guiding pneumothorax treatment decisions.
• Further integration and validation are warranted to optimize AI-assisted pneumothorax management.