Artificial Intelligence Could Predict Chest Tube Drainage Necessity for Spontaneous Pneumothorax

Artificial intelligence (AI) is increasingly utilized in the medical field, primarily for diagnostic purposes. Although AI has demonstrated efficacy in pneumothorax detection using chest X-rays (CXR), it has yet to be applied for decision-making regarding subsequent treatment. This study aims to develop and evaluate an AI-based system capable of predicting the necessity of chest tube drainage (CTD) in spontaneous pneumothorax patients based on CXR and clinical data. A two-stage AI model was developed: (1) segmentation and quantification of pneumothorax size from CXR using deep learning and (2) prediction of CTD necessity using machine learning models integrating pneumothorax size and patient clinical parameters. The AI model was trained using CXR images and clinical information from 163 pneumothorax patients. Model performance was assessed using area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, and inference time. The AI model demonstrated high segmentation accuracy for pneumothorax (DSC, 76.95%; MAE, 5.06%). In predicting CTD necessity, the AUROC for the AI model incorporating pneumothorax ratio and clinical data was 89.68 (95% CI, 78.57-98.02), outperforming models without pneumothorax ratio (AUROC, 84.13) or with pneumothorax ratio alone (AUROC, 71.03). The sensitivity and specificity of the optimized AI model were 80.95% and 100%, respectively. The mean inference time was 0.75 ± 0.06 s, demonstrating potential for real-time clinical application. This study presents an AI-based clinical decision support system capable of accurately predicting the need for CTD in spontaneous pneumothorax patients. By integrating AI-driven pneumothorax quantification and clinical parameters, the model improves decision-making efficiency and accuracy. Future studies with larger datasets and prospective validation are warranted to further refine and validate this approach.