Application of Deep Learning in Pneumothorax Diagnosis: A Lightweight U-Net Model with Efficientnet Encoder

Pneumothorax is a life-threatening condition caused by air in the pleural cavity. Accurate and timely diagnosis is essential but often limited by radiologist experience. This study presents a deep learning model combining EfficientNet and U-Net to detect pneumothorax from chest Xrays. Trained on public datasets, the model achieved strong performance with a ROC_AUC of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.86, \text{mIoU}$</tex> of 0.20, and mDice of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 3 0}$</tex>. The results indicate that the proposed model can support automated, accurate diagnosis, aiding clinical decisionmaking and improving patient outcomes.