Prediction of ipsilateral and contralateral pneumothorax using a simple chest X-ray

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonology

Background:

• Spontaneous pneumothorax (SP) management requires accurate prediction of recurrence and contralateral occurrence.
• Simple chest X-rays are a primary tool for diagnosing SP.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting ipsilateral recurrence and contralateral occurrence of SP.
• To identify key features in chest X-rays associated with SP development using Grad-CAM.

Main Methods:

• Retrospective analysis of 1,086 SP cases from July 2017 to June 2023.
• Application of machine learning algorithms (Gradient Boosting, Logistic Regression, KNN, Random Forest, Neural Network) using X-rays and clinical data (age, sex, smoking, COPD, surgery).
• Utilized Gradient-weighted Class Activation Mapping (Grad-CAM) for X-ray region analysis.

Main Results:

• Machine learning models demonstrated fair performance in predicting SP development.
• Top models varied by prediction type (ipsilateral recurrence, contralateral occurrence) and patient group (young/older) and side (right/left).
• Grad-CAM analysis highlighted apical lung portions as strongly associated with SP development.

Conclusions:

• Machine learning models integrating simple chest X-rays and clinical data offer a viable approach for predicting SP.
• Findings confirm the clinical relevance of apical lung regions in SP pathogenesis and prediction.