Deep learning for visceral pleural invasion in non-small cell lung cancer

The authors aimed to evaluate and compare the diagnostic performance of deep learning models and a conventional clinical-radiological feature model for predicting visceral pleural invasion (VPI) in non-small cell lung cancer using preoperative CT scans. While the clinical model demonstrated higher specificity, it had low sensitivity, whereas the best-performing deep learning model showed improved sensitivity but lower positive predictive value. Overall, neither approach achieved optimal clinical utility, indicating the need for further advancements in noninvasive VPI prediction methods.