Dual-attention deep learning model accurately differentiates pulmonary invasive mucinous adenocarcinoma from inflammatory nodules on CT

A dual-attention deep learning model differentiated pulmonary invasive mucinous adenocarcinoma from inflammatory nodules on CT, achieving AUC 0.99, sensitivity 100%, specificity 94% in internal validation. Needs external validation.

• Retrospective single-center study with 443 patients (1,409 CT slices) using an 80/20 internal random split.
• SE-DAS ResNet outperformed standard ResNet50 (AUC 0.914, P<0.05); a real-time inference platform was developed.
• Limitation: No external validation; reliance on internal split may yield optimistic performance estimates.

RadPigeon summaries are original and for information only. They are not clinical advice.