PET/CT six-channel early fusion deep learning best for lung cancer subtype discrimination

PET/CT six-channel early fusion deep learning model achieved highest mean AUC 0.766 for differentiating squamous cell carcinoma from adenocarcinoma in non-small cell lung cancer (n=220); external validation needed.

• The six-channel early fusion model outperformed PET-only, CT-only, and dual-branch fusion in mean AUC, balanced accuracy, specificity, and Matthews correlation coefficient.
• Dual-branch fusion had the highest sensitivity (0.755) but lower AUC than the PET-only model.
• The study was retrospective and lacks external validation, so clinical application is not yet supported.

Automated summary

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