Registration-Guided GAN Synthesizes Virtual Contrast-Enhanced CT for Lymph Node Detection

A registration-guided GAN synthesized virtual contrast-enhanced CT from noncontrast CT that improved detection of enlarged hilar/mediastinal lymph nodes: AUC rose from 0.81 to 0.90 and 0.74 to 0.88 (P<.05) for experienced radiologists on an external test set.

• Reg-GAN vCECT images achieved lower median mean absolute error (13.60 internal, 14.91 external) compared with Non-Reg-GAN (17.84, 19.49) and Public GAN (15.52, 16.37) (all P<.001).
• Enlarged lymph nodes were defined as short-axis diameter >10 mm.
• The retrospective study trained on 700 patients, tested internally on 100 and externally on 63 patients.

Automated summary

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