ViT-radiomics fusion improves US lymph node classification in multicenter study

A ViT-radiomics fusion model achieved an AUC of 0.95 (95% CI 0.91-0.98) for classifying benign vs malignant lymph nodes on ultrasound in a multicenter study, outperforming radiomics alone (AUC 0.79). Domain adaptation boosted external performance from 0.73 to 0.85.

• The model fuses vision transformer (ViT) deep learning features with handcrafted radiomics features.
• With few-shot domain adaptation using limited external data, the external-test AUC improved from 0.73 to 0.85.

Automated summary

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