Deep learning on breast DWI predicts HER2 status with AUC 0.91 in external test set

On an external test set, a hybrid CNN-Transformer deep learning model predicted human epidermal growth factor receptor 2 (HER2) status from breast diffusion-weighted MRI with AUC 0.91 (accuracy 0.84), outperforming clinical experts (AUC 0.65 and 0.63). Internal test AUC 0.93. Re…

• Retrospective, two-center study with 239 women; external test set of 44 patients from a separate center.
• On internal test, AUC 0.93 vs best benchmark model AUC 0.89; accuracy 0.90 vs 0.85.
• Hybrid CNN-Transformer uses DWI inputs (ADC map, b=0 and b=800 s/mm²) without additional data.

Automated summary

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