Deep-Learning with Domain-Specific Pretraining for Breast Cancer Neoadjuvant Chemotherapy Response Prediction from
Christoph Fürböck1,2,3, Ivana Janickova1,2,3, Georg Langs1,2,3
1Department of Biomedical Imaging and Image-guided Therapy, Computational Imaging Research Lab, Medical University of Vienna, 1090 Vienna, Austria.
Cancers
|May 13, 2026
Summary
A deep-learning model using pre-treatment ultrasound images accurately predicts neoadjuvant chemotherapy (NAC) response in breast cancer. Domain-specific pretraining significantly improved model performance, highlighting AI
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Predicting neoadjuvant chemotherapy (NAC) response in breast cancer is crucial for personalized treatment planning.
- Current prediction methods often lack accuracy and accessibility.
- B-mode ultrasound is a widely available imaging modality.
Purpose of the Study:
- To evaluate a deep-learning model's ability to predict NAC response using pre-treatment B-mode ultrasound images.
- To compare different model training strategies: training from scratch (SC), transfer learning (TL), and domain-specific pretraining (USP).
Main Methods:
- Retrospective study of 245 female patients (253 lesions) treated with NAC.
- ResNet18-based deep-learning models were trained using pre-treatment ultrasound images and clinical features.
- Model performance was assessed using accuracy, specificity, and sensitivity.
Main Results:
- The best-performing model, USP Image, achieved 0.76 accuracy, outperforming other models and those using clinical features (p<0.05).
- Domain-specific pretraining (USP) significantly improved model performance compared to SC and TL.
- AI analysis identified distinct imaging features associated with complete response (CR) and non-CR.
Conclusions:
- Deep learning with domain-specific pretraining on ultrasound images can accurately predict NAC response in breast cancer.
- Ultrasound serves as a cost-effective tool for AI-driven predictive oncology.
- This approach supports personalized treatment planning for breast cancer patients.