A new AI framework called TRACE is proposed for breast ultrasound diagnosis. It uses structured radiology reports to guide the training process and enables image-only diagnosis at test time. The framework refines image-derived concepts through a teacher-guided editing mechanism within an ordered concept space. To address incomplete annotations, it introduces Strategic Concept Missing Training (SCMT) and trains an image-only self-editor via edit distillation for autonomous concept refinement. Experiments show that TRACE achieves superior performance and improved cross-domain robustness compared to existing methods. --- Why it matters: This matters because current deep learning methods for breast ultrasound diagnosis lack interpretability and robustness, making it difficult to trust their results. TRACE addresses these issues by providing a more interpretable and robust framework, which is crucial for clinical applications where accurate diagnoses are essential. Source: https://arxiv.org/abs/2608.20809