Lightweight Transformer Distills Spatiotemporal Radiomics for Breast Cancer Prediction

A lightweight Transformer model with knowledge distillation analyzing spatiotemporal radiomics from ultrafast DCE-MRI achieved AUC 0.959 and 92% accuracy for benign vs malignant breast lesion differentiation, an 8% improvement over conventional temporal radiomics.

• A gradient-dynamics-based feature selection algorithm was used to identify discriminative kinetic trajectory patterns from high-dimensional time-series data.
• Attention heat maps visualized critical enhancement phases and spatiotemporal patterns, providing interpretability of model decisions.

Automated summary

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