A new study shows quantitative ultrasound, powered by a deep learning pipeline, can accurately predict three-year survival in patients with locally advanced breast cancer before treatment begins.
This non-invasive biomarker addresses limitations of conventional ultrasound by using normalized power spectra to generate system-independent parameters. Researchers integrated data scaling, oversampling, feature selection, and classification into a single framework to reduce data leakage and improve analytical efficiency.
Among the parameters, average acoustic concentration was most predictive. The model achieved 95% recall and 91% precision in identifying survivors, distinguishing patients likely to survive from those at higher risk over three years.
Predicting survival before treatment could guide adjustments in therapeutic intensity and support more personalized treatment strategies. The findings suggest quantitative ultrasound, supported by deep learning, has potential as a clinically relevant prognostic biomarker.