Longitudinal DCE-MRI deep-learning system supports breast cancer neoadjuvant chemotherapy assessment across multicenter cohorts

Deep-learning system using pre/post neoadjuvant chemotherapy DCE-MRI and clinical data predicted residual tumor burden with AUCs 0.8885 internal, 0.8628 external, 0.8695 I-SPY2; NAC response AUCs 0.8400/0.8299; subtype AUCs 0.8066/0.7748/0.7661 in 1,515 patients.

• The MANAS model was developed and tested on a multicenter dataset of 1,515 patients with locally advanced breast cancer, using pre- and post-NAC DCE-MRI plus clinical information.
• It combined three prediction tasks: molecular subtype, neoadjuvant chemotherapy response, and residual tumor burden.
• Confidence-based decision criteria were used to defer low-confidence cases to clinicians and were reported to enhance predictive performance.

Automated summary

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