Multimodality AI-Enhanced Breast Imaging: Transforming Detection and Diagnosis (2025)-M8-CBR05-2025

Video Transcription

Video Summary

The session reviews how AI is reshaping breast imaging across screening/triage, treatment-response prediction, and risk assessment.<br /><br />Linda Moy summarizes evidence for AI in screening mammography, emphasizing lessons from older CAD systems that lowered accuracy. Recent meta-analyses and large prospective European trials show AI can increase cancer detection (~10–40%), keep recall rates similar or slightly higher, improve positive predictive value, and reduce workload when used for triage or as a second reader in double-reading settings. However, fully standalone AI screening is not yet ready, partly due to interval cancers. Retrospective studies suggest AI detects most screen-detected cancers and can identify a meaningful fraction of interval cancers (often aggressive), potentially improving early-stage detection. She also highlights workflow tradeoffs (removing “easy” cases may leave radiologists with only complex reads) and a new “safeguard” workflow where AI-flagged normal reads get an extra review, boosting detection but raising recalls.<br /><br />Fernando Collado-Mesa covers multimodal AI to predict response to neoadjuvant therapy. MRI is currently best for response assessment, but prediction requires integrating imaging, pathology, and clinical data over time. Systematic reviews show multimodal and longitudinal models perform best, yet most studies are small and retrospective; no FDA-cleared tool exists. Key gaps include prospective trials, external validation, fairness, and regulatory pathways.<br /><br />Liz Morris explains AI-enhanced, multimodal risk prediction. Imaging-based AI risk models outperform traditional epidemiologic tools, work better across ethnicities, and enable dynamic, short-term personalized screening—potentially guiding both escalation (supplemental imaging) and de-escalation (less frequent screening).

Keywords

AI breast imaging

screening mammography AI

computer-aided detection (CAD) limitations

double-reading triage workflow

interval cancer detection

recall rate and positive predictive value

multimodal longitudinal response prediction

neoadjuvant therapy MRI assessment

AI-based breast cancer risk prediction