Breast AI in Clinical Practice…Are We There Yet? (2024)-T1-CBR01-2024

Video Transcription

Video Summary

The session on breast AI in clinical practice explored how AI is being integrated into cancer detection and supplemental imaging, alongside ongoing AI trials. Fredrik Strand discussed differences in AI application based on geographic settings, highlighting approaches like AI as an independent reader or as part of a triage to optimize radiologist involvement. In studies, AI alone sometimes detected fewer cancers but required less recall and workflow, while combining AI with radiologists improved detection rates and reduced workload. Results varied based on initial cancer detection rates and recall levels.<br /><br />Etta Pisano emphasized regulatory concerns, notably in the U.S., where autonomous AI isn't allowed under MQSA regulations mandating radiologist involvement. The efficacy and safety proof are crucial for FDA approval, and AI's real-world performance sometimes deviates from initial expectations, as seen in past implementations like CAD (Computer-Aided Detection).<br /><br />Linda Moy discussed hurdles in AI adoption, noting the exponential increase in FDA-approved AI devices but highlighting market penetration challenges. She suggested identifying specific use cases tailored to practices and potential AI benefits in improving accuracy and reducing workload. However, hurdles such as biases in training data, clinical trial needs, reimbursement concerns, and educational requirements were also noted as barriers to broad AI implementation in breast imaging.

Keywords

breast AI

cancer detection

AI trials

radiologist involvement

regulatory concerns

FDA approval

AI adoption

training data biases