AI systems in breast imaging yield near-perfect negative predictive value but increase recall rates compared to radiologists, particularly for intermediate-risk classifications.
Key Details
- 1Study published in AJR compared Transpara v1.7.1 AI (ScreenPoint Medical) to 11 breast radiologists.
- 2Digital mammography cohort: 26,693 exams; DBT cohort: 4,824 exams.
- 3AI classified most cases as low risk but doubled or more the recall rate compared to radiologists (up to 41.8% for digital mammography at intermediate/elevated threshold).
- 4AI and radiologists both achieved NPVs of 99.8–99.9%.
- 5Sensitivity for AI at intermediate/elevated risk threshold: 94% (digital mammography), specificity fell to 58.6%.
- 6Researchers emphasized need for strategies to reduce false-positives, especially in intermediate-risk cases.
Why It Matters
While AI systems can improve workflow efficiency in breast cancer screening by safely excluding negative cases, their tendency to elevate recall rates could lead to unnecessary follow-ups and anxiety. Addressing false-positive rates is critical for successful clinical integration and radiologist workload management.
Source
AuntMinnie
Related News
•Radiology Business
Stanford and Rad Partners Propose Framework for Evaluating Radiology AI Models
Stanford and Rad Partners developed a structured framework for pre-deployment evaluation of radiology AI models to guide purchasing decisions.
•AuntMinnie
Deep Learning CT Muscle Segmentation Enhances Sarcopenia Detection in COPD
A deep learning model for CT-based 3D pectoralis muscle segmentation outperforms standard 2D techniques in assessing muscle loss among COPD patients.
•AuntMinnie
Stanford Launches Merlin: 3D AI Model for Abdominal CT Interpretation
Stanford researchers introduce Merlin, a 3D vision-language AI model for interpreting abdominal CT scans, demonstrating strong performance across multiple radiology tasks.