Multicenter, Multivendor Validation of an FDA-approved Algorithm for Mammography Triage

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Artificial intelligence (AI) algorithms offer potential to enhance cancer detection and optimize radiologist workflows.
• Evaluating AI performance in mammography is crucial for clinical adoption.

Purpose of the Study:

• To assess the performance of a commercial AI-based triage algorithm on screening mammograms.
• To evaluate the algorithm's effectiveness across different breast densities and lesion types.

Main Methods:

• A retrospective study analyzed 1255 screening mammograms using a commercial AI algorithm (cmTriage).
• The AI algorithm flagged exams as "suspicious" or not, with performance measured by area under the curve (AUC), sensitivity, and specificity.
• Data included diverse breast densities and lesion types (masses, microcalcifications).

Main Results:

• The AI algorithm achieved an overall AUC of 0.95 for case identification, consistent across densities (AUC 0.95) and lesion types (masses: 0.94, microcalcifications: 0.97).
• Default sensitivity was 93% with 76.3% specificity.
• Real-world performance testing showed 86.9% sensitivity and 88.5% specificity, comparable to practicing radiologists in the Breast Cancer Surveillance Consortium (BCSC) study.

Conclusions:

• AI-based triage software demonstrates performance on par with practicing radiologists for lesion detection in mammography.
• AI can potentially improve reader specificity and streamline radiologist workflow, leading to faster turnaround times and enhanced patient care.