HealthMammo

HealthMammo is an artificial intelligence-powered software designed to analyze 2D FFDM screening mammograms from validated systems, identify suspicious findings, and notify clinicians to enable worklist prioritization. Its methodology involves parallel processing of mammography studies using an AI algorithm, and its clinical utility is supported by performance data demonstrating high sensitivity, specificity, and an AUC of 0.9661 in a retrospective study.

HealthMammo by Zebra Medical Vision is a software solution employing an artificial intelligence algorithm to analyze 2D Full Field Digital Mammography (FFDM) screening exams from validated FFDM systems. Its purpose is to assist MQSA-qualified interpreting physicians by identifying mammograms with suspicious findings at the exam level and flagging them within the Picture Archiving and Communication System (PACS) or a dedicated Zebra Worklist. This notification facilitates prioritization of cases potentially requiring urgent review, operating in parallel with the standard care workflow without altering images or removing cases from the radiologist's worklist. The system includes modules for data input/validation, the core AI algorithm for analysis, PACS/workstation integration, and error reporting, aiming to reduce the time to alert physicians about potentially critical findings, though it does not provide a diagnosis or treatment recommendations. The device's performance was evaluated using a retrospective cohort of 835 anonymized 2D FFDM screening mammograms, encompassing 435 biopsy-confirmed cancer cases and 400 normal cases with two-year negative follow-up, considering confounding factors like lesion type, breast density, age, and histology. HealthMammo demonstrated an overall Area Under the ROC Curve (AUC) of 0.9661. Performance was validated at three operating points: 'standard mode' (sensitivity 89.89%, specificity 90.75%), 'high sensitivity' mode (sensitivity 94.02%, specificity 83.50%), and 'high specificity' mode (sensitivity 84.14%, specificity 94.00%). The average processing time for analysis and notification was reported as 2.9 minutes per study.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the performance of HealthMammo vary across different breast densities, particularly in dense breasts where lesion detection is more challenging?; What is the real-world impact on radiologist workflow efficiency and reading times when HealthMammo is integrated, beyond simple case prioritization?; Are there specific types of suspicious findings (e.g., architectural distortions vs. masses vs. microcalcifications) for which the HealthMammo algorithm demonstrates higher or lower sensitivity and specificity?; How does HealthMammo's performance, potentially validated primarily on Hologic systems as suggested in the technical comparison, generalize to 2D FFDM images from a broader range of FFDM system manufacturers, especially compared to its vendor-agnostic predicate?; What are the computational resource requirements for implementing and scaling HealthMammo in diverse clinical settings, from small clinics to high-volume screening centers?