MammoScreen BD

MammoScreen BD is an AI-powered software designed to classify breast tissue composition into ACR BI-RADS 5th Edition density categories from full field digital mammography and digital breast tomosynthesis images. Its methodology involves leveraging artificial intelligence to assist interpreting physicians, validated by high agreement with radiologist consensus, and it incorporates a Predetermined Change Control Plan for ongoing algorithmic improvements and expanded compatibility.

Therapixel's MammoScreen BD, an artificial intelligence-powered software application, has received market clearance by demonstrating substantial equivalence to an existing device. This application is designed to analyze breast tissue composition and assign breast density categories according to the ACR BI-RADS 5th Edition, intended for use by interpreting physicians as supplementary information for asymptomatic women aged 40 and older undergoing full field digital mammography and digital breast tomosynthesis screening. It is crucial to note that MammoScreen BD is not intended as a standalone diagnostic tool or the sole basis for patient management decisions. The determination of substantial equivalence was supported by the review and clearance of the device's Predetermined Change Control Plan (PCCP), which enables future algorithm improvements and expanded compatibility without requiring new premarket notifications, provided changes adhere to the plan. Performance testing confirmed the device's safety and effectiveness, showing high agreement with radiologist consensus for breast density classification, evidenced by a quadratically-weighted Cohen’s kappa of 89.03 for the four-class task. Further validation included assessing consistency across different image views, reproducibility between modalities, and reliable performance across various patient subgroups, including age, breast thickness, and race.

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 AI algorithm specifically process mammographic images to determine breast density, and what are the underlying deep learning architectures or methodologies employed?; Beyond the high Cohen's kappa, what were the specific false positive and false negative rates for each breast density category, and how might these impact clinical decision-making?; What is the long-term impact of integrating MammoScreen BD into clinical workflows on radiologist efficiency and potential reduction in inter-reader variability for breast density assessment?; How will the Predetermined Change Control Plan specifically allow for advancements, such as integrating new imaging modalities or adapting to evolving BI-RADS criteria, while maintaining regulatory compliance?; Were there any specific patient subgroups (e.g., extremely dense breasts, specific races/ethnicities not explicitly mentioned) where the device's performance showed noteworthy deviations or limitations?