Visage Breast Density

Visage Breast Density is an image processing software that employs a trained convolutional neural network to automatically assess breast density from mammography and digital breast tomosynthesis images, categorizing them according to the ACR BI-RADS Atlas 5th Edition. It serves as an adjunctive tool for radiologists, integrating with the Visage 7 platform to aid in the evaluation of breast tissue composition, with performance validated through comparison to radiologist consensus and a predicate device.

Visage Breast Density is a software application designed to assist radiologists in assessing breast tissue composition by automatically evaluating breast density from mammography studies. It utilizes a convolutional neural network (CNN) that has been trained on a substantial database of mammography exams. When applied to a mammography image, the CNN calculates likelihoods for four distinct breast density categories (A, B, C, or D) as defined by the ACR BI-RADS Atlas 5th Edition. The classifications from individual images within a study are then consolidated to provide an overall density assessment for the entire mammography study. This software functions as an add-on module for the Visage 7 medical imaging platform, which handles the distribution, viewing, processing, and archiving of medical images. The breast density assessment is performed on mammography studies stored on the Visage 7 server, with the results displayed via the Visage 7 client and stored back on the server. Indicated for use with compatible full-field digital mammography and digital breast tomosynthesis systems, Visage Breast Density provides adjunctive information and is not intended as a standalone diagnostic tool. Performance testing against two independent datasets, using consensus from board-certified radiologists as ground truth, demonstrated that the software achieves accuracies comparable to a predicate device for both categorical and binary (dense vs. non-dense) classifications.

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. What specific architectural details of the convolutional neural network (CNN) contribute to its accuracy in classifying breast density categories, and how was the training dataset curated to ensure robustness across diverse patient populations?; How does the software handle variations in image quality or acquisition parameters from different mammography and digital breast tomosynthesis systems when assessing breast density?; What is the quantitative performance (e.g., sensitivity, specificity, AUC) of Visage Breast Density for each of the four BI-RADS categories and for the binary 'dense' vs 'non-dense' classification, beyond 'similar accuracy' to the predicate?; How does the merging algorithm combine individual image classifications into a general classification for the entire mammography study, especially in cases of intra-study density heterogeneity?; What impact does the integration with the Visage 7 system have on workflow efficiency for radiologists, and are there any limitations in its use as purely adjunctive information?