MammoScreen 2.0

MammoScreen 2.0 is an AI-powered software utilizing deep learning algorithms, trained on extensive mammography datasets, to assist radiologists in detecting and characterizing suspicious findings like soft tissue lesions and calcifications in both FFDM and DBT images. It aims to improve diagnostic accuracy by providing a quantifiable suspicion score and visual aids, with clinical studies demonstrating enhanced radiologist performance (AUC) when using the device as a concurrent reading aid.

MammoScreen 2.0 is a computer-assisted detection and diagnosis software designed as a concurrent reading aid for physicians interpreting screening mammograms, including both Full-Field Digital Mammography (FFDM) and Digital Breast Tomosynthesis (DBT). Its purpose is to help identify potential soft tissue lesions and calcifications and assess their level of suspicion. The software automatically processes the four standard mammographic views, generating a report with a "MammoScreen score" from one to ten for individual findings, each breast, and the overall exam, where higher scores indicate increased suspicion. Technologically, MammoScreen 2.0 employs medical image processing and machine learning, specifically deep learning modules trained on extensive databases of biopsy-proven cancers and normal tissue, to recognize suspicious features. This version (2.0) updates the algorithmic components of its predicate for improved FFDM accuracy and to enable DBT image processing. Clinical validation for MammoScreen 2.0 involved two multi-reader multi-case (MRMC) studies, one for FFDM and one for DBT, comparing radiologist performance with and without the software. For FFDM, utilizing data from the predicate device (MammoScreen 1.0) to which MammoScreen 2.0 demonstrated non-inferior standalone performance, radiologists using the software improved their average Area Under the ROC Curve (AUC) from 0.77 to 0.80. The standalone FFDM software performance (AUC 0.79 for version 1.0) was non-inferior to unaided radiologists (AUC 0.77). For DBT, radiologists using MammoScreen 2.0 improved their average AUC from 0.79 to 0.83. The standalone MammoScreen 2.0 performance on DBT (AUC 0.84) was found to be superior to the average performance of unaided radiologists (AUC 0.79). The software is intended to provide interpretable results, with suspicion scores traceable from the exam level down to specific findings.

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 "MammoScreen score" calibration (1-10 scale) correlate with specific BI-RADS categories or probabilities of malignancy?; What were the specific sensitivities and specificities achieved by radiologists with and without MammoScreen 2.0 in the FFDM and DBT studies, beyond the overall AUC improvement?; The document mentions "large databases" for training deep learning modules; what is the diversity (e.g., ethnicity, breast density, equipment vendors) of these training datasets, and how might this impact generalizability?; How does MammoScreen 2.0 handle images from different mammography system vendors, and what are the specific "compatible mammography systems" mentioned in the indications for use?; What is the average processing time per case for MammoScreen 2.0 for FFDM and DBT modalities, and how might this impact clinical workflow?