ProFound AI Software V2.1

ProFound™ AI Software V2.1 is a computer-assisted detection tool that employs deep learning algorithms to identify suspicious soft tissue densities and calcifications in digital breast tomosynthesis images, providing radiologists with visual cues and quantitative scores indicating the algorithm's confidence of malignancy. Its methodology involves processing 3D DBT slices for pattern recognition of lesions, with its scientific validation demonstrated through non-clinical performance studies confirming efficacy and non-inferiority across various DBT systems, including newly supported Siemens platforms.

ProFound™ AI Software V2.1 is a radiological computer-assisted detection and diagnosis (CAD) tool designed to aid physicians in interpreting digital breast tomosynthesis (DBT) exams. The software utilizes deep learning algorithms for pattern recognition to identify and mark suspicious soft tissue densities (such as masses, architectural distortions, and asymmetries) and calcifications within 3D DBT slices. It provides visual overlays on the images and assigns two types of scores: a 'Certainty of Finding' score for each detected lesion and a 'Case Score' for the overall exam, both ranging from 0 to 100, reflecting the algorithm's confidence in the malignancy of a detection or case, respectively, to guide radiologists in their assessment. A key update in version 2.1 is the expansion of compatibility to include Siemens Mammomat Inspiration and Revelation DBT systems, alongside previously supported Hologic and GE systems, while maintaining the core lesion detection technology and scoring system of its predecessor. Non-clinical performance for these new Siemens compatibilities was evaluated through standalone studies using Siemens Standard and Empire Reconstruction DBT image datasets, which demonstrated that case-level sensitivity, lesion-level sensitivity, false positive rates, and specificity met predefined design specifications. Furthermore, comparative non-inferiority testing against its performance with Hologic DBT systems showed that ProFound AI V2.1 maintains comparable efficacy (in terms of sensitivity, Area Under Curve (AUC), and False Positives Per Image volume (FPPI)) when used with the newly supported Siemens systems. It is important to note that the Certainty of Finding and Case Scores are relative, based on a model population with 50% cancer prevalence, and are intended as a relative level of concern rather than absolute probabilities of malignancy for individual patients in typical screening populations.

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 'Certainty of Finding' score calibration, based on a 50% cancer prevalence model, impact its practical utility and interpretation in real-world screening populations where prevalence is significantly lower?; What specific deep learning architectures are employed in the ProFound AI V2.1 algorithm, and how were they optimized for detecting diverse lesion types such as masses, architectural distortions, asymmetries, and calcifications?; Beyond sensitivity, specificity, AUC, and FPPI, were other performance metrics like radiologist reading time reduction or impact on biopsy rates assessed in the non-clinical studies for the newly supported Siemens DBT systems?; The documentation mentions a 'stratified bootstrap procedure' for performance estimation; could you elaborate on the stratification variables used and how this procedure addresses potential biases in the test datasets for Siemens systems?; How does the ProFound AI V2.1 software account for variations in image acquisition protocols, image quality, or reconstruction parameters across the different compatible Digital Breast Tomosynthesis systems (Hologic, GE, Siemens) to ensure consistent detection performance?