ProFound AI Software V3.0

ProFound AI® Software V3.0 is an advanced computer-assisted detection and diagnosis tool that utilizes deep learning to analyze digital breast tomosynthesis images, identifying and scoring potential malignant soft-tissue densities and calcifications to support radiologists. Non-clinical studies for this version demonstrated improved specificity on Hologic and GE modalities and non-inferior overall performance compared to its predecessor, indicating its utility in enhancing breast cancer detection.

ProFound AI® Software V3.0 is a radiological computer-assisted detection and diagnosis (CAD) software designed to aid interpreting physicians during the concurrent review of digital breast tomosynthesis (DBT) examinations. Its primary function is to identify suspicious soft tissue densities (masses, architectural distortions, asymmetries) and calcifications within 3D DBT images by employing deep learning algorithms and pattern recognition. The software marks the contours of detected potential lesions on tomosynthesis slices and provides two types of scores: a "Certainty of Finding" score (0-100%) for each detected finding and a "Case Score" (0-100%) for the entire study. These scores indicate the algorithm's confidence in the presence of malignancy, derived from a model based on 50% cancer prevalence, and serve as relative indicators of suspicion rather than absolute probabilities of malignancy in typical screening populations. This version, V3.0, features software enhancements that led to a significant increase in specificity when used with GE and Hologic DBT modalities, as demonstrated in non-clinical validation testing compared to ProFound AI V2.0. Standalone studies also concluded that ProFound AI V3.0's performance (in terms of case sensitivity, false positive rate per 3D volume, and Area Under the localized ROC Curve) with Hologic and GE DBT systems is non-inferior to the baseline performance of ProFound AI V2 with Hologic DBT images. This suggests that the clinical benefits established in the original pivotal reader study for V2 are applicable to V3.0 when used with these specific Hologic and GE DBT systems. The software is compatible with designated DBT systems from Hologic, GE, and Siemens.

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, calibrated on a 50% cancer prevalence population, translate to clinical decision-making in a typical screening environment with much lower cancer prevalence, and what guidance is provided to avoid misinterpretation?; What specific algorithmic modifications in ProFound AI V3.0 led to the significant increase in specificity for GE and Hologic modalities, and were these improvements also observed or tested for the supported Siemens DBT systems?; Given that the Certainty of Finding and Case Scores are generally higher than the actual probability of malignancy in lower prevalence populations, what were the observed positive predictive values of these scores in the validation studies that reflect typical screening scenarios?; The non-clinical validation demonstrated non-inferiority in case sensitivity and FP rate compared to a V2 baseline. What were the absolute sensitivity and specificity figures for ProFound AI V3.0 itself on the Hologic and GE datasets used for validation?; How does ProFound AI V3.0 account for variations in image acquisition protocols, reconstruction algorithms (e.g., standard vs. Empire for Siemens), and image quality across the different compatible DBT systems, and is there data on performance variability between these systems?