iCAD ProFound AI

ProFound Detection runs while the radiologist reads, not as a separate queue. It highlights malignant soft tissue densities and calcifications on each tomosynthesis slice or 2D image and scores both the lesion and the case. iCAD trained the model on more than 20 million images, 88,000 cases, and 13,500 biopsied cancers from 130 contributing centers. The practical effect is a shorter first pass: the eye goes to the flagged region instead of scanning every slice cold.

Every study gets a score, and the scorecard shows which findings rank highest. That ranking is the part radiologists use when time is short, because a case with one high-scoring lesion gets attention before a case with several low ones. In the V3 service pack iCAD added color coding and a cap on how many marks appear per 3D view, so a noisy study doesn't bury the reader in boxes.

Version 4.0 added two changes aimed at precision. A prior mammogram can be pulled into the analysis of the current one, which cuts the number of marks that turn out to be nothing. The algorithm also correlates the CC and MLO views of the same lesion, so a finding that shows up in both views scores differently from one that doesn't. iCAD reports better accuracy on hard cases compared with V3, and fewer false positives alongside it.

PowerLook Density Assessment measures how much fibroglandular tissue a breast contains and maps the result to one of the four BI-RADS 5th edition categories. That matters because visual density calls vary widely between radiologists, and the variation drives inconsistent supplemental screening advice. Automating the category makes reporting repeatable year to year and supports the dense breast notification laws now in force in many US states.

ProFound AI Risk estimates the likelihood of a breast cancer diagnosis within one or two years using the mammogram plus the patient's age. It reads density, microcalcifications, tissue masses, and left-right asymmetries, then compares those against incidence tables from 15 countries. iCAD claims up to 2.4 times the accuracy of questionnaire models such as Tyrer-Cuzick and Gail. Treat that figure as the vendor's own comparison, and note the regulatory gap: no FDA clearance, US investigational use only.

ProFound Cloud is the hosted version of the suite, launched in mid 2024 as a software-as-a-service option. It runs detection, density, and risk from the same platform and removes the need to maintain AI servers inside the hospital network. For smaller imaging groups that can't justify on-premise hardware, this is the realistic entry point. Larger hospital systems often keep the on-site install for data residency reasons.

The software is built to sit inside the reading workflow rather than beside it. It pulls studies from PACS, returns results to the workstation, and works across mammography systems from several manufacturers instead of locking a site to one vendor. Integration effort varies with the PACS and the age of the reading stations, which is why iCAD sells through a demo and implementation process rather than a download.