iCAD ProFound AI Mammography: FDA Clearance History and Device Record

Executive Summary

ProFound Detection V4.0 is the current FDA-cleared generation of iCAD's AI-powered mammography software line, authorized under 510(k) submission K240417 with a decision date of November 8, 2024. The device is classified under FDA product code QDQ — Radiological Computer Assisted Detection/Diagnosis Software For Lesions Suspicious For Cancer — within the Radiology medical specialty. It is indicated for use with digital breast tomosynthesis (DBT) only; the 2D mammography version of ProFound AI is CE Marked but not FDA cleared.

The developer, iCAD, Inc., was acquired by DeepHealth, a wholly-owned subsidiary of RadNet, Inc., on July 17, 2025. As of Q2 2026, the product line operates under the DeepHealth brand. The installed base covers more than 1,500 healthcare provider locations across 50+ countries, with an estimated 10 million or more mammograms processed annually under the DeepHealth OS platform.

This profile consolidates the complete multi-generation FDA clearance record for the ProFound AI / ProFound Detection line — including K-numbers, decision dates, submission types, product codes, AI architecture details, and version-specific performance claims — in a single structured reference. Vendor press releases cover individual clearance events in isolation; this record is intended to serve procurement evaluation, regulatory analysis, and clinical due diligence.

Developer Background: iCAD and the DeepHealth/RadNet Acquisition

iCAD, Inc. was founded as a computer-aided detection (CAD) company and established an early position in breast imaging AI through its SecondLook Digital platform, which was the first FDA-cleared 2D mammography CAD solution. The company subsequently rebranded its core platform to PowerLook, and the PowerLook Tomo Detection product represented iCAD's entry into AI-powered DBT — the first such system to receive FDA authorization.

Over the following years, iCAD expanded the portfolio beyond cancer detection to include breast density assessment (PowerLook Density), cancer risk evaluation (ProFound AI Risk), and breast arterial calcification assessment. The DBT cancer detection product was rebranded from PowerLook Tomo Detection to ProFound AI for DBT, and subsequently to ProFound Detection with the V4.0 generation.

RadNet announced its intent to acquire iCAD on April 15, 2025, and the transaction was completed on July 17, 2025. DeepHealth — RadNet's digital health subsidiary — serves as the umbrella brand integrating iCAD's AI portfolio with DeepHealth's cloud-native operating system, DeepHealth OS, which is designed to unify data across clinical and operational screening workflows. The iCAD corporate website header confirms the completed integration: "iCAD is now part of DeepHealth." Full brand and product-name consolidation may be ongoing as of the article date.

• SecondLook Digital: first FDA-cleared 2D mammography CAD solution (pre-DBT era)
• PowerLook: rebranded platform name used for the DBT-era product line
• PowerLook Tomo Detection 1.0: first-generation DBT AI for cancer detection, initially a GE-exclusive product
• ProFound AI for DBT: rebranded product name introduced with the December 2018 clearance, continuing through V3.0
• ProFound Detection V4.0: current product name as of the November 2024 FDA clearance
• DeepHealth / RadNet acquisition: completed July 17, 2025; iCAD AI portfolio now integrated into DeepHealth OS

Device Classification and Regulatory Framework

ProFound Detection and its predecessor versions are regulated as Software as a Medical Device (SaMD) under FDA's Center for Devices and Radiological Health (CDRH). The applicable product code is QDQ, which designates Radiological Computer Assisted Detection/Diagnosis Software For Lesions Suspicious For Cancer, under the Radiology medical specialty. This classification applies to both K203822 (V3.0) and K240417 (V4.0), as confirmed in the FDA accessdata records for each submission.

A structurally important distinction runs through the iCAD clearance history: the original PowerLook Tomo Detection was authorized via the Premarket Approval (PMA) pathway, which requires the highest level of FDA review and independent clinical evidence. All subsequent ProFound AI and ProFound Detection versions were authorized via the 510(k) pathway, which demonstrates substantial equivalence to a legally marketed predicate device rather than requiring independent clinical proof of safety and effectiveness. This pathway distinction is central to understanding both the predicate chain for subsequent versions and the evidentiary standard applied at each clearance.

FDA Clearance Timeline: Four Generations

The iCAD DBT AI product line has accumulated four distinct FDA authorization events spanning approximately eight years. Each generation built on the prior clearance as a predicate, enabling the 510(k) pathway for all versions after the initial PMA. The timeline below documents the confirmed regulatory facts; unverified K-numbers are flagged explicitly.

Generation 1: PowerLook Tomo Detection (2016/2017, PMA)

PowerLook Tomo Detection was the first AI-enabled DBT cancer detection system to receive FDA authorization. According to the product table in Potnis et al. (2022), published in JAMA Internal Medicine and indexed at PMC10623674, the clearance year is listed as 2017 and the pathway as Premarket Approval. The study population was multicenter with an enriched dataset and external validation; reported outcomes included sensitivity, specificity, AUC, recall rate, and radiologist reading time.

Generation 2: ProFound AI for DBT V1 (December 2018, 510(k))

FDA clearance for the first ProFound AI for DBT product was announced on December 7, 2018. The clearance was based on a multi-reader, multi-case (MRMC) reader study in which 24 radiologists read 260 tomosynthesis cases with and without ProFound AI assistance. The study reported an 8% average improvement in case-level sensitivity, a 7% average reduction in unnecessary patient recalls, and a reading time reduction of more than 50% on average.

Generation 3: ProFound AI Software V3.0 (K203822, March 2021, Special 510(k))

ProFound AI V3.0 was authorized under 510(k) K203822 as a Special 510(k). The FDA received the submission on December 29, 2020, and issued a decision of Substantially Equivalent on March 12, 2021. The applicant is listed as iCAD, Inc., 98 Spit Brook Rd., Suite 100, Nashua, NH 03062. Product code: QDQ.

Per iCAD's V3.0 clearance announcement, the model was developed using more than five million images from 30,000 cases, including approximately 8,000 biopsy-proven cancers. Validation was performed on approximately one million images from 3,500 cases, including 1,200 biopsy-proven cancers. iCAD reported, based on internal data, improvements of up to 10% in specificity and up to 1% in sensitivity compared to prior versions, along with up to 40% faster processing on the PowerLook platform.

Generation 4: ProFound Detection V4.0 (K240417, November 2024, Traditional 510(k))

The current cleared version, ProFound Detection V4.0, was authorized under K240417 as a Traditional 510(k). The FDA received the submission on February 13, 2024, and issued a decision of Substantially Equivalent on November 8, 2024. The applicant is iCAD, Inc. Product code: QDQ. Regulation Medical Specialty: Radiology. The submission was not reviewed by a third party and is not a combination product.

V4.0 introduces two notable architectural additions: training with advanced deep learning CNNs and an option to incorporate a prior exam into the current analysis — a feature absent in V3.0. iCAD's internal performance data, corroborated across multiple trade press outlets that published the same press release, reports a 6.3% improvement in AUC over V3.0.

AI Architecture and Training Data by Version

All generations of the ProFound AI / ProFound Detection line use deep learning convolutional neural networks (CNNs) as the core architectural approach. The algorithm is trained to detect malignant soft tissue densities and calcifications within DBT image stacks. Algorithm outputs include:

• Certainty of Finding scores: Assigned to individual detected findings, indicating the algorithm's confidence that a specific region of interest represents a malignant lesion.
• Case Scores: A 0–100% scale value representing the algorithm's overall confidence that the case contains a malignancy. This score is presented at the case level, not the finding level.

Training dataset scale increased substantially across generations. V3.0 was trained on more than five million images from 30,000 cases, with approximately 8,000 biopsy-proven cancers, and validated on approximately one million images from 3,500 cases including 1,200 biopsy-proven cancers. V4.0 was described by iCAD as using "advanced deep learning CNNs" without a specific training dataset count disclosed in publicly available press materials. The prior-exam integration feature introduced in V4.0 allows the algorithm to incorporate longitudinal imaging context into its analysis, representing a qualitative architectural advance over prior versions.

Reported Clinical Performance by Clearance Version

Performance figures vary in their evidentiary basis across versions. V1/V2 figures derive from a named peer-reviewed reader study (Conant et al. 2019, Radiology: Artificial Intelligence). V3.0 and V4.0 figures are sourced from iCAD internal data and have not been independently peer-reviewed. All figures should be interpreted in the context of the retrospective, enriched-dataset design used for FDA clearance submissions — see the Evidentiary Limitations section.

Two conference presentations cited by third-party sources reported real-world performance data: Madden et al. (SBI 2023) reported a 63% cancer detection rate increase for general radiologists, and Schilling et al. (ECR 2023) reported a 23% improvement for breast subspecialist radiologists. These figures originate from conference abstracts, not peer-reviewed publications, and were cited on a vendor distribution partner page. They should be interpreted as preliminary real-world signals rather than validated clinical evidence.

Intended Use and Indications for US FDA Clearance

All FDA clearances across all four generations apply exclusively to the digital breast tomosynthesis (DBT) version of the product. ProFound AI for 2D Mammography — a separate product in the iCAD portfolio — is CE Marked for use in Europe but is not FDA cleared for use in the United States. This distinction is confirmed by multiple sources including the Blackford Analysis product page and iCAD press materials.

The clinical rationale for focusing the US clearance on DBT is that tomosynthesis has become the preferred breast cancer screening modality in the United States, with DBT offering improved lesion visibility compared to conventional 2D digital mammography. The cleared indication — Radiological Computer Assisted Detection/Diagnosis Software For Lesions Suspicious For Cancer (product code QDQ) — applies to both malignant soft tissue densities and calcifications within DBT image stacks.

Evidentiary Limitations of the Clearance Record

Every FDA authorization across all four ProFound AI / ProFound Detection generations was granted on the basis of retrospective reader studies using enriched datasets — datasets in which the proportion of cancer cases is deliberately elevated above what would be encountered in routine screening populations. No prospective clinical outcomes — including cancer stage at detection, interval cancer rates, or mortality — were required or reported for any version.

This is not unique to iCAD. A 2022 analysis published in JAMA Internal Medicine (Potnis et al., PMC10623674) reviewed nine AI products for breast cancer screening cleared or approved by FDA between January 1, 2017 and December 31, 2021. The review found that all nine products were cleared on the basis of previously collected, retrospective datasets; seven used enriched datasets; and none reported clinical outcomes such as cancer stage, interval cancer detection, or mortality. PowerLook Tomo Detection (iCAD) is listed in the product table as cleared in 2017 via PMA, with a multicenter enriched dataset and external validation, reporting sensitivity, specificity, AUC, recall rate, and radiologist reading time as outcomes.

The JAMA Internal Medicine paper notes that enriched datasets can produce AUC values higher than what would be observed under real-world screening conditions, where cancer prevalence is substantially lower. This structural limitation applies to the performance figures cited across all four clearance generations in this profile.

Current Deployment Context and Recent Developments

Following the July 2025 acquisition, the iCAD portfolio operates within RadNet's DeepHealth OS platform. The combined entity reports deployment across more than 1,500 healthcare provider locations in more than 50 countries, with an estimated 10 million or more mammograms processed annually. In the five years prior to the V4.0 clearance, iCAD estimated that over 40 million mammograms had been read using its AI, with nearly 30% being tomosynthesis cases.

Several partnerships were announced in the months preceding the acquisition completion:

• Microsoft Precision Imaging Network (PIN) collaboration (announced April 29, 2025): Offers radiology providers automated patient reporting on mammography results through integration with Nuance PowerScribe.
• RamSoft partnership (announced March 5, 2025): Commercial preferred distributor and integration partnership across North America.
• Koios Medical partnership (announced February 26, 2025): Multi-modality AI suite for breast cancer detection, extending the product ecosystem beyond DBT.

Clearance Record Summary Table

The table below consolidates the complete FDA authorization record for the iCAD DBT AI product line across all four generations. Unverified K-numbers are flagged; readers requiring confirmed identifiers for those versions should consult FDA accessdata directly.