|
Mammo AI Momentum Builds, CAC Guidelines, and GE Finalizes Intelerad
March 19, 2026
|
|
|
|
|
Together with
|
|
|
|
“What stands out here is that the impact doesn’t come from replacing radiologists, but from re-designing the screening workflow around AI support. In many healthcare applications, the biggest gains appear when AI helps prioritise cases, reduce workload and surface subtle findings earlier.”
|
|
Nuno Silva of UnifAI Technology, on recent studies of mammography AI.
|
|
|
Radiology operations face a compounding problem: the coordination work surrounding every scan runs almost entirely on people, a manual burden that doesn’t scale. Could agentic AI help? This webinar on Wednesday April 22 at 12 pm ET examines how agentic AI takes on that coordination burden – and what changes across radiology organizations when it does.
|
|
|
|
Momentum is building toward routine clinical use of AI for breast cancer screening. Several new studies offer even more support for mammography AI, including research published today in Nature Medicine in which AI reduced radiologist workload by over 60% by excluding low-risk studies from human review.
Breast screening has become one of the most promising use cases for AI, with the potential to reduce radiologists’ workload while improving their ability to detect cancer.
- For example, the recent MASAI study found that ScreenPoint Medical’s Transpara AI algorithm could replace the second human reader in a double-reading protocol, reducing workload by 44% and improving cancer detection rates by 28%.
The new research in Nature Medicine also used Transpara, as part of the AITIC study in Spain with the goal of seeing if AI could triage low-risk studies so they don’t require review by human radiologists.
- AITIC had a prospective design, involving 31k women with screening exams split between 2D mammography (17k) and digital breast tomosynthesis (14k).
Women in the control arm of the study got conventional double reading by two radiologists – the standard mammography paradigm in Europe.
- The intervention arm used a partially autonomous AI approach: cases that AI interpreted as low risk were classified as normal and were not reviewed by radiologists, while all other cases were double-read by radiologists using AI support.
In analyzing the results, researchers found…
- Workload in the AI arm was 64% lower than conventional double reading.
- AI’s workload reduction was similar between DBT and conventional digital mammography (-66% and -62%, respectively).
- The AI arm’s cancer detection rate per 1k women was 15% higher (7.3 vs. 6.3 cancers).
- But the recall rate was also 15% higher.
It’s worth noting that the AITIC study differed from MASAI in its inclusion of DBT screening exams, whereas MASAI only included 2D digital mammography.
- While 2D mammography is the norm in Europe, much of the U.S. has switched to DBT for breast screening, so the AITIC results offer good news for U.S. breast imaging practices considering AI adoption.
The Takeaway
The AITIC study’s new results are powerful confirmation of findings from the recent MASAI trial and support broader clinical deployment of mammography AI. Taken together with positive findings from last week’s Nature Cancer articles (see The Wire section in this newsletter), they paint a picture of a technology that’s ready for prime time.
|
|
Transform Your Workflow with Intelligent Imaging
Radiology departments face growing challenges: higher workloads, fewer staff, and tighter budgets. Discover how intelligent imaging technology from Siemens Healthineers streamlines processes, reduces stress, and improves outcomes. Access the white paper now and transform your workflow.
|
|
A New Solution for Radiology Reporting
Sectra Reporting delivers a fully built-in, AI-enhanced reporting experience that seamlessly connects PACS imaging data and measurements through intuitive templates. Discover how it can meet your needs today.
|
|
New-Generation Platform for Managing Multi-Omics Data
QP-Insights from Quibim is a new-generation advanced platform for the management, storage, and analysis of large-scale multi-omics data and medical images for clinical studies and research projects. Learn more on this page.
|
|
- More on Mammography AI: Several important papers in Nature Cancer focused on mammography AI last week. In the GEMINI study, NHS researchers tested 17 different workflows with Kheiron Medical Technologies’ Mia algorithm (now part of DeepHealth) in 10.9k women. The optimal workflow combined AI triage of normal cases, while for the rest, AI assisted during live reading and flagged suspicious cases not recalled by human readers. This workflow improved cancer detection by 10% compared to human double reading while reducing workload by 31%, with slightly lower recall rates (-0.8%).
- Google AI for Mammography: In another Nature Cancer study, a mammography AI algorithm developed by Google turned in good performance when used to replace a second human reader in a study of 45.6k women in the NHS system. Researchers found that the combination of a human reader with AI had slightly higher sensitivity than two human double readers (49% vs. 48%) and comparable specificity (both 97%). More importantly, AI reduced by 46% the number of screening exams read by humans.
- AI Deployment Recalibration: The third study in Nature Cancer also used Google’s mammography AI algorithm, this time in the AIMS trial in the U.K. The study included retrospective (116k women) and prospective arms (9.3k women), and in the retrospective arm AI had higher sensitivity than a first reader (0.54 vs. 0.44) with comparable specificity (0.94 vs. 0.95), and a 24% higher cancer detection rate per 1k women (9.33 vs. 7.54). The prospective phase revealed deployment challenges (higher recall rate) that required recalibrating AI sensitivity.
- BRAIx Predicts Breast Cancer Risk: Breast cancer risk prediction is another promising use case for AI, and researchers describe their use of the BRAIx AI Reader algorithm in a new paper in The Lancet Digital Health. BRAIx was developed by a consortium of Australian universities, and the study tested its prognostic power over four years in 1.6k women getting screening mammograms in Australia and Sweden. BRAIx was more accurate in predicting risk than conventional risk factors like breast density, and could be used to personalize screening.
- DBT AI Training Dataset Released: A new open-source dataset for training AI algorithms to analyze digital breast tomosynthesis images has been released. The dataset is a collaboration between iMerit, Segmed, and Advocate Health and includes imaging studies from 558 women, with associated longitudinal and outcomes data, split almost evenly between malignant and benign cases. Average tumor size is 1.34 cm, which makes the dataset suitable for training AI algorithms to detect early-stage cancer.
- GE Completes Intelerad Acquisition: GE HealthCare completed its $2.3B acquisition of Intelerad this week. The acquisition was announced in November 2025 and is part of GE’s strategy to expand its position in cloud-based enterprise image management, especially in the outpatient sector. GE estimates that Intelerad will deliver $270M in revenues in its first year and will grow at a rate in the low double digits thereafter. Intelerad will operate as part of GE’s Imaging business.
- FDA Extends Review of Lantheus Radiotracer: The FDA notified Lantheus that it needs another three months to complete its regulatory review of a PET radiotracer the company is developing to image somatostatin receptor-positive neuroendocrine tumors. The agency is reviewing LNTH-2501 for NET imaging, and now projects the review to extend to late June. The extension relates to manufacturing data submitted by Lantheus and is not connected to the radiotracer’s safety or efficacy.
- AI Guides Appropriate Imaging: Could AI be used to help referring physicians order more appropriate imaging exams – and cut healthcare costs in the process? Researchers from China describe their AMIR-GPT model, which they developed based on ACR Appropriateness Criteria to score imaging exam appropriateness. In tests against general-purpose large language models like ChatGPT, AMIR-GPT scored highest for agreement with ACR guidelines, but with an accuracy score of 33%, it could use some fine-tuning to improve performance.
- New Statin Guidelines Reinforce CAC’s Value: New ACC/AHA lipid guidelines released last week on the use of statins to prevent heart disease reinforce the role of CT-derived coronary artery calcium scores for assessing cardiac risk. The guidelines show that CAC scores have evolved from decision aids to being “embedded in the framework of risk assessment, treatment initiation, and treatment intensity,” as one cardiologist wrote. CAC has moved “from the margins of guidelines to the center of preventive cardiology.”
- AI Helps Triage MSK Fractures: Clinicians in Norway used Gleamer’s BoneView AI algorithm to help radiographers (radiologic technologists) triage emergency musculoskeletal X-rays for immediate radiologist review. Researchers describe the workflow in a new study in European Journal of Radiology with 1.2k patients, in which initial AI evaluations were reviewed by radiographers, who could either discharge patients or escalate cases for radiologist review. Standalone AI had sensitivity of 0.95 and specificity of 0.90, with radiographers overriding AI results in 19% of cases.
- Which Lung Nodules Are Most Dangerous? A new study in CHEST offers guidance on which nodules on CT lung cancer screening scans are most likely to be malignant. Researchers analyzed 26.4k patients from the NLST study who underwent baseline scans and one round of follow-up. Compared to patients with pre-existing nodules that didn’t grow, those with new nodules had almost 4X the odds of a lung cancer diagnosis in two years (OR = 3.9) and those with pre-existing nodules that were growing had nearly 20X the odds (OR = 19.7).
- Improved Sybil AI Better Predicts Lung Cancer: A new version of the Sybil deep learning model that included epidemiological data showed better accuracy for predicting lung cancer risk from low-dose CT scans. In a new study in CHEST, researchers tested Sybil-Epi on 22.5k people who received LDCT baseline screening scans, finding the newer model had better overall performance than the original Sybil after six years of follow-up (AUC = 0.83 vs. 0.80), with much better performance for patients with no detected nodules (0.76 vs. 0.64).
- Lung Cancer Screening to Start in Germany: The long-awaited start of Germany’s national CT lung cancer screening program is set for April 1, making the country the latest nation to adopt population-based lung screening. The program covers active and former heavy smokers ages 50 to 75, who can receive low-dose CT scans every 12 months. The scans are eligible for reimbursement under a new health insurance benefit, but reimbursement levels still haven’t been set yet.
- Stanford’s 3D CT Vision Language Model: Most vision language models to date have focused on analyzing 2D medical images, but Stanford University researchers (including Cognita Imaging founder Louis Blankemeier, PhD) developed a VLM called Merlin that’s able to analyze 3D CT scans. They describe their work in a paper published in Nature, in which the VLM was tested on 44.1k CT scans and completed tasks ranging from classification of findings and phenotypes to five-year risk prediction and report generation.
- HOPPR Adds NVIDIA AI Models: Foundation model developer HOPPR added tools from NVIDIA to their AI Foundry algorithm development environment. AI Foundry users now have access to NVIDIA’s NV-Reason tool for generating structured, analytical reasoning alongside model output, as well as NV-Generate for creating synthetic DICOM imaging datasets to support AI algorithm development. HOPPR launched AI Foundry at RSNA 2025 as a secure platform for developing radiology AI applications.
- NVIDIA Models Aid Philips MRI: NVIDIA’s accelerated computing hardware and AI development tools are helping Philips build more powerful MRI applications. The companies began working together in 2025 to create an MRI foundation model that can perform image generation, segmentation, and interpretation as part of a single intelligent workflow. The solution would operate with NVIDIA’s NV-Generate, NV-Segment, and NV-Reason models, and would perform tasks such as predictive image preview that would enable clinicians to see an AI-generated preview of a patient image before the scan begins.
- Strings Adds Agentic AI for Radiology Workflow: 3DR Labs’ Strings subsidiary added agentic AI features to its flagship Strings radiology workflow and automation software. The new features include multiple enhancements to help radiology practices work more efficiently, such as exam routing, eliminating manual data entry and search tasks, and orchestrating micro-workflows. 3DR Labs acquired Strings in 2025.
|
|
Bring Your Radiology AI into Your Clinical Workflows
CARPL enables healthcare providers and researchers to develop, test, and deploy their own AI models within existing clinical infrastructure. From seamless data ingestion and de-identification to model training, packaging, and live deployment, CARPL provides an end-to-end environment tailored for radiology.
|
|
Radiology Case Report
A man in his 40s presented with a known metastasis within his abdomen. Learn how contrast-enhanced MRI helped to diagnose the extent of his disease.
|
|
From Image Exchange to Enterprise Interoperability
Easily share, access, and exchange any medical image with solutions from Medicom. Deliver critical studies directly into the EHR to ease clinician burden, simplify your tech stack, and unlock the research potential hidden in clinical imaging data.
|
|
- Cardiac CT Online Training Course: Medality’s online cardiac CT training course is designed for busy clinicians working towards Level 2 Cardiac CT certification. Endorsed by SCCT, it provides a flexible, practice-focused approach to develop your coronary CTA interpretation and reporting skills while earning CME.
- How CHU Bordeaux Integrated AI into Their Echo Lab: Bordeaux University Hospital evaluated the real-world use of AI in echocardiography from Us2.ai. Results showed strong agreement between AI-generated and human measurements, particularly for ejection fraction and Doppler-based parameters, highlighting AI’s potential to streamline workflow and reduce variability.
- Enterprise Imaging Done Differently: Legacy radiology solutions were not designed to carry healthcare organizations into the future. From their first line of code, Mach7 Technologies was designed to meet the imaging needs of the entire healthcare enterprise. Learn more about their unique approach today.
- Mastering Enterprise Imaging in the Cloud: Reserve your seat at this March 31 webinar hosted by Merge and SIIM to learn how VNAs serve as the backbone of enterprise imaging and cloud strategies by enabling seamless data orchestration, accessibility, and management across diverse systems and modalities.
- Elevating Breast Cancer Detection: Breast Suite from DeepHealth is a new package of AI-powered solutions delivering increased breast cancer detection rates, risk stratification tools, and viewing and reporting workflow acceleration. Find out how it can benefit your practice today.
- Why Radiology Reporting Needs a Reset: Radiologists are under growing pressure, yet many reporting tools still slow them down. Kailo Medical believes reporting should support clinical thinking, not add to the workload. Discover how their KailoAir solution can help you reset your reporting.
- Experience the Next Evolution: From Day 1, AGFA HealthCare has understood that striking the critical balance between clinical efficiency and quality patient care starts with the physician experience. Learn about their history of firsts on this page.
- Exploring AI Readiness in Diagnostic Imaging: In collaboration with AHRA, Philips surveyed imaging leaders to assess AI adoption, value, and barriers. The State of AI in Diagnostic Imaging whitepaper delivers expert insights and practical guidance to move AI from promise to real-world clinical impact. Check out the white paper today.
- A Passion for Change: United Imaging’s passion for change was on display at RSNA 2025 with the launch of new products across multiple modalities, including the new uSonique ultrasound family shown as works-in-progress. Find out what drives the company on this page.
- The Power of a Smooth Go-Live: Don’t gamble on your healthcare institution’s go-live: take control of your PACS migration with ENDEX from Enlitic. Discover how ENDEX uses AI to standardize, normalize, and cleanse your imaging metadata before migration.
- Future-Proofing Your Reporting: The PS360 end-of-life is approaching – is your practice ready? A mandatory legacy migration doesn’t have to disrupt patient care. Join Rad AI next week, March 25, to navigate the hidden operational hurdles and secure your roadmap for a safe, supported transition. Register now.
- Interpretation Efficiency in Radiology – A Critical Strategy: Healthcare institutions are at a critical stage, where an emphasis on interpretation efficiency needs to be a priority. Check out this white paper from Visage Imaging and Signify Research on strategies to optimize your interpretation efficiency.
- Driving Efficiency in Radiology: Radiology’s workforce is under pressure. Discover evidence-based strategies for driving efficiency without burning out your staff in this e-book from Riverain Technologies.
- A New Era of Imaging Technology: MosaicOS is the cloud-native and AI-native operating system from Mosaic Clinical Technologies designed to expand capacity, cut reporting time, and deliver faster, smarter patient care. Discover how it can improve your radiology operations today.
- Low Dose, Remarkably Open Design: The Scenaria View CT scanner from Fujifilm Healthcare Americas is a powerful premium CT solution that provides dependable routine application capabilities, while coronary artery imaging is improved with the Focus Edition’s optional Cardio StillShot 3D motion correction technology.
- Reimagining Cloud-Native Cardiology Workflow: Find out how Intelerad’s next-generation cloud-based InteleHeart solution delivers an all-in-one cardiology platform that unifies viewing, reporting, analytics, and workflow orchestration.
|
|
|
|
|
