Category: Artificial Intelligence

Big Prostate AI Study Supports Screening Role

Posted on by Brian Casey

A new international study testing AI’s performance in analyzing prostate MRI exams could lay the groundwork for the technology’s broader role in prostate cancer screening. In the SCARLET-1 study in Radiology, researchers found that an open-source AI algorithm performed well – and consistently – across different research studies. 

Prostate cancer screening stands on the cusp of population-based adoption, but concerns remain about false positives, particularly how to determine which men should undergo biopsy and which ones can be directed to PSA monitoring. 

  • MRI has proven invaluable for working up patients with elevated PSA levels, but it’s expensive and requires a level of radiologist expertise that’s usually only available at high-volume centers. 

AI could help by automating key parts of the MRI process, enabling the modality to be offered more broadly to support prostate cancer screening.

  • The SCARLET-1 study investigated AI’s potential by combining results from research studies conducted in 22 countries around the world: PRIME, STHLM3-MRI, and IP1-PROSTAGRAM.

All three studies used an updated version of the PI-CAI AI algorithm, developed as part of a challenge sponsored by the Prostate Imaging-Cancer AI consortium in 2024.

  • In all, 2k patients received prostate MRI on 1.5T and 3T scanners, and the PI-CAI algorithm was used to analyze exams at different sensitivity and specificity thresholds to detect cases suspicious for prostate cancer rated at Gleason grade group ≥ 2, the threshold for clinically significant cancer. 

When compared to interpretations by local radiologists, prostate AI…

  • Achieved good AUC across the three studies (AUC = 0.84 to 0.89).
  • Had significantly higher agreement with an external reference standard than local radiologists (13 to 16 percentage points).
  • Maintained a positive net benefit across different thresholds, in particular those intended to reduce biopsies (patient must have > 30% chance of cancer).
  • Predictably, sensitivity thresholds affected AI’s performance, with higher sensitivity detecting more cancers but also generating more biopsies. 

The Takeaway

The SCARLET-1 study confirms that AI of prostate MRI performs as well as radiologists in real-world settings across multiple different locations. That’s important in making prostate cancer screening available in areas where specialty expertise may not be available to interpret imaging exams.

Blackford Finds Buyer with Sopris Acquisition

Posted on by Brian Casey

One of radiology’s longest-running mysteries was resolved late last week when a U.S. private equity firm acquired AI platform company Blackford Analysis from Bayer. Sopris Capital said it would fold Blackford into another radiology AI property it owns, Azra AI, as it builds its healthcare technology portfolio. 

The news comes just over a year after Bayer made the stunning announcement that it was backing away from the AI platform business and would “discontinue offerings and services” in the sector, including Blackford. 

  • The move reversed several years of Bayer’s investment in digital platforms, including the company’s own Calantic Digital Solutions division, which had been formed in 2022. 

The announcement left Blackford twisting in the wind. Some industry experts questioned the long-term viability of the AI platform model, in which platform companies act as middlemen between algorithm developers and end-users, adding a markup that made AI deployment more expensive. 

  • But others still see AI platforms and algorithm orchestration as a necessary foundation for successful AI deployment, especially given the explosion of clinical algorithms and the complexity of IT integration.

For its part, early reports following last year’s Bayer announcement suggested that Blackford’s business model was sound, with the company reporting a strong pipeline of prospective customers. 

  • And despite the departure of key executives like founder Ben Panter, Blackford continued operations under new managing director James Holroyd. The company even exhibited at SIIM 2026 – not exactly a sign of a firm about to go under. And not long after SIIM, Blackford announced a new partnership with Azra AI – perhaps a portent of last week’s announcement. 

So what do we know about Sopris? With its main offices in Colorado and Florida, Sopris first invested in Azra AI in 2022 to help the company acquire Digital Reasoning and expand into oncology. 

Sopris sees the Blackford acquisition as adding an AI orchestration capability to its growing imaging IT portfolio, combining Blackford with Azra AI. 

  • Blackford will remain a separate legal entity with operations in Scotland and the U.S., and will continue serving its existing OEM partners and health system customers “with no disruption to current deployments or operations.” 

The Takeaway

The Sopris acquisition of Blackford means that one of the radiology AI sector’s most established companies will continue operations for the foreseeable future. It also raises the intriguing prospect that a new player is entering the game to drive AI consolidation. 

Is It Time for Mammography AI?

Posted on by Brian Casey

Is AI ready for widespread use in mammography screening? That question was the focus of a point-counterpoint debate in AJR that discussed the pros and cons of broad-based clinical AI deployment for one of radiology’s most high-profile exams. 

Mammography screening is perhaps the clinical use case with the most potential to help radiologists drowning under a rising tide of imaging exams.

  • Several large-scale research studies have been published recently – mostly in Europe – showing that, for most normal mammograms, AI could reduce the need for a second reader, as is common in European screening programs.

Indeed, mammography AI developer Lunit last week announced a contract with a Swedish health system that would do just that, having AI act as an autonomous second reader. 

  • While the U.S. doesn’t typically use second readers for breast screening, AI could still play an important role in assisting radiologists in interpreting mammograms, predicting patient risk, and acting as a sort of clinical spell-check to reduce misses.

So does that mean mammography AI is ready for prime time? Taking the skeptical view were Duke University breast radiologists Eun Langman, MD, and Vilert Loving, MD, who believe caution is warranted, for the following reasons…

  • Most of AI’s reported diagnostic performance gains have been with 2D digital mammography, and may not apply to DBT mammography, which has become the standard in the U.S.
  • While autonomous AI interpretation of mammograms has been touted, there are practical barriers to its widespread application, such as the legal implications of AI misses.
  • There is no evidence showing AI reduces breast cancer mortality, and there are many other interventions that could do so at a lower cost.

Taking the other side of the argument was MGH breast radiologist Manisha Bahl, MD, who noted that…

  • Evidence supporting mammography AI’s effectiveness has come not only from randomized controlled trials like MASAI, but also from studies of routine population-based screening like the PRAIM study in Germany. 
  • Studies of mammography AI in the U.S. – where DBT is routinely used – have shown cancer detection gains comparable to those in Europe.
  • In addition to diagnostic support, AI can speed radiologist workflow, such as through autonomous AI reporting of some mammograms.

The Takeaway

In the debate over mammography AI deployment, both sides present persuasive arguments. But as is the case with AI applications outside healthcare, the technological imperative to leverage this powerful new technology is likely to swamp any attempts to put up guardrails restricting its use.

Measuring AI’s Impact in Real-World Radiology

Posted on by Brian Casey

One of the criticisms of AI in radiology is that there isn’t much data on its effectiveness in real-world environments. Well, a new study in JACR aims to fill that gap by detailing how AI adoption at a large imaging network in Europe affected key radiology performance metrics. 

Plenty of studies have been published on AI’s positive contribution to radiologists’ diagnostic performance. 

  • But in most cases this research was performed under controlled conditions, or involved a single AI algorithm working on a particular task. That’s left many wondering how well AI would perform under actual conditions. 

In the new study, researchers tracked the performance of 10 AI algorithms from seven vendors that were implemented over the Incepto Medical AI orchestration platform at 3R Swiss Imaging Network, a 20-center outpatient radiology network in Switzerland. 

  • Over 4.5 years, researchers tracked the network’s performance before and after AI implementation for the major radiology key performance indicator, report turnaround time (TAT). They also tracked AI adoption rates and infrastructure latency, or how delays in data transfer between PACS networks affected AI usage.

In all, 389k AI-assisted imaging exams were processed over the implementation period from 2021 to 2025, with researchers finding…

  • AI’s biggest impact was in high-volume modalities, reducing TAT for trauma radiography (-26%) and knee MRI (-18%).
  • AI improved the network’s diagnostic capacity, performing the work of almost half a full-time-equivalent radiologist for trauma radiography alone (0.46 FTE). 
  • AI usage rates were high, with 91% reporting active AI adoption and 66% reporting regular use.

On the downside, latency in data transfer between PACS networks slowed down AI performance, and in some cases AI results arrived after reports were finalized – too late to make a clinical impact…

  • Median total latency was 2.06 minutes per exam, with 72% attributable to data routing and tasks like fetching studies from PACS.
  • AI’s “too late rate” was 7.2% overall, ranging from 13% for chest CT to 3% for knee MRI, with 6.8% for trauma X-ray.

How did the network’s radiologists feel about AI? Researchers surveyed over 50 radiologists working for the network, finding that most of the differences in before-and-after opinions weren’t statistically significant. 

  • The only difference was a modest gain in the perception that AI made radiologists more productive (from 2.57 to 2.94 on a five-point Likert scale).

The Takeaway

The new study fills in a major gap in real-world experience regarding radiology AI implementation. It also reveals areas for improvement, as slow data transfer rates between networks can prevent AI results from reaching radiologists in time. 

Top 10 AI Vendors by FDA Approvals

Posted on by Brian Casey

The FDA updated its list of regulatory authorizations for AI-enabled medical devices, covering approvals granted through the end of June 2026. The new numbers show radiology maintaining its strong lead over other medical specialties. 

The FDA issues quarterly updates on marketing authorizations for AI-enabled products, a category that includes dedicated software applications as well as AI-enhanced hardware, like scanners with onboard AI functionality. 

  • The list is closely watched by healthcare industry observers, and radiology has typically secured more than three-quarters of the authorizations – not surprising given the specialty’s image-intensive nature. 

The most recent list tracks FDA actions through the end of June 2026, indicating that the FDA has…

  • Authorized 1,614 AI-enabled medical devices since it started keeping track, up 5.9% from Q1 2026. 
  • Authorized 1,230 radiology devices, or 76% of all AI-enabled medical authorizations. 
  • In the second quarter of 2026, the FDA authorized 89 AI-enabled medical devices, slightly fewer than the 92 devices it authorized in the first quarter of 2026. 
  • For the quarter, 66 authorizations (74%) were for radiology devices, similar to the Q1 ratio (75%).
  • GE HealthCare maintained its AI authorizations lead with 134, thanks in part to approvals secured by companies it has acquired over the years. 
  • Next are Siemens Healthineers at 101, then Philips at 62, Canon at 51, United Imaging at 45, Aidoc at 34, and DeepHealth at 32, with all numbers including acquisitions. 
  • Rounding out the top 10 are Samsung (21), RapidAI (20), and Hyperfine (13).

The Takeaway

The new numbers track pretty closely with data on AI-enabled regulatory authorizations the FDA released for the first quarter. It will be interesting to see whether the agency’s pace changes as it comes to a final decision on regulating generative AI algorithms used for direct clinical care – a task for which it is currently soliciting feedback. 

Harrison’s Teleradiology Play

Posted on by Brian Casey

AI developer Harrison.ai is expanding its business model by supporting a teleradiology venture called Frontier Radiology that’s scaling up and hiring radiologists. But the move drew unwanted attention from a critical article in the Australian news media this week.

Harrison is best known for its suite of AI algorithms for radiology applications, as well as its Harrison.Rad 1.5 foundation model that can draft reports for radiologist review.

  • Harrison is based in Australia, but has recently made the U.S. market a major focus, with co-founder and managing director Dimitry Tran relocating to California to spearhead the initiative.

But it was Harrison’s newfound U.S. focus that apparently drew the attention of ABC News, an Australian news outlet that on September 7 published a critical article that questioned the company’s moves, in particular its involvement with Frontier Radiology.

  • One bone of contention in the story was whether Harrison’s U.S. emphasis was appropriate given the Australian government’s 2025 investment of US$23M in the company. The story also claimed Harrison was planning layoffs of Australian staff and that its integration of AI with Frontier’s clinical services could be a conflict of interest.

But in an interview with The Imaging Wire, Dimitry Tran clarified Harrison’s strategic direction and its involvement with Frontier Radiology.  

  • Tran noted that Frontier is owned and operated by a radiologist – Joshua Ewell, DO – in accordance with U.S. rules requiring physician ownership of entities providing clinical services. 

Ewell’s LinkedIn profile says Frontier will be an “AI-native radiology group” built entirely around AI foundation models.

  • Harrison is providing Frontier with non-clinical services, including its suite of AI algorithms, which are already helping Frontier teleradiologists achieve industry-leading report turnaround times. But Frontier radiologists are free to use any AI solutions they wish – including those of Harrison’s competitors. 

Tran noted that close cooperation between AI developers and imaging services providers is hardly unusual anymore in radiology. 

  • Indeed, two of the specialty’s largest U.S. providers – Radiology Partners and RadNet – have formed their own AI divisions to provide algorithms to both their own radiologists and outside customers. 

The ABC News article conflated a number of recent developments into a narrative that doesn’t reflect reality, Tran believes. 

  • For example, the layoffs that occurred earlier this year at Harrison were connected to the company’s transition from an aggressive R&D phase into a commercialization push and weren’t related to Harrison’s U.S. entry.

And the Australian government’s funding was part of an investment that gave it a single-digit equity stake in the company – a stake it retains to this day and that will prove increasingly profitable with Harrison’s growing success.

  • Even as it supports Frontier, Harrison plans to continue its focus on AI algorithm development and commercialization, especially of the Harrison.Rad foundation model – while keeping the “vast majority” of its employees in Australia. 

The Takeaway

Putting aside the ABC article’s negative spin, Harrison’s move into teleradiology offers an intriguing twist on the growing integration between AI and imaging services providers. Given ongoing workforce shortages and rising imaging volume, it’s perhaps the best way to move the chains toward finding relief for beleaguered radiologists. 

Where Do Humans Belong in the AI Loop?

Posted on by Jack Troy

AI has its skeptics and superfans when it comes to interpreting medical images, inspiring a wide range of approaches to checking its work, but a new RadioGraphics paper argues the right level of oversight lies somewhere in the middle. 

There’s a strong consensus on the importance of monitoring AI post-deployment, even as the right level of rigor remains an open question. 

  • The FDA and many of its global peers require institutions to track AI performance, implement human oversight, and have a corrective action plan. 
  • Past studies show oversight mechanisms are top of mind for many radiologists. 

The international researchers behind the article say a human-on-the-loop (HOTL) model is the sweet spot for radiology departments as they balance safety and reliability with efficiency. 

  • Under this framework, radiologists don’t have to review every AI output.
  • At the same time, the proposed model doesn’t take AI results at face value.
  • HOTL threads the needle by alerting humans to drift and accuracy problems.
  • If performance drops, the work of reading these scans reverts to radiologists. 

The model acknowledges that even the most cleverly designed radiology algorithms can be sensitive to changing inputs.

  • Scanner upgrades, workflow adjustments, and shifts in patient mix can leave AI that aced training tests out of step with clinical realities. 

HOTL has its perks, but the authors note a few challenges. 

  • Excessive alerts risk desensitizing monitoring teams, so it’s important not to set the notification threshold too low or treat minor deviations as urgent. 
  • Declines in subgroup performance can go unnoticed if overall metrics stay stable.

The Takeaway

For radiology departments, the ideal AI oversight plan protects patients without making the technology more trouble than it’s worth. It seems the HOTL system outlined in RadioGraphics checks those boxes by focusing on trends over individual outputs. Even so, the debate over how to best supervise AI is sure to remain lively. 

Top 10 AI Vendors by FDA Approvals

Posted on by Brian Casey

Who are the top 10 radiology AI vendors, based on the number of FDA regulatory authorizations? The agency provided some clarity this week with an update to its list of authorized AI-enabled medical devices through the end of Q1 2026. 

The FDA updates the list on more or less a quarterly basis, and it’s become a closely watched barometer for tracking not only the health of the AI industry but also which companies have received the greatest number of authorizations.

  • As we’ve noted in the past, the list includes both standalone AI algorithms as well as medical hardware that has AI functionality embedded in it, like a mobile X-ray machine with an onboard AI feature for detecting fractures.

The updated list tracks marketing authorizations through the end of March 2026, and shows that the FDA has…

  • Authorized 1,524 AI-enabled medical devices since it began keeping track in 1995, up 5.1% from Q4 2025. 
  • Authorized a total of 1,164 radiology devices, or 76% of all AI-enabled medical authorizations. 
  • In the first quarter of 2026, the FDA authorized 92 AI-enabled medical devices, or 28% more than in the fourth quarter of 2025.
  • For the quarter, 69 authorizations (75%) were for radiology devices, about the same ratio as in Q4 2025 (76%). 
  • GE HealthCare held its lead as the company with the most radiology AI authorizations at 130 (including recent acquisitions that had AI authorizations of their own).
  • Next is Siemens Healthineers at 95, then Philips at 58, Canon at 48, United Imaging at 40, Aidoc at 33, and DeepHealth at 29, with all numbers including acquisitions. 
  • Rounding out the top 10 are Samsung (21), Rapid.ai (20), and Hyperfine (13).

The Takeaway

The FDA’s new numbers on AI marketing authorizations show that the agency is keeping pace with rapid developments in the healthcare AI industry. Indeed, the FDA is even accelerating its pace of product approvals compared to its last update, with radiology still securing the lion’s share of authorizations.

AI for Breast Cancer Risk

Posted on by Jake Fishman

Artificial intelligence may be capable of identifying subtle mammographic signs of breast cancer years before conventional diagnosis, according to a new study published in Radiology. Researchers from Sweden found that three commercially available AI algorithms for mammography screening generated elevated cancer scores as early as 10 years before diagnosis, with detection signals strengthening as diagnosis approached.

Predicting breast cancer risk offers the prospect not only of detecting cancer earlier, but also of tailoring mammography screening to women most likely to benefit from it.

  • Clinical risk calculators like Tyrer-Cuzick and breast density analysis are available, but AI-based algorithms are showing promise by predicting risk from screening mammograms.

In the new study, researchers analyzed 89k mammograms from 31.4k women collected over a 10-year period, drawn from Sweden’s national screening program, where women aged 40-74 undergo biennial mammography interpreted by two radiologists.  

  • During the study period, 12.1k women (39%) were ultimately diagnosed with breast cancer. Three commercially available AI algorithms were used to generate risk scores (Vara AI from Vara, Lunit Insight MMG from Lunit, and MammoScreen from Therapixel). (It’s worth noting all three were originally designed for cancer detection rather than risk prediction.) 

AI scores increased progressively over time in women who later developed cancer, while remaining relatively stable among cancer-free participants…

  • At 90% specificity, AI systems flagged 19%-20% of future breast cancer cases six years before diagnosis.
  • Detection increased to 23%-25% at four years before diagnosis.
  • Performance rose further to 35%-39% at two years before diagnosis.
  • Even 10 years before diagnosis, the systems identified 13%-17% of future cancers.
  • Across all pre-diagnostic examinations, AI achieved AUC values of 0.63-0.67, outperforming mammographic density alone (AUC = 0.57).

The findings suggest that AI tools developed for cancer detection may also have value as early-alert systems for identifying women who could benefit from closer surveillance or supplemental imaging.

  • While prospective validation is still needed, sequential AI scoring may ultimately help identify women who would benefit from supplemental imaging, closer surveillance, or earlier intervention.

The Takeaway

The study adds to growing evidence that mammography AI can extend beyond cancer detection to long-term risk stratification. By identifying subtle imaging patterns years before diagnosis, AI-derived detection scores could provide an additional layer of longitudinal risk monitoring and help guide more personalized screening strategies.

AI Reduces Mammography Workload

Posted on by Brian Casey

Using AI to triage low-risk breast screening exams that don’t need extra review could remove more than three-quarters of mammography cases from radiologists’ workload and allow them to spend more time on high-risk cases. That’s according to a new study in Radiology: Artificial Intelligence that confirms other recent studies. 

Much of recent mammography AI research has focused on its ability to triage low-risk cases to avoid additional radiologist review – saving precious personnel resources.

  • This is particularly valuable in Europe, which uses a double-reading paradigm in which two radiologists review all mammography cases (the U.S. employs single readers but tends to screen women annually rather than every two years). 

The new study comes from France, which employs a slightly different paradigm from the rest of Europe. Double reading is conducted only for lower-risk BI-RADS 1 and 2 cases, while BI-RADS 3-5 go directly to diagnostic workup. 

  • As such, double reading occurs with cases that have low cancer prevalence, which can make it more difficult for radiologists to detect cancers that don’t occur very often.

But what if you offloaded low-risk double reading to AI? 

  • In the new paper, researchers tried that with Therapixel’s MammoScreen AI algorithm, which was employed retrospectively to analyze mammograms from 42.4k women acquired from 2015 to 2019.

AI results were compared to standard radiologist double reading, with the following findings…

  • AI classified 77% of cases as low-risk, meaning these could be safely triaged from the double-reading paradigm.
  • AI missed only one cancer in the low-risk group, a rate the researchers characterized as “small but measurable.” 
  • Eleven cancers were found in the group AI classified as non-low-risk, which would have undergone double reading anyway in the AI triage paradigm.
  • Rates of interval cancer (cancer that occurs between screening rounds) were 5X higher in the cases AI classified as non-low-risk compared to low-risk (2.16 vs. 0.47 cancers per 1k exams). 

Using AI to classify and remove low-risk cases from double reading could therefore save significant resources from the French mammography screening program, with a “small but non-zero risk” of missed cancers.

The Takeaway
The new results track with findings from other recent studies that apply AI to mammography screening, particularly in Europe. While the French reading paradigm is unique, it’s instructive to see that AI maintains its ability to reduce radiologist workload across different types of breast cancer screening programs.

Get every issue of The Imaging Wire, delivered right to your inbox.

This content is exclusive to subscribers

Log in or join by entering your email below.

Completely free. Every Monday and Thursday