Key radiology IT trends worth watching

Rather than standalone diagnostic tools, many vendors are now emphasizing AI that streamlines reporting, automates clerical tasks and integrates more smoothly into daily workflows. These efficiency gains are built on years of informatics standards development within the imaging community, they both said.

Opportunistic AI gains attention

Another theme gaining traction is proactive opportunistic imaging AI, which watches all scans coming into the PACS and reviews them for incidental findings that have value in preventive health. This includes looking at all chest CTs to automatically perform a coronary artery calcium scan and look for lung nodules, even though the exam was performed for a different reason. These piggyback screenings with alerts to radiologists are expected to find heart or lung disease much earlier than waiting for patients to become symptomatic. 

“Vendors are looking at more of a proactive approach. AI has been more of a triage tool, but now they're looking at how can we prevent the event from happening in the first place,” Devlin said.

Other potential AI tools could focus on ordering physicians, clerical staff and technologists. Related care improvements could also help with staff retention and satisfaction amid workforce shortages and increasing patient volumes.

“From a technologist perspective, that's huge because if they can get through their work tasks faster, that would give them more time to scan more patients,” Devlin said.

FDA-cleared AI vs. practical deployment

Despite the sheer number of AI products on display at RSNA, Towbin cautioned that not all cleared applications are widely used.

“There are about a thousand FDA-cleared apps. Not many of those are in active use or routine use,” he said.

But he emphasized many high-impact AI applications do not require FDA clearance because they do not perform diagnostic detection. For example, tools that digitize paper forms or automate administrative processes can be deployed immediately to reduce burden and improve care.

At large practices such as Radiology Partners, Devlin said AI adoption has accelerated.

"Radiologists used to see AI as something nice to have and now it's something that they really desire to help them with their diagnoses. So, it's become a must have,” she said.

Towbin noted that pediatric hospitals like his face a different challenge: At children’s hospitals, there are not many applicable AI solutions. Many options are not cleared for pediatric use, so these hospitals require internal development and customized deployment.

Enterprise imaging and breaking down silos

As health systems look to tighten integration of all data into a unified electronic medical record system, enterprise imaging is seen as a key goal. Both SIIM leaders highlighted persistent challenges in image sharing and enterprise integration. Towbin said that while technology for image sharing exists, it is not universally implemented. He also pointed to ongoing efforts to develop standards and structured data elements to better measure radiology’s impact on population health. 

Siloed systems remain a concern, particularly outside radiology. Dermatology, pathology and other specialties often use non-DICOM formats and lack advanced viewers, complicating enterprise imaging strategies.

At the same time, Devlin said vendors are increasingly forming partnerships to bridge these gaps and reduce workflow challenges. 

Cloud-native platforms and next-generation AI

Cloud computing continues to evolve, with more vendors now describing their systems as “cloud native” rather than merely cloud compatible. Many IT vendors have unveiled cloud-native platforms over the past two years as it becomes a driver for better IT capabilities.

Towbin also highlighted two emerging AI approaches gaining attention: agent-based workflows and foundation models.

AI agents are capable of managing multistep workflows autonomously to increase workflow efficiency. These began to appear across RSNA vendors on the expo floor. Meanwhile, foundation models promise broader functionality.

“Those are the types of algorithms that have the promise to really start easing the burden instead of a narrow algorithm to solve one problem, one task that becomes expensive to implement and difficult to use,” Towbin said.

However, human interaction with these large, multitask algorithms remains a challenge, particularly when it comes to efficiently approving AI-generated outputs without adding excessive extra clicks.

Devlin underscored the importance of efficiency in AI-generated reporting and hopes it will make systems more streamlined rather than having to proofread and adjudicate what the AI creates.

Large language models enter the workflow

Large language processing tools are also becoming embedded in radiology applications. These systems can automatically format dictated reports, assign appropriate billing codes and distinguish between clinical dictation and casual conversation during interruptions. This is eliminating the need for transcription services and helping improve report turnaround times.

A also is increasingly being used to summarize patient records and extract data from EHRs to present concise problem lists and medication histories. Towbin said this reduces the time-consuming need to manually search through charts and prior reports.

SIIM’s role in shaping the future

Towbin emphasized that SIIM plays a distinct role compared to RSNA by focusing on enterprise imaging across all specialties and bringing together more technical stakeholders on the IT side of medical imaging.

SIIM meetings tend to feature deeper technical discussions among imaging informatics professionals, researchers and physicians aimed at solving real-world problems and shaping product development, he said.

Devlin added that SIIM’s collaborative environment fosters professional networking and practical problem-solving.