The Radiological Society of North America’s 2025 Annual Meeting in Chicago showcased a medical imaging industry in infrastructure transition, with more than 700 exhibitors presenting advances that signal a shift from pilot programs to enterprise-scale deployment. Under the conference theme “Imaging the Individual,” vendors unveiled cloud-native diagnostic viewers, foundation model AI reaching clinical deployment, and hardware with AI embedded at the signal acquisition stage—developments that collectively represent fundamental architectural changes in how healthcare organizations capture, process, and interpret medical images.
The announcements reflect mounting pressure on radiology departments facing workforce constraints and expanding imaging volumes. According to the Bureau of Labor Statistics, radiologist positions are projected to grow 2% annually through 2033, while imaging study volumes continue climbing at 3-5% per year. Meanwhile, administrative burden consumes an estimated 64% of radiologist time—documentation, peer review, quality assurance, and care coordination tasks that pull physicians away from image interpretation. Technology vendors are responding with solutions designed to automate workflow bottlenecks, accelerate diagnostic turnaround times, and extend specialist expertise across broader patient populations.
HotSpot Take:
With foundation model FDA submissions follow cloud-native viewer launches and hardware manufacturers embed AI at the acquisition stage, medical imaging infrastructure has moved decisively from optional enhancement to essential architecture.
Cloud-Native Viewers Reach Diagnostic Readiness
GE HealthCare introduced Genesis Radiology Workspace with Genesis View, a zero-footprint diagnostic viewer designed for remote reading without local software installation. The 510(k)-pending platform represents GE’s attempt to address the operational friction that has limited teleradiology adoption—specifically, the complexity of maintaining multiple on-premise PACS installations across distributed reading sites.
“Genesis View is designed to be seamlessly integrated with existing PACS, offering a flexible zero-footprint solution that supports diagnostic reading from virtually any location,” the company stated. The viewer includes user-defined AI prioritization, FHIRcast interoperability for synchronizing clinical context across applications, and native integration with GE’s existing AI algorithms. The company reports more than 100 FDA-authorized AI-enabled solutions across its imaging portfolio, positioning it as the largest FDA-cleared AI catalog among medical imaging vendors.
Fujifilm Healthcare demonstrated its Synapse PACS with embedded AI and 3D visualization capabilities designed for multi-site health systems requiring consistent reading environments across facilities. The platform addresses a persistent challenge in enterprise imaging: maintaining diagnostic consistency when radiologists work across multiple hospitals, each potentially using different PACS vendors and workflow configurations.
University Radiology Group, a New Jersey-based practice serving more than 20 hospitals, implemented Synapse PACS to standardize mammography workflows across sites with different equipment vendors. The deployment consolidated previously fragmented reading environments into a single platform, according to the company, reducing the cognitive load of switching between disparate systems during interpretation sessions.
The cloud-native PACS architecture demonstrated at RSNA 2025 differs fundamentally from traditional on-premise systems that have dominated radiology for two decades. Zero-footprint viewers eliminate local software installation and maintenance, enabling diagnostic reading from any internet-connected device while centralizing security updates and feature enhancements. For distributed radiology practices and teleradiology services covering multiple hospitals, this architecture reduces IT overhead while improving radiologist mobility.
Foundation Models Move From Research to Clinical Deployment
Aidoc announced a breakthrough FDA submission for a multi-triage device capable of detecting and prioritizing double-digit abdominal conditions simultaneously from CT imaging—the first regulatory filing for a foundation model-powered diagnostic platform designed for routine clinical use. The submission follows FDA Breakthrough Device Designation granted in September 2025 for the company’s CARE Foundation Model, which demonstrated 97% sensitivity and 98% specificity in internal validation studies.
“Foundation models represent the next generation of medical AI—moving from narrow, single-disease algorithms to comprehensive diagnostic assistants that mirror how radiologists actually think.” – Michael Braginsky, Chief Product Officer, Aidoc
“Foundation models represent the next generation of medical AI—moving from narrow, single-disease algorithms to comprehensive diagnostic assistants that mirror how radiologists actually think,” said Michael Braginsky, Aidoc’s chief product officer. The CARE model processes complete CT studies holistically rather than analyzing individual slices in isolation, identifying multiple pathologies and their spatial relationships within a single inference pass.
Aidoc’s aiOS platform has analyzed more than 100 million patient cases since deployment, according to the company, with 18 FDA clearances spanning neurovascular, cardiovascular, pulmonary, and abdominal imaging. The platform operates across 150+ U.S. health systems and 1,600 hospitals globally. In February 2025, Aidoc received the first FDA clearance for a foundation model-powered solution—a rib fracture detection algorithm that identifies fractures across multiple anatomical locations simultaneously.
The foundation model approach addresses a critical limitation of narrow AI algorithms: the proliferation problem. Radiologists at large health systems can face dozens of point-solution algorithms, each requiring separate launch procedures, distinct user interfaces, and individualized result interpretation. Foundation models consolidate multiple detection tasks into unified platforms that integrate more naturally into existing reading workflows.
AWS announced a multi-year investment in Aidoc to accelerate foundation model development and cloud infrastructure scaling. The partnership builds on existing collaborations with NVIDIA for GPU-accelerated inference and Quibim for quantitative imaging biomarkers, creating a technology stack designed for enterprise-scale deployment across global hospital networks.
AI Embedded at Hardware Level, Not Just Post-Processing
Philips unveiled Verida, described as the world’s first detector-based spectral CT fully powered by AI. The system reconstructs 145 images per second—a throughput the company states enables 270 exams daily, compared to approximately 180 exams for conventional CT scanners. The performance gain derives from AI embedded at the signal acquisition stage rather than applied during post-processing, fundamentally altering the scanner’s data collection methodology.
Traditional spectral CT systems separate X-ray photons by energy level using dual-source configurations or rapid kV switching, then reconstruct spectral images through computationally intensive algorithms. Verida’s detector-based approach captures spectral information simultaneously at the detector level, with AI managing the separation and reconstruction in real-time. This architectural shift reduces scan times while maintaining spectral imaging quality, according to the company, enabling spectral CT deployment for routine protocols rather than specialized exams.
Philips also introduced BlueSeal Horizon 3.0T MRI—the world’s first helium-free 3.0T magnet system. Beyond the environmental and operational benefits of eliminating helium dependency, the system incorporates SmartSpeed Precise, an AI-powered imaging acceleration tool. The company claims the technology delivers 3x faster imaging with 80% sharper images compared to conventional acceleration techniques, enabling high-resolution scans within time windows previously achievable only at lower resolutions.
SmartHeart provides one-click automation for 14 cardiac views completed in 30 seconds, addressing the technical complexity that has limited cardiac MRI adoption in community hospitals. The tool automates scan planning, slice positioning, and sequence selection—tasks that typically require specialized cardiac MRI technologists and can extend exam preparation by 15-20 minutes.
The Advanced Visualization Workspace 16 with Cardiovascular Suite integrates Philips’ imaging analysis tools with clinical AI applications for cardiac, vascular, and neurological assessment. The company announced a partnership with Cortechs.ai for quantitative neuroimaging, bringing automated brain volumetrics and longitudinal analysis into the advanced visualization platform.
GE HealthCare’s Genesis product line spans hardware and software, with the company reporting more than 100 FDA-authorized AI-enabled solutions—the broadest FDA-cleared AI portfolio among imaging vendors. The Imaging 360 platform provides machine learning tools for fleet management, predictive maintenance, and utilization optimization, extending AI application beyond image interpretation into operational domains.
Enterprise Platforms: Breadth Versus Depth Strategies
DeepHealth, a RadNet subsidiary, presented DeepHealth OS—a cloud-native operating system designed to consolidate previously siloed imaging AI applications into unified workflows. The platform integrates five major acquisitions completed in 2025: iCAD for breast cancer detection, CIMAR UK for cloud image management, See-Mode for thyroid AI, Quantib for quantitative imaging, and Aidence for lung cancer screening.
The company announced an expanded collaboration with GE HealthCare, including global distribution of DeepHealth’s Breast Suite through GE’s customer channels and integration of the Thyroid Suite into GE’s ultrasound product line. These partnerships position DeepHealth as a comprehensive imaging AI vendor rather than a point-solution provider, with disease-specific suites spanning breast, thyroid, neuro, and chest imaging.
DeepHealth’s Breast Suite demonstrated a 21% increase in cancer detection rates in a real-world analysis published in Nature Health covering more than 579,000 women, according to the company. The Thyroid Suite achieved greater than 94% acceptance of AI measurements across 4,070 nodules while reducing exam time by 30%. TechLive remote scanning capabilities, FDA-cleared for ultrasound, have connected more than 400 scanners and achieved a 42% reduction in MR room closures through remote technologist supervision.
The company reports 2,000+ customers, 5,000+ radiologists using its platforms, and more than 10 million mammograms processed annually—deployment metrics that suggest enterprise-scale adoption beyond pilot programs. DeepHealth’s strategy emphasizes comprehensive platform coverage: providing end-to-end solutions from image acquisition through diagnosis, rather than narrow point solutions for specific pathologies.
RapidAI announced a strengthened partnership with AWS focused on co-developing foundational model infrastructure for neurovascular and cardiovascular imaging. The company positions Rapid Edge Cloud as “the market’s only cloud-native platform with on-premise capabilities,” addressing healthcare organizations’ conflicting requirements for cloud scalability and on-premise data sovereignty.
RapidAI operates across 2,500+ hospitals in 100+ countries with 700+ clinical studies supporting its neurovascular algorithms. The Rapid Enterprise Platform includes Rapid Aneurysm for intracranial aneurysm detection, Lumina 3D for advanced visualization, and Rapid Aortic for cardiovascular assessment. The company is expanding from neurovascular specialization into orthopedic, cardiovascular, and oncology domains—a breadth strategy contrasting with DeepHealth’s depth approach.
The strategic difference reflects broader market positioning debates in medical imaging AI: whether comprehensive platforms spanning multiple anatomies and modalities provide greater clinical value than deeply optimized algorithms for narrow use cases. DeepHealth’s acquisition strategy suggests platform breadth wins enterprise contracts, while RapidAI’s clinical study volume indicates deep clinical validation drives adoption. Both approaches gained traction at RSNA 2025, suggesting the market accommodates multiple architectural strategies.
Advanced Visualization and AI Orchestration
Fujifilm demonstrated its Synapse AI Orchestrator, designed to provide a common user experience across multiple AI engines from different vendors. The orchestrator addresses a persistent challenge in enterprise AI marketplace deployments: maintaining workflow consistency when health systems deploy algorithms from 5-10 different vendors, each with distinct user interfaces and result presentation formats.
The orchestrator acts as a middleware layer, routing studies to appropriate algorithms based on exam type, clinical indication, and protocol parameters, then consolidating results into a unified display integrated with the PACS reading interface. For radiologists working across multiple anatomies and pathologies, this architecture reduces cognitive switching costs while maintaining access to best-of-breed algorithms rather than forcing dependence on a single vendor’s AI portfolio.
Philips’ Advanced Visualization Workspace 16 integrates cardiovascular quantification tools with third-party AI applications through standardized interfaces, enabling health systems to combine Philips’ native capabilities with specialized algorithms from partners like Cortechs.ai. The platform supports structured reporting templates that automatically populate measurements and AI findings, reducing documentation time while improving report consistency.
Both Fujifilm’s orchestrator and Philips’ visualization workspace reflect industry recognition that no single vendor provides optimal AI solutions across all clinical domains. Health systems require interoperability frameworks that enable multi-vendor AI deployment while maintaining consistent user experiences—architectural requirements that are reshaping how imaging informatics platforms are designed.
Addressing the Workforce Crisis Through Technology
The technology announcements at RSNA 2025 consistently emphasized workflow automation and productivity enhancement—reflecting acute awareness of radiology workforce constraints. With administrative burden consuming 64% of radiologist time and radiologist supply growth trailing demand increases, vendors positioned their solutions as multipliers of existing specialist capacity rather than diagnostic curiosities.
GE’s Genesis View enables radiologists to provide subspecialty interpretation across multiple facilities without traveling between sites, effectively extending specialist availability to community hospitals lacking on-site expertise. Philips’ SmartHeart reduces cardiac MRI complexity to levels manageable by general radiologic technologists, expanding cardiac imaging access beyond academic medical centers with dedicated cardiac imaging staff.
DeepHealth’s TechLive remote scanning allows experienced sonographers to supervise multiple ultrasound exams simultaneously across different locations, addressing sonographer shortages while maintaining quality standards. The 42% reduction in MR room closures translates directly to increased patient access and revenue capture during peak demand periods.
Foundation models from Aidoc and others promise to reduce the number of AI algorithm launches required per study—consolidating 10-15 narrow algorithm interactions into single platform queries. For radiologists interpreting 50-100 studies per shift, reducing per-study algorithm management from 2-3 minutes to 30-45 seconds creates meaningful time savings that accumulate across daily workloads.
The workforce crisis framing suggests vendors recognize that pure diagnostic accuracy improvements no longer suffice as product differentiation. Health systems require solutions that demonstrably improve radiologist productivity, reduce technologist burden, and extend specialist expertise to underserved populations—operational improvements that directly address staffing constraints.
Strategic Implications for Healthcare Systems
The infrastructure investments demonstrated at RSNA 2025 carry significant strategic implications for healthcare organizations evaluating imaging technology roadmaps. Cloud-native architectures fundamentally alter total cost of ownership calculations by eliminating on-premise server infrastructure, reducing IT staffing requirements, and accelerating feature deployment cycles. Organizations accustomed to 18-24 month PACS upgrade cycles must reconsider investment horizons when cloud platforms enable continuous feature updates.
Foundation model AI shifts procurement strategy from acquiring multiple narrow algorithms to selecting comprehensive platform partners. This consolidation may reduce algorithm licensing costs while increasing vendor concentration risk. Health systems must evaluate whether unified platforms from single vendors provide sufficient clinical coverage or whether best-of-breed multi-vendor strategies justify additional integration complexity.
Hardware with embedded AI represents capital equipment decisions with different obsolescence curves than traditional imaging systems. When AI capabilities are fundamental to scanner performance rather than optional add-ons, imaging equipment refresh cycles may compress from 10-12 years to 7-8 years as AI capabilities advance. This acceleration affects capital planning and depreciation strategies across radiology departments.
The strategic divergence between DeepHealth’s comprehensive platform approach and RapidAI’s deep clinical specialization reflects broader uncertainty about optimal market positioning. Large health systems may prefer comprehensive platforms that standardize workflows across anatomies, while specialized imaging centers might prioritize deeply validated algorithms for their core clinical domains. Market dynamics over the next 18-24 months will clarify which architectural approach gains greater enterprise adoption.
Looking Ahead: From Pilot Programs to Production Deployment
RSNA 2025 marked a transition point where medical imaging AI moved decisively from pilot programs to production deployment at scale. The infrastructure announcements—cloud-native viewers, foundation models, embedded hardware AI, enterprise orchestration platforms—represent fundamental architectural shifts rather than incremental feature improvements.
Near-term developments will focus on integration: connecting these new infrastructures with existing EHR systems, billing platforms, and care coordination tools. Radiologists increasingly operate as consultants within broader care teams, requiring imaging results to flow seamlessly into clinical workflows rather than remaining isolated in PACS environments. Interoperability standards like FHIRcast and DICOMweb, demonstrated across multiple vendor platforms at RSNA 2025, will enable this integration while maintaining vendor diversity.
Longer-term trajectories point toward predictive imaging: identifying disease earlier in its progression when interventions prove most effective. Foundation models that analyze complete patient imaging histories rather than individual studies will detect subtle longitudinal changes indicating emerging pathology. Early stage localized breast cancer carries a greater than 99% survival rate according to the American Cancer Society—outcomes that worsen dramatically when diagnosis occurs at advanced stages. Imaging AI that enables earlier detection could shift survival curves across multiple cancer types.
The ultimate measure of these infrastructure transformations will be patient outcomes: lives saved through earlier diagnosis, reduced healthcare costs through prevention, and improved access to specialist expertise in underserved regions. Technology demonstrations at conference exhibits matter only insofar as they translate into measurable improvements in the humans those technologies serve. RSNA 2025 suggested the infrastructure is finally ready to deliver on that promise.
– This original article was created with AI support.
