AI Orchestration in Emergency Radiology – Implementation in the Valencia Health Region

The Valencia Health Region deployed a vendor-neutral AI orchestration system across 29 hospitals to improve emergency radiology. Validated at Hospital General Universitario Dr Balmis, it streamlines triage, accelerates diagnoses and reduces radiologists’ workload. The system processes over 5,700 studies daily, delivers results in 1:35 minutes on average, supports EMRAM and DIAM digital maturity goals and is expanding to other imaging specialties.

Key Points

• Valencia deployed AI orchestration in 29 hospitals for emergency radiology.
• Hospital Dr Balmis validated the system before regional implementation.
• AI delivers triage results in an average of 1 minute and 35 seconds.
• Over 580,000 radiology studies have been processed since deployment.
• The system aligns with EMRAM and DIAM digital transformation frameworks.

Introduction

The integration of artificial intelligence (AI) into medical imaging is reshaping emergency radiology by improving efficiency, increasing diagnostic precision and accelerating clinical decision-making. As part of a regional strategy to modernise services, the Generalitat Valenciana deployed an AI orchestration system, powered by Idonia, across all 29 hospitals in its healthcare network, serving a population of around five million people. The initiative was intended to ensure consistent radiological services, strengthen workflows and maintain interoperability with existing clinical and IT infrastructures.

The Hospital General Universitario Dr Balmis in Alicante played a central role in validating the system before it was extended to the entire network. The hospital tested its integration into daily practice, ensuring that the system could manage multiple algorithms while allowing clinical teams to retain control over selection and use. Initial implementation focused on algorithms designed to detect bone fractures and chest pathologies within emergency radiology. Medical images were automatically transmitted to the algorithms, which processed and prioritised cases in real time. By classifying studies as positive, negative or uncertain, the system supported emergency physicians in triaging cases and ensuring timely attention to patients requiring urgent care.

The deployment demonstrated measurable benefits. Automated prioritisation reduced diagnostic review times and eased the workload on radiologists, who could concentrate on complex cases rather than routine negative studies. By providing rapid results and highlighting potential findings, the system enhanced decision-making and improved overall efficiency within the emergency department. The successful validation at Hospital General Universitario Dr Balmis provided the foundation for rapid adoption across the wider regional network.

In addition, the orchestration layer was designed to strengthen workflow transparency. It enabled real-time monitoring of both original and AI-processed images and incorporated an analytics dashboard that allowed clinicians to track performance and refine processes. These functions supported continuous improvement, reinforced clinical confidence in the system and contributed to building a more resilient emergency radiology service.

Transforming Emergency Radiology with AI: Faster Triage, Smarter Decisions

The Hospital General Universitario Dr Balmis was the first site in the Valencia Health Region to introduce the orchestration system, providing a model for deployment across the wider hospital network. The hospital faced the challenge of rising patient volumes and increasing demand for rapid diagnostics, which placed considerable pressure on radiologists already working at full capacity. The introduction of automated triage aimed to improve efficiency in emergency workflows and ensure timely care for patients with the most urgent conditions, while also supporting continuity with primary care services.

At the centre of the implementation was the objective of enhancing triage and prioritisation in the emergency department. By integrating directly with the hospital’s picture archiving and communication system (PACS), the orchestration layer enabled real-time case classification, allowing critical studies to be identified and reviewed first. This change shortened waiting times for radiological diagnoses, supported faster clinical decision-making and helped reduce delays in patient management (Figure 1). On average, results were available in one minute and thirty-five seconds, enabling more rapid attention to critical patients. The system also added value by highlighting findings that might otherwise have been overlooked, strengthening diagnostic certainty and supporting more efficient use of resources.

The implementation followed a structured but flexible approach designed to limit disruption to hospital operations. It involved close collaboration between administrators, radiologists, emergency physicians, IT staff and technical teams to ensure alignment with clinical priorities. Traditionally, radiologists were required to manually review large volumes of images, creating bottlenecks and increasing workload. The orchestration system automated this process by classifying studies into positive, negative or uncertain categories (Figure 2). Clinicians were then able to direct their attention to the most urgent cases, reducing cognitive burden and accelerating decision-making.

The impact has been measurable at scale. Since its introduction, the system has processed more than 580,000 studies, averaging 5,700 each day (Figure 3). The automation of triage has reduced the workload of radiologists and emergency physicians, increased diagnostic capacity and enabled departments to manage greater volumes of imaging studies compared to traditional manual workflows.

From an operational perspective, the orchestration system has also helped reduce costs associated with radiology workflows. Automating triage and case prioritisation decreased the time required per study, leading to greater overall efficiency. Shorter diagnostic delays and smoother patient pathways further strengthened the efficiency gains, producing measurable economic benefits for the health system. These outcomes have reinforced the long-term sustainability of the region’s digital transformation strategy.

Addressing Key Challenges in AI-Driven Emergency Radiology

Despite measurable improvements, embedding AI into emergency radiology workflows required addressing several challenges to ensure effective adoption and long-term impact. Four areas were particularly important: interoperability, data management, data protection and clinical acceptance.

Ensuring Interoperability and Scalable Integration

Successful deployment of AI required real-time connectivity with hospital PACS, electronic health records (EHR) and other IT systems so that AI-generated insights could be incorporated directly into clinical workflows. A middleware orchestration layer was introduced to serve as a central point of integration between AI algorithms and hospital systems. This avoided one-to-one connections, reduced IT complexity and allowed hospitals to retain autonomy in choosing, testing and switching between different AI models as needs evolved. At Hospital General Universitario Dr Balmis, this included an algorithm designed to detect bone fractures and chest pathologies. Prioritisation results were embedded directly into PACS and radiology interfaces, enabling physicians to view AI outputs without additional steps. Images were anonymised or pseudonymised before being processed and securely transmitted to AI models, with results reintegrated into workflows once deanonymised (Figure 4).

Managing Large Volumes of Imaging Data

High demand for imaging required scalable solutions capable of processing large volumes of studies without disrupting workflows. The orchestration system automated anonymisation, structuring and secure transmission of imaging data. Cloud-based infrastructure provided scalability and high performance, adjusting dynamically to meet hospital requirements and variable demand. As a result, AI-generated outputs were delivered in an average of one minute and thirty-five seconds. The scalable design also supported rapid regional deployment, enabling implementation across three hospitals per week until all 29 hospitals were covered, without compromising performance or service continuity.

Safeguarding Data Privacy and Security

Maintaining strict data protection standards was essential for compliance and clinical trust. The orchestration system incorporated end-to-end security measures aligned with GDPR, ISO 27001, ISO 27017, ISO 27018, SOC 2/3 and Spain’s Esquema Nacional de Seguridad (ENS). All images were automatically anonymised before AI analysis, removing identifiers while preserving diagnostic quality. Data transfers were encrypted end-to-end to ensure secure communication between hospital systems and AI models. These measures enabled AI outputs to be embedded into workflows without compromising patient confidentiality.

Supporting Clinical Adoption

For the system to be fully adopted, AI-generated insights needed to be intuitive, reliable and aligned with existing physician practices. Results were therefore embedded directly into established reporting systems, avoiding additional steps for radiologists and emergency physicians. An intuitive colour-coded triage interface was introduced to highlight urgent cases. Validation phases enabled physicians to compare AI outputs with conventional diagnostics, reinforcing confidence in clinical reliability. A study conducted by the Generalitat Valenciana confirmed the system’s effectiveness, reporting positive predictive values of 76.8% for thoracic lesions and 61.3% for bone lesions, with negative predictive values of 83.0% and 95.0%, respectively. These results demonstrated that the system accelerated workflows while maintaining diagnostic accuracy, supporting trust in AI as a complement to clinical expertise. Continuous collaboration with hospital teams ensured that outputs were adapted to real-world needs and reinforced usability.

Scaling AI in Healthcare: Regional Strategy and Future Developments

Beyond addressing immediate implementation challenges, the Generalitat Valenciana has committed to ensuring that AI adoption remains scalable and sustainable across the healthcare system. This strategy is designed to enable the 29 hospitals of the Valencian Community to expand their AI capabilities over time while maintaining clinical integrity and alignment with healthcare priorities.

To achieve this, a structured framework for integration has been developed. It enables hospitals to:

• Flexibly adopt and evaluate AI solutions, with the ability to integrate, validate and switch between algorithms as needed, avoiding dependence on a single provider.
• Strengthen system robustness and risk management by incorporating monitoring and logging mechanisms for real-time anomaly detection, supporting uninterrupted AI-driven workflows.
• Use AI-generated data for continuous improvement by analysing algorithm performance and patient demographics to refine triage processes, optimise workflows and improve outcomes.

The next phase across the regional network will focus on enhancing triage efficiency and broadening AI applications beyond emergency radiology. Planned areas of expansion include neurological and cardiovascular imaging, with the goal of improving diagnostic precision and accelerating treatment pathways. Real-time analytics will also be developed further to generate actionable insights, guide resource allocation and support personalised alerts that improve patient flow. Continuous clinical validation and collaboration with hospital teams will remain central to this approach, ensuring usability, trust and sustainable adoption.

The initial success at Hospital General Universitario Dr Balmis provided the model for wider replication. The deployment across 29 hospitals was completed in a short timeframe, demonstrating that the orchestration system could be implemented rapidly and consistently across the network. This experience underlined the potential of AI-supported workflows to improve efficiency, enhance diagnostic processes and strengthen clinical practice at scale.

The initiative has also been aligned with established international digital maturity frameworks. On the Digital Imaging Analytics Maturity (DIAM) scale, the hospital has advanced towards Stage 5 through systematic use of radiological data and AI-driven insights to optimise workflows. The orchestration approach has contributed to progress towards Stage 7 requirements by enabling predictive analytics that support triage, resource allocation and diagnostic accuracy.

In parallel, integration with the Electronic Medical Record Adoption Model (EMRAM) has advanced the digital positioning of hospitals in the region. Real-time electronic exchange of imaging reports between clinicians and patients fulfils many of the Stage 7 compliance statements. The system also aligns with Stage 5–6 requirements for data integration, supporting interoperability between AI models, PACS and clinical decision-support systems. By facilitating secure image sharing, improving accessibility and enhancing patient-centred imaging, the orchestration system contributes to higher-level EMRAM objectives related to efficiency, safety and quality of care.

These advances strengthen digital governance and foster a more integrated radiology ecosystem. Alignment with Stage 6 of the Analytics Maturity Adoption Model (AMAM) is evident through the incorporation of AI-based recommendations into clinical decision-making, supporting guideline compliance and improving operational and patient outcomes. Looking ahead, the scalability of the orchestration system provides a pathway towards Stage 7, where predictive analytics are used to evaluate standardised care pathways, inform leadership decisions and enable the personalisation of treatment strategies to support equitable health outcomes.

Why an AI Orchestrator is Essential: The Vendor Neutral Orchestrator (VNO) Concept

With more than 300 CE- and FDA-certified AI algorithms available for medical imaging, integrating multiple solutions into hospital workflows presents considerable challenges. While many AI systems offer PACS connectivity, maintaining separate connections for each algorithm is inefficient and can create additional complexity. An interoperability platform is therefore essential, not only to support AI integration but also to enable secure image exchange, streamline referrals, enhance data analytics and improve accessibility for clinicians and patients.

The orchestration system was developed to act as a single connection point for multiple AI algorithms, reducing IT complexity and limiting the need for infrastructure modifications. Clinicians can activate or deactivate AI models as required, creating flexibility to test, compare and validate different approaches without dependence on one vendor. The system manages interoperability, data flow and infrastructure optimisation, allowing hospitals to integrate AI solutions efficiently (Figure 5).

The orchestration layer is designed as a vendor-neutral solution, allowing healthcare providers to integrate, validate and switch between algorithms from different sources. This approach avoids lock-in to proprietary systems, promotes interoperability and ensures that AI adoption remains flexible and scalable as clinical needs change.

Additional functions include automated anonymisation and rehydration of data. Before analysis, all patient identifiers and quasi-identifiers are removed to ensure compliance with GDPR and other healthcare privacy regulations. Once AI results are generated, data are securely re-associated in a traceable workflow. AI-generated findings are separated from original reports, giving radiologists the opportunity to review and validate outputs before they are included in patient records, which supports trust in AI-assisted diagnostics.

The orchestration system also regulates data volumes to prevent algorithm overload. Direct connections between PACS and algorithms can saturate processing capacity, slowing response times. By filtering irrelevant images and regulating transmission timing, the system prevents bottlenecks and maintains real-time efficiency. The ability to select destination algorithms further increases flexibility, enabling users to direct studies to specific models based on clinical needs or quality control purposes.

By streamlining AI integration and promoting interoperability, the orchestration model has provided a scalable and secure basis for radiology transformation in the Valencia Health Region. It also contributes to compliance with the emerging European AI Act by supporting transparency, accountability and risk management. The system’s monitoring and logging functions make AI-driven decisions auditable, while its data security and anonymisation measures ensure compliance with GDPR. This positions the region to continue advancing responsible and ethical use of AI in healthcare.

Conclusion

The deployment of a vendor-neutral AI orchestration system across the 29 hospitals of the Valencia Health Region has transformed emergency radiology workflows by optimising case triage, accelerating diagnostic review and enhancing clinical efficiency. By embedding AI into decision-making processes, hospitals have been able to reduce response times for urgent cases, allocate medical resources more effectively and lessen the cognitive burden on radiologists, contributing to a more resilient emergency care service.

Beyond its immediate operational impact, the initiative supports broader digital transformation objectives. The AI-driven workflow aligns with international frameworks such as the Electronic Medical Record Adoption Model (EMRAM) and the Digital Imaging Analytics Maturity model (DIAM). It enables advanced data integration consistent with EMRAM Stage 5 and supports the development of predictive analytics in line with DIAM Stages 5–6, strengthening the region’s position in the adoption of AI for diagnostic imaging.

Idonia’s AI Orchestration Platform has also demonstrated measurable operational and financial value. Automation of triage and prioritisation has reduced workflow costs, increased throughput and improved patient flow. On average, AI-generated results are delivered within one minute and thirty-five seconds, allowing patients requiring urgent care to receive medical attention more rapidly. Since implementation, more than 580,000 studies have been processed, with daily averages of around 5,700. These performance indicators reflect reductions in diagnostic delays, improvements in resource allocation and reduced workload for radiologists, reinforcing the sustainability of AI-driven healthcare transformation in the region.

Future plans include extending AI applications beyond emergency radiology into areas such as neurological and cardiovascular imaging. Further development of real-time analytics and AI-based decision support will enhance diagnostic precision, support workflow optimisation and advance patient-centred care. Continued collaboration with clinical teams and ongoing validation of algorithms will ensure reliability, usability and trust in daily practice.

The initial validation of the orchestration system at Hospital General Universitario Dr Balmis provided the foundation for regional rollout and demonstrated its capacity to be adopted at scale. Its successful expansion across all hospitals in the network has established a model for scalable, interoperable and compliant AI adoption, positioning the Valencia Health Region as a reference point for responsible and sustainable digital transformation in healthcare.

Conflict of interest

None