Discover how Aidoc’s AI platform offers seamless end-to-end integration into a facility’s existing IT infrastructure enabling implementation of AI at scale.
Discover how Aidoc’s AI platform offers seamless end-to-end integration into a facility’s existing IT infrastructure enabling implementation of AI at scale.
22 Hospitals, One Stroke Network: The Role of AI at the BG Klinikum Unfallkrankenhaus Berlin
Table of Contents
Prof. Dr. med. Sven Mutze and Alexander Böhmcker on using AI to support connected stroke care in Germany.
What happens when stroke imaging from 22 hospitals arrives fast — sometimes even at the same time — and urgent cases need to be prioritised for radiologist review?
At the BG Klinikum Unfallkrankenhaus Berlin (ukb), this is part of everyday practice. The hospital operates a large teleradiology network and houses a certified supra-regional stroke unit with the full range of interventional stroke therapy. Patient images arrive from a total of 22 connected hospitals, are interpreted with AI support, and patients with large vessel occlusions are — depending on regional distances — rapidly transferred to the ukb for thrombectomy.
This hub-and-spoke model provides a practical example of how AI can support connected and centralised stroke care. To understand how it works in clinical practice, and what other hospitals can learn from the experience, we spoke with Prof. Dr. med. Sven Mutze, Director of the Institute for Radiology and Neuroradiology at the ukb, and Alexander Böhmcker, VP Europe at Aidoc.
Together, they explore what happens when AI is embedded across a connected stroke network, including prioritising cases, supporting transfer decisions, aiding quality assurance, and integrating AI into the existing clinical and IT infrastructure.
What happens when a stroke network becomes “AI-enabled”?
With large volumes of imaging arriving at the central stroke unit from multiple connected hospitals, prioritising stroke cases can be challenging.
“Images arrive from a total of 22 hospitals, at times simultaneously,” explains Prof. Mutze. “In many cases, it is clinically uncertain what is causing particular symptoms, which also makes prioritisation in reporting difficult.”
At the ukb, Aidoc’s AI continually analyses all relevant CT scans and rapidly flags suspected cases of intracranial haemorrhage (ICH) or vessel occlusion (VO) via a desktop widget, surfacing the suspected cases to the radiologist for priority review.
For patients with an unclear time window, AI-supported CT perfusion analysis via the Aidoc aiOS operating system also provides mismatch and corresponding volumes quickly.
Because Aidoc’s AI solutions have also been deployed across the connected hospitals, the same approach extends beyond the hub.
“AI gives us a significant time gain through the notification and prioritisation of critical pathologies,” says Prof. Mutze. “Since the Aidoc AI solutions have been rolled out to all connected hospitals in the teleradiology network, corresponding vessel occlusions are also flagged promptly and reliably at the external sites. This allows a rapid transfer to be initiated if needed and suitable patients to be brought to thrombectomy.”
The effect of prioritisation is also visible in the ukb’s routine teleradiology workflow. An internal analysis across 21 connected clinics found a 21-minute reduction in the median waiting time for studies prioritised through the AI-enabled workflow1.
This should not be read as a 21-minute reduction in time to treatment, nor as evidence that AI replaces clinical judgement. What it illustrates is how an AI signal can move a potentially urgent study forward in a shared network workflow, so that radiologists review it sooner and specialist teams can concentrate on the cases that may require escalation or transfer.
The result is a connected pathway: imaging at the spoke, AI-supported prioritisation, specialist assessment at the hub and, where appropriate, transfer for thrombectomy.
More than prioritisation: an additional safety net
Prioritisation is only part of the added value.
“The ability to prioritise challenging, harder-to-identify vessel occlusions and smaller haemorrhages is markedly increased. Among vessel occlusions, patients with smaller occlusions of peripheral vessels — e.g., M2 — benefit in particular.” says Prof. Mutze.
In an offline, retrospective standalone-algorithm evaluation of 4,946 emergency head CT examinations in the ukb teleradiology network, the standalone AI identified additional intracranial haemorrhages that had not been described in the initial unaided radiology reports (representing a 12.2% increase in ICH detection when AI complemented radiologist interpretation; 85% of which occurred during on-call hours). Conversely, the standalone AI missed some haemorrhages that the unaided radiologists had identified. These retrospective findings suggest the potential benefit of AI complementing the radiologist to maximize haemorrhage detection in a high-volume setting, above all during on-call hours.
For a network like the ukb’s, this means AI acts as an additional safety net, supporting rather than replacing clinical interpretation, and supports quality assurance on top of prioritisation.
Seamless integration in the clinical workflow is key
For Prof. Mutze, effective hub-and-spoke care depends on more than the AI itself.
“The most important prerequisite for smooth operations within hub-and-spoke stroke care structures is the deep integration of neurology, radiology/neuroradiology, and interventional neuroradiology.”
The IT infrastructure also needs to support this integration. At the ukb, AI is integrated via the aiOS orchestration platform into the existing PACS-driven workflow through Aidoc’s orchestration layer.
“The rapid export of images from the PACS — or directly from individual scanners — via this orchestrator yields results typically within 2–6 minutes, which are then also displayed immediately to the radiologist.”
For Alexander Böhmcker, this is a key consideration when implementing AI across complex hospital environments.
“What we see increasingly is that hospitals are thinking beyond individual AI applications and asking how technology can support an entire network,” says Alexander. “The ukb model is a good example of that. When you have multiple hospitals connected to a specialist centre, AI can help ensure that potentially critical cases are prioritised consistently across the network, while the clinical teams retain responsibility for assessing the case and deciding what happens next.”
What can other stroke networks learn?
Since the start of the collaboration in 2020, more than 430,000 patient studies have been processed by Aidoc’s AI solutions at the ukb1. The ukb’s experience offers a practical example of what a connected, AI-supported stroke network can look like: multiple hospitals linked to a specialist centre, with AI supporting the identification and prioritisation of potentially critical cases across the network.
For hospitals considering similar models, the experience also highlights that the technology itself is only one part of the equation.
“You need to think about how the AI fits into the existing imaging infrastructure, how results reach the right people, and how it supports the clinical pathway from the spoke through to the hub,” says Alexander Böhmcker. “Getting those pieces working together is what makes the model scalable.”
Importantly, AI remains a supporting tool. “Each of these findings must be validated by a physician,” says Prof. Mutze. “These experts are supported by AI, but they must of course still analyse and assess each case individually before a therapy decision and verify its technical feasibility.”
For Prof. Mutze, the value of this experience goes beyond simply identifying individual findings more quickly.
“AI has thus become a supporting tool that is hard to do without, one that informs clinical workflows and supports timely treatment decisions,” he says.
The ukb experience shows what’s possible when AI is embedded into an existing clinical and IT infrastructure: connecting hospitals, supporting specialist expertise and helping potentially time-critical patients move through the pathway more efficiently.
Looking ahead: funding connected stroke care
Looking ahead, Germany’s Krankenhaus-Transformationsfonds (KHTF) may open a route for hospitals to develop AI-enabled stroke and telemedicine networks. Fördertatbestand 3 (funding category 3) is dedicated to the formation of interoperable telemedicine network structures between hospitals, which makes it potentially relevant to hub-and-spoke models such as the ukb’s.
Hospitals register their funding needs with their Land, which in turn applies to the Federal Office for Social Security (BAS), and approval depends on a project meeting the federal minimum requirements for telemedicine network structures and interoperability standards set by the Federal Ministry of Health.
For stroke networks weighing their next step, however, the policy direction and the clinical experience at the ukb point the same way: towards connected, interoperable pathways rather than isolated tools.
Aidoc is happy to provide further information and to support stroke networks in exploring and preparing a potential KHTF application.
Prof. Dr. med. Sven Mutze, Director of the Institute for Radiology at the ukbAlexander Böhmcker, VP Europe, Aidoc
Explore how Aidoc can help increase hospital efficiency, improve outcomes and demonstrate ROI.
Nicola Emery, PhD
Nicola Emery, PhD, is a marketing expert with experience in the pharmaceutical, life sciences and technology sectors. With a strong scientific background, she brings a data-driven, analytical and creative approach to her role as Director of Marketing, Europe.
