Each colour maps a published value to the exact source passage used to support it. Only bounded excerpts are public; administrators can inspect the complete captured source.
Title
University Hospital Bonn deploys an isolated on-premise open-weights LLM for radiology, piloted with 22 radiologists
The system served the open-weights DeepSeek-R1 model via vLLM.
…and incorporated into the deployment concept. 2.6 Study Design and Participants We conducted a real-world pilot evaluation over a one-week period in July 2025, following Institutional Review Board approval (2025-167-BO). Twenty-two radiol…
Description
University Hospital Bonn built an isolation-first, self-hosted LLM service on a single on-premise server with eight NVIDIA A100 80 GB GPUs, serving DeepSeek-R1 via vLLM behind Open WebUI, with network segmentation and egress filtering so unanonymized patient data can be processed. In a one-week pilot in July 2025, 22 radiologists used ten prompt templates; stability was rated high and source-anchored tasks such as translation and report correction rated best, while open-ended conclusion generation produced the most critical errors. Median time-to-first-token was 3.13 seconds. The authors state the system is the basis for an official service for a hospital with over 10,000 employees.
The infrastructure is deployed on a single on-premise server within the hospital data center equipped with eight NVIDIA A100 80 GB PCIe graphics processing units (GPUs).
The system served the open-weights DeepSeek-R1 model via vLLM.
whereas open-ended conclusion generation based on findings resulted in the highest frequency of critical errors
The system served the model with a low time-to-first-token of 3.13 seconds (median 3.13 seconds [IQR 2.00-4.81]).
is the current base to serve open-weights LLMs as an official service of a German University Hospital with over 10,000 employees.
Company
University Hospital Bonn
now a permanent service for over 10,000 University Hospital Bonn employees processing PHI.
Country
Germany
an official service of a German University Hospital with over 10,000 employees
Industry
Healthcare
…Sprinkart1, Julian Alexander Luetkens1, Benjamin Wulff2,*, Alexander Isaak1,* 1Department of Diagnostic and Interventional Radiology, University Hospital Bonn 2Department of Research-IT, University Hospital Bonn * contributed equally †Cor…
Problem
Data protection and information security requirements are a major barrier to using LLMs in clinical workflows: well-known proprietary LLMs are offered through cloud APIs that require data to leave the hospital network, and hospitals in many countries do not permit external LLM services with clinical data.
data protection and information security requirements remain a major barrier to integration of LLMs into clinical workflows.
Consequently, hospitals do not permit the use of external LLM services with clinical data in many countries.
Solution
The hospital designed an isolation-first, containerized LLM inference stack on a single on-premise server with eight NVIDIA A100 80 GB PCIe GPUs. It relies on strict network segmentation, host-enforced egress filtering and active isolation monitoring, and serves the open-weights DeepSeek-R1 model via vLLM behind an Open WebUI frontend. The architecture is model-agnostic, and the intended use is radiology text support without autonomous clinical decision-making.
…s and Methods: The isolation-first, containerized LLM inference stack relies on strict network segmentation, host-enforced egress filtering, and active isola- tion monitoring preventing unauthorized external connectivit…
The deployment architecture is model-agnostic and serves open-weights models through vLLM
We defined the intended use and operational boundaries of the system as radiology text support without autonomous clinical decision-making.
Business value
In a one-week pilot with 22 radiologists, participants rated system stability exceptionally high and usability strong, and no external connectivity events were detected. Source-anchored tasks such as report translation, linguistic correction and radiology guideline recommendations received the highest ratings, while open-ended conclusion generation produced the most critical errors. The authors note it was a single-centre pilot with user-reported errors.
Participants rated system stability exceptionally high
No external connectivity events were detected by the active isolation monitoring layer during the evaluation period.
It was a single-center pilot involving a modest sample of 22 radiologists over a one-week period
AI capabilities
Generative AI
The system served the open-weights DeepSeek-R1 model via vLLM.
Technology
DeepSeek-R1, vLLM, Open WebUI
The system served the open-weights DeepSeek-R1 model via vLLM.
…ul completion to assignment to a dedicated Active Directory (AD) or LDAP group, which is configured in Open WebUI as a prerequisite for login. 3.3 In-Field Evaluation of Locally Deployed LLMs in Clinical Routine To ensure…
Implementation approach
The hospital ran a real-world pilot over one week in July 2025 with 22 radiologists using ten predefined prompt templates. Institutional approval required coordinated review by the board of directors, data protection, information security, works council and compliance. The authors describe the system as now a permanent service for over 10,000 University Hospital Bonn employees.
…and incorporated into the deployment concept. 2.6 Study Design and Participants We conducted a real-world pilot evaluation over a one-week period in July 2025, following Institutional Review Board approval (2025-167-BO). Twenty-two radiol…
Institutional approval required coordinated review by board of directors, data protection, information security, works council, and compliance.
now a permanent service for over 10,000 University Hospital Bonn employees processing PHI.
Onboarding approach
Compliance required a mandatory AI literacy course before access; the hospital developed the training materials and linked course completion to a dedicated Active Directory or LDAP group that is a prerequisite for login in Open WebUI.
a mandatory AI literacy course prior to access.
…ul completion to assignment to a dedicated Active Directory (AD) or LDAP group, which is configured in Open WebUI as a prerequisite for login. 3.3 In-Field Evaluation of Locally Deployed LLMs in Clinical Routine To ensure…
Headline outcome
Six of the ten evaluated tasks (email generation, translation, linguistic correction, report simplification, guideline application, follow-up report generation) had zero reported critical errors across all interactions.
Use case type
Document processing
…anonymized PHI. The system was rated stable and user friendly during the pilot. Source text-anchored tasks, such as report corrections or simplifications, and radiology guideline recommendations received the highest utility ratings,…
