LUMC at a Glance
Size/Number of beds: 900
Employees: 10,000
Number of radiological examinations per year: 137,000
Background
Leiden University Medical Centre (LUMC), a leading academic hospital in the Netherlands, faced the common challenges of increasing radiology workloads and the need for more efficient resource allocation. Like many healthcare institutions, LUMC realised that, despite being a tier 3 academic hospital, a significant percentage of specific radiological examinations showed no pathology findings, which represented a significant opportunity to optimise workflow and use its resources more effectively. For chest X-rays, one of the most commonly performed radiological examination, retrospective validation showed that between 40% – 50% of the X-rays contained no findings and thus could be characterized as normal X-rays.
Challenge
In order to maintain high quality of patient care for patients coming in for chest X-rays, the radiology department at LUMC needed to:
- Manage high yearly volumes of ~ 23 000 chest X-rays per year
- Maintain diagnostic quality while improving efficiency
- Support less experienced staff, particularly during night shifts
- Create a scalable solution that could be implemented across regional partnerships.
Solution
LUMC and Oxipit began a collaborative partnership in 2018, initially utilising Oxipit’s Quality solution before expanding to include ChestLink. The hospital created a strategic, multi-phase approach to integrate the AI solution into existing workflows and applied rigorous validation protocols to ensure a smooth and seamless roll out. Together with Oxipit and PACS vendor, the LUMC focused on deep PACS integration where Oxipit results were directly and timely available in the PACS viewer in clinical routine. The initial phase positioned the AI as a secondary check and quality assurance tool, which was particularly beneficial for people during their first night shifts and less experienced staff members.
The system’s quality assurance framework proved particularly valuable, providing feedback when discrepancies were detected between radiologist reports and AI findings. In a few cases, this led to a re-evaluation of the images that were taken, demonstrating the AI system’s practical value in enhancing patient care.
After gaining buy in with clinicians, the team at LUMC was then able to conduct its own extensive validation using ChestLink. This included a retrospective analysis of approximately 20,000 chest X-rays, comparing radiologist reports with Oxipit’s findings using natural language processing algorithms. This helped to realistically estimate the amount of high-certainty no-findings reports and will enable to team to benchmark for future autonomous AI activity.
What’s impressed us most is how seamlessly Oxipit’s AI solutions integrate into our existing workflow. The implementation process has been a true collaboration, with strong support from their team in ensuring everything runs smoothly and securely. By implementing autonomous AI for normal case detection, we’re addressing a significant workflow challenge while maintaining complete control over quality and safety. For us, the value has been particularly evident in supporting less experienced staff and ensuring quality control. But what’s crucial is that we’re not just looking at efficiency within radiology—we’re considering the entire healthcare pathway. This technology allows us to scale the number of scans we can process with the same number of staff, while maintaining the high-quality standards our patients expect.
Results and Impact
The implementation of Oxipit’s ChestLink and Quality solutions at LUMC has delivered significant operational improvements across multiple areas of the radiology department. With chest X-rays typically requiring 3-4 minutes per examination for reporting – including image loading, analysis, dictation and sign-off – the ability to filter out 15-20% of normal cases with 99.97% sensitivity has the potential to create real time savings. This efficiency gain is particularly impactful given that approximately 45 – 50% of scans in LUMC’s hospital show no pathology.
The system has proven especially valuable during off-hours and night shifts, where less experienced radiologists particularly benefit from the AI’s support as a secondary check. This enhanced coverage has improved the department’s ability to provide rapid results to patients, potentially reducing wait times from days to minutes for normal cases.
Clinical outcomes:
- Confidently filtered normal cases, allowing radiologists to focus on chest X-rays with findings
- 15276 of cases processed in 2023 through the AI solution
- Maintained high accuracy rates which aligned with radiologist’s findings
Although radiologists were initially sceptical, through careful change management and ongoing dialogue they are now highly satisfied with Oxipit’s solutions. This transformation required a significant shift in mindset from the initial concept of full automation to a more nuanced understanding of AI as a tool to enhance rather than replace radiological expertise.
The key to gaining acceptance was acknowledging that concerns weren’t just about technical validation but often stemmed from professional considerations about patient care and training of radiology residents. Through a methodical approach that included standardised implementation protocols and extensive stakeholder engagement, the team gradually built trust in Oxipit’s AI solutions. They proved particularly valuable in supporting less experienced staff, while the integration into the hospital’s PACS system allowed radiologists to filter cases effectively within their familiar workflow. The quality control aspect, where AI cross-checks findings, has helped build confidence in the system’s reliability.
“We’ve seen firsthand how AI can transition from a helpful tool to an essential part of radiology. We know that we can go one step further by leveraging the autonomous AI feature of ChestLink, to unlock even greater potential, streamline workflows and improve patient care.”
Future plans
LUMC has an ambitious roadmap for using ChestLink’s fully autonomous capabilities. Using ChestLink autonomously will allow LUMC to automatically report on chest X-rays showing no findings, indicating high confidence in a healthy patient. This would allow LUMC to potentially produce finalised patient reports without any intervention from the radiologist, reducing their workload, creating greater time efficiencies and enabling them to focus on cases with pathologies.
The exact implementation of such an automated workflow paradigm is currently being investigated. The department of radiology is together with the hospital and University of Leiden actively exploring the legal framework that would allow automated reports to be sent directly to referring physicians without requiring radiologist intervention. Furthermore, as a step between different possibilities are explored where different non-physician practitioners such as radiographers or physician assistants will evaluate normal images that have been checked with AI. By freeing up tasks through automating normal cases, radiologists can focus on more complex cases, enhancing overall workflow quality and productivity.
To increase the value of the AI solution even further, the LUMC is also conducting research to optimise the AI’s sensitivity thresholds, aiming to reduce false positives while maintaining high accuracy rates. Currently, with a high sensitivity threshold of 99.97%, there are significant number of chest X-rays with no findings that are being flagged as abnormal. Reducing the sensitivity threshold will increase the fraction of normal detected chest X-rays, albeit at the expense of increased probability in false negative diagnoses. Tuning the thresholds to reduce false positive rates as much as possible while preventing unacceptable increase in false negative rates will enhance the value of Oxipit’s AI solution.
LUMC’s implementation of Oxipit’s AI solutions represents a significant step forward in radiology practice. Our experience shows that with careful implementation and strong stakeholder engagement, AI can successfully enhance radiological services while maintaining high quality standards. The project has laid groundwork for broader regional implementation, and we look forward to working with Oxipit on future autonomous AI initiatives.
LUMC is hoping to set up a regional radiology network with nearby hospitals. This regional approach, with Oxipit’s AI technology at its core, will eventually enable smaller facilities to benefit from advanced technological capabilities while maintaining consistent quality standards across the network. The aim is to create a more robust and efficient healthcare ecosystem that can better serve patient needs across the region, while allowing radiologists to focus their expertise on more complex cases requiring specialised attention.
Having AI solutions like that of Oxipit available in peripheral hospitals is particularly important, where the percentage of normal cases is typically higher (sometimes even up to 80%) compared to academic centres (50%). At the same time, LUMC is carefully considering various operational models, including the possibility of having radiographers, rather than radiologists, oversee the AI system for certain workflows. This “process shifting work-related tasks,” as described by the team, represents a fundamental shift in radiological practice, where the goal is not just to enhance existing workflows but to completely reimagine how radiological services are delivered across the region to ensure a high-quality sustainable radiology service can still be delivered in the future.