From recent publications and lectures that took place at the latest ECR in July 2022, it is clear that the healthcare landscape is changing dramatically. New insights and technologies are pushing healthcare knowledge to new heights, but at the same time, these are also putting additional stress on radiologists, who must be vigilant not to succumb to the increasing workload that could negatively affect the diagnostic quality of their medical image analysis.
Medical imaging faces an annual growth of more than ten percent and a fourfold increase in the workload of radiologists. Between 2006 and 2020, an increase of more than 500% was recorded for CT scans. But at the same time, the entire healthcare sector is looking into budget sanitations to secure the necessary financial savings. These challenges will only increase over time. Making it crucial for the sector to start seriously investigating viable, cost-effective, and above all, sustainable solutions.
We believe that artificial intelligence (AI) can provide significant benefits to those seeking to overcome these existing and increasing difficulties in medical imaging. The question is how to make that happen.
Why would you want to integrate AI into a clinical setting?
Before exploring the potential of existing data science tools, one must start by analysing the bottlenecks in the workflow. Next, you need to investigate where AI-based solutions could help tackle these problems and explore how they can be incorporated into everyday clinical practice.
The ultimate goal should be to optimise the value of patient care, which can only be achieved by:
- Improving the quality of services without increasing the cost of delivering them
- Reducing the costs without negatively impacting the overall quality of care, or
- Simultaneously reducing the costs and improving the care package
Achieving this goal requires the consideration of many options, including
- Streamlining the workflow for both radiologists and radiologic technicians, for example, by reducing the reading time that is required for imaging examinations
- Reduction in total radiation dose and usage of contrast agents
- Faster detection of diseases by prioritisation of examinations with urgent findings
- Improving the accuracy of diagnoses by providing objective quantitative data and automatically measuring longitudinal follow-up
- Improvement in the presentation of examinations, e.g., based on personalised "hangings" of cross-sectional examinations
- Improving the timely communication of results, and automated follow-up of related actions
However, the totality of the cost is not determined solely by the cost of the selected AI model that offers you those features to make these variables happen. The selection can only be done following an in-depth multidisciplinary analysis of the business case, involving all stakeholders to weigh in on whether or not this is the right solution for the short, medium, and long term. It’s not only about the financial aspect, you also need to factor in equally essential factors such as technical infrastructure, maintenance, and staff capacity for preparing the implementation.
How do you do it?
Your first step should always be determining the goals one wants to achieve with the algorithm. This could be improving overall workflow efficiency, optimising diagnostic quality, or both. After all, what anyone really wants is an improved situation for both the clinical department and the patients.
Next, you need to consider the cost of implementing an AI solution in your department. The cost will depend on the type of AI model you wish to implement, its complexity, the existing digital infrastructure, the volume and type of analyses that need to be made, and the results you have in mind. Will you require a short-, medium- or long-term solution? Different setups ultimately lead to different price ranges, as no two clinical environments are the same. After all, cheap is expensive, especially in healthcare, but more on that in a future blog...
But above all, you must assess the value you hope to get from your investment in AI. That value translates on two different levels: people and practice.
What should a proper AI implementation do for different people in your healthcare facility?
It all depends on the existing situation and needs in your institution. Different people and different departments have different needs and wants. Let's look at this from different perspectives:
Ideally, radiologists should be able to try out existing AI solutions in advance via a "plug and play” setup. Once approved, final adoption into clinical practice should be facilitated through efficient project management and training of all users. Higher ROI can be achieved through optimal integration with existing systems, alongside continued methodological improvements based on feedback and long-term management agreements with the providers of these solutions.
AI can improve the overall efficiency of the hospital's diagnostics and care services and offers the potential to improve the working environment of the healthcare providers involved while saving costs.
- IT Team and PACS management
IT can benefit from the lower costs offered by a centrally organised project management solution, especially if a clear distinction is made between the tasks of the institution's IT management and those of the AI vendor or platform offering the AI solution. Crucial here is that the necessary support is provided via one clear dashboard and that optimal security of the whole can be guaranteed at all times.
- PACS Admin
An AI solution should be seamlessly integrated into the existing PACS and RIS (or even EMR) to enable optimal use of data. This should also ultimately improve the AI solution's results, both in the short and long term.
- DPO (Data Protection Officer)
A centralised radiology AI platform offers the advantage of being able to work with one trusted party, under one overall data processing agreement. This is also coupled with the proactive collection of Data Protection Impact Assessments.
When working with one platform and a centralised approach for multiple hospitals, the hospital purchasing department is in a better bargaining position when it comes to volume pricing. It also makes financial administration a lot easier.
Truth time - how has it delivered so far in practice?
In a 2021 paper, TC Kwee et al. stated that while there isn’t much evidence to support the speculation that AI will decrease workloads, “it can already have an important impact on political and strategic decisions.”
AI has the potential to improve the entire radiology workflow, including initial acquisition, analysis, pathology detection, and treatment prognosis. AI can positively impact the workflow process from start to finish. Its value chain and benefits include
- Scheduling - Optimising study scheduling, predicting and preventing patient non-attendance, and improving wait times.
- Protocol Optimisation - AI can help improve and automate scanning protocols and e.g. enable the reduction of radiation dose and/or limit contrast usage
- Modality Ops & Workflow - The ability to detect examination anomalies in real-time via a dashboard, as well as enable improvement in imaging quality control and improve the radiology technician’s workflow
- Reporting & Communication - AI enables the optimisation of case assignment to radiologists, automatic creation of structured reports, longitudinal examination of lesions and multi-level reporting
- Billing - AI can increase billing accuracy
Everyone’s bottom line: benefiting the most important ‘p’ - the patients
By offering the most effective healthcare technology, we hope to help radiologists and hospitals offer better services by helping their professionals do what they do best:
- Delivering excellent care to their patients.
- Better health technology for better health care.
Our final question is for you.
When you weigh costs and value, where is your tipping point?
Talk to Osimis today, we can help you find that return on your investment.
- Bruls and Kwee, Insights Imaging (2020) 11:121 https://doi.org/10.1186/s13244-020-00925-z
- Kwee, T.C., Kwee, R.M. Workload of diagnostic radiologists in the foreseeable future based on recent scientific advances: growth expectations and role of artificial intelligence. Insights Imaging 12, 88 (2021).
- Ahn, J. S. et al. Association of Artificial Intelligence–Aided Chest Radiograph Interpretation with Reader Performance and Efficiency. JAMA Netw Open. 2022;5(8):e2229289. doi:10.1001/jamanetworkopen.2022.29289
📸 Credits to DALL-E, an AI-based image generator we used to generate this image using the terms "futuristic radiologist makes the diagnosis of an apple with the help of artificial intelligence in the style of René Magritte"