There is an extensive offering when it comes to artificial intelligence (AI), with over 75% growth by 2021 in the number of products for AI-powered radiology workflow. It’s no surprise when you consider there is so much pixel-based analysis and image processing data available from digital imaging reports. After all, good tools help accelerate the adoption of AI products and ultimately help more medical professionals see the tangible benefits AI brings to the quality of their healthcare services.
In my experience, AI isn’t just a tool that’s been sitting on the edge of fringe technology. There’s a good reason it keeps popping up in news and business reports; AI can be used as a time-saving technology when it comes to examining patients and analysing imaging exams. It is a technology partner designed to improve a radiologist's work-life balance by taking repetitive, boring and/or time-consuming tasks out of their hands and helping radiologists do what they do best and where their expertise is most pertinent. Integrating AI into a radiology workflow comes with quite a few benefits, including:
- Speeding up the pace of a radiology workflow process
- Increasing the accuracy and efficiency of imaging examinations
- Accelerated reporting of urgent findings by prioritising positive examinations
- Identifying human diagnostic errors and providing feedback to the radiologists
- Increasing the overall productivity of a radiology setting by automating procedures
- Improving the sensitivity of detecting pathologies
- Performing the examinations in a more sustainable way by reducing time, contrast injection and radiation
Recently, Osimis has observed a noticeable shift from on-premise standalone solutions to platforms based in the cloud but the market growth has still been quite modest. There are likely numerous reasons for this, including the technical aspects of transitioning to a cloud-based offering, the managerial challenges of overseeing such a shift, convincing decision-makers of the tangible medical benefits and obvious concerns about the financial and security implications.
Seamless integration will prove crucial for improving that adoption rate and driving the trend in assistive AI. Especially for predicting disease progression, interoperability of data and systems is essential, which would be a huge improvement for AI.
So, how can we all constructively contribute to the proliferation of AI in radiology? The industry experts from our panel were on hand to share their thoughts. Here’s a taster of what they had to say on the topic, and the key topics they discussed on the day:
Prof. Giovanni Briganti discussed how today’s healthcare is not the healthcare of yesteryear and why it is important to ask the right questions in order to find the right answers. What do clinicians really want and how can AI help them achieve those goals? How well established is AI in Europe and what needs to be done to progress that infrastructure? His input covers all these topics and more.
Prof. Dr. Sergey Morozov shared his insights on how to better understand the benefits AI can bring and the importance of identifying the five stages of radiology reporting automation. He also covered why it is important to compare the ratio between diagnostics vs skill level ratio and how one weighs up against the other. And what different clinical AI scenarios need to be considered? Prof. Dr. Morozov also addressed the ‘7 Ps’ for optimal success and the importance of striking the right balance between vendor qualification, radiology methodology and quality assessment.
Dennis Groen joined us to share a real-life use case of Gleamer’s BoneView, an AI-powered tool for identifying fractures and other bone trauma. This fascinating insight delves into the challenges the radiologist in this use case encountered prior to the introduction of AI, including ad hoc interruptions in the daily job for double checks. They also wanted fewer patient callbacks to the hospital due to night shift misdiagnoses. Gleamer moved through a two-phase process of testing, measuring first by analysing the issues and evaluating the improvements AI delivered into this medical setting.
- Ensuring better documentation of the pilot from the start – you need to honestly identify key struggles to better validate the AI results you produce.
- Compare with traditional doctor-led study to mirror their reporting outcomes and practices.
- Compare, learn, centralise and filter - Feed the algorithm with carefully curated data, but be aware that the more you feed into it, the more expensive the learning curve will be. Assess those running costs beforehand.
- GDPR-compliant data anonymisation processes are essential if you want to safeguard personal information, so consider opting for a SaaS integration.
What did this trigger?
So, what’s the perfect solution? The ideal answer: a single gateway, offering a vendor-neutral marketplace with all the anonymisation you could ask for. Of course, the reality isn’t quite so simple, especially when you consider the need for carefully managed governance, quality assurance and validation.
The conversation certainly triggered attendees to inquire about
- how we can translate the stats into impact
- the adequacy of today’s image quality standard being up to par to truly improve diagnostics
- the impact of AI on the jobs of future radiologists – are they gateways to new skills and expertises?
- the anthropomorphisation of AI
In short, AI is a technology with incredible potential but making it the right fit for every radiology department takes a lot of study, trial and communication. By remaining accountable and ethically minded as to how we develop, deploy, and validate AI, we can contribute to a major sea change across the healthcare sector.
Want the full story? Click here to watch the session recording and hear the full thoughts and insights from the esteemed panel:
