Some background
The work presented in the Gemini study paper was produced while I was a postdoctoral research fellow at the University of Aberdeen. At the time, and even now, I think there is a type of magic that happens when you are in the right place at the right time with the right team.
My involvement in artificial intelligence (AI) evaluation for healthcare started in 2020 through serendipity. I just finished my PhD at the University of Aberdeen and successfully interviewed for a post at the University of Southampton. There was an underspend in the iCAIRD (Industrial Centre for Artificial Intelligence Research in Digital Diagnostics) project, and I was told it could be re-allocated for a research post for me at Aberdeen. I declined the post in Southampton with a verbal offer from Aberdeen but without a contract in hand, as the role in the iCAIRD project sounded exciting and innovative. This bet paid off. I threw myself into iCAIRD, learning about breast cancer screening, developing my data science skillset, collaborating with clinicians and industry partners, and exploring stakeholder views.
The Gemini study
With the same team, based on the iCAIRD results, we designed and ran the Gemini (Grampian's Evaluation of Mia in an Innovative National breast screening Initiative) study just published in Nature Cancer. In this study, we evaluated 17 different AI workflows for breast cancer screening using a combination of live use and simulations. Like a human reader, the AI can assess women’s mammograms for potential signs of cancer. For the primary workflow, we showed that AI use can increase cancer detection by 10.4% while not increasing the number of women invited for additional investigations, and reduce workload by up to 31%. Other workflows show different benefits, demonstrating that AI use can be tailored to an individual site’s needs.
“Magic” ingredients
I am now a Lecturer in Data Science at the University of Glasgow. In this role, alongside my co-authors, I put the finishing touches on the manuscript. I have reflected on the ingredients that lay behind the “magic” and which made our collaborative work a success, and believe there were 3 key elements:
1) A large, interdisciplinary team. The number of people involved in this study exceeds 20. For an interdisciplinary project such as this one, you need expertise beyond a single individual’s, spanning across NHS (clinical), academia (data science, research methods), industry, and patient and public input. By highlighting this, I would like to help counteract the prevailing narrative of extraordinary individuals single-handedly progressing science, and showcase the value of collaboration over competition.
2) Communication. The academics’ incentives in pursuing this project differed to that of the industry and NHS partners. Understanding each other’s points of views, priorities and objectives, helped facilitate effective communication. This is where relationship building heavily comes into play and where the strength of the human touch shines.
3) Putting egos aside. I believe that most of us, most of the time, successfully put our egos aside during the project to focus on getting the work done, and keep the ultimate goal in mind: improve service delivery for women attending screening.
Looking ahead
Beyond the academic output, our collaboration has also received external recognition: we received the Innovative Collaboration Award at the 2021 Scotland’s Life Sciences Awards, and later the Multiparty Collaboration award at the 2023 Scottish Knowledge Exchange Awards.
As I look back on this journey, from a serendipitous opportunity to a multi-year collaboration, I consider two conflicting thoughts. On the one hand, you can’t ensure success. On another, you can skew things in your favour by lining up key ingredients.
The Gemini work is one chapter in a much longer journey toward integrating AI safely and effectively into healthcare. My hope is that our work contributes not only evidence, but also a model for how interdisciplinary teams can collaborate effectively towards the goal of translation to clinical practice. And in this way, hopefully the magic can live on.