神马福利影片

It was a pleasure (and a privilege) to be invited to speak at the 鈥淭ransforming Digital Pathology & AI: The Path Forward鈥� conference in Edinburgh this year, particularly given the calibre of speakers. Sharing the stage with leaders in computational pathology such as Prof. Jeroen van der Laak, Dr Rapha毛lle Luisier (University of Bern), and the TIA Centre鈥檚 very own Dr Fayyaz Minhas made it clear from the outset that this would be a meeting worth attending.

Now in its fifth year, the conference has clearly established itself as a key fixture in the digital pathology calendar. It also helps that it was hosted at Murrayfield Stadium. While I admit to being more of a football fan than a rugby enthusiast, even I couldn鈥檛 fail to be impressed by the scale of the place, the 67,000-seater structure dominating the skyline as I stepped off the tram (a very convenient 5-minute journey from Haymarket).

A slight navigation mishap, courtesy of a well-meaning but directionally unhelpful security guard, meant I ended up doing a warm-up lap of the stadium before finding the entrance. Still, I made it in time for a cup of tea, so no real harm done.

The conference itself, held in the Presidential and Thistle Suite, brought together a mixed audience of consultant pathologists, trainees, industry representatives, and a healthy contingent of computer scientists, though the balance tipped firmly towards pathology.

The opening keynote from Prof. Jeroen van der Laak (Radboud UMC) set the tone nicely, offering an overview of the current state of the field, somewhere between promise and proof, with exciting developments on the horizon but a clear need for robust clinical validation. Throughout the day, it was particularly encouraging to see the continued impact of work originating from the TIA Centre. Multiple talks referenced or built upon tools like the TIA Toolbox (led by Dr Shan Raza) and HoVer-Net (developed back in 2019!), which remain widely used and influential.

My own talk went down relatively well. I began with my earlier work on oral epithelial dysplasia (OED), focusing on predicting cancer progression and identifying potential biomarkers, before transitioning to more recent efforts in multimodal fusion and emerging agentic AI pipelines. Encouragingly, this sparked a good Q&A and several follow-up conversations that carried on into lunch. These sort of conversations at conferences, where you can find new collaborators, are often just as valuable as the talks themselves.

Fayyaz鈥檚 talk, as expected, was both engaging and thought-provoking. Beginning with Plato鈥檚 allegory of the cave as an analogy for how we interpret tissue samples (and the limitations of partial observations), he weaved through concepts of causality, and the challenges of drawing robust conclusions from complex biological data.

The subsequent panel discussion, featuring Fayyaz, Jeroen, Scott Meleney (UKAS), and Hadassah Sade (AstraZeneca), was another highlight. The central question being 鈥渨here are we really with AI in pathology? Hype or clinical reality?鈥� The consensus leaned (perhaps unsurprisingly) towards 鈥渟till in the hype phase鈥�. However, as Fayyaz pointed out, this isn鈥檛 necessarily a negative. The current level of hype brings attention, investment, and momentum, or 鈥渙omph鈥�.

In the discussions, funding emerged as a key barrier. One UK-based pathologist highlighted the challenge of even digitising pathology workflows, let alone integrating AI tools on top. Jeroen countered with the longer-term view that, when deployed correctly, digital pathology and AI have the potential to deliver significant health-economic benefits. The challenge, of course, is proving this convincingly through prospective studies and real-world clinical deployment. There was also broad agreement that digital pathology is becoming essential for educating the next generation. As Jeroen noted, new trainees are unlikely to choose labs still reliant on microscopes alone when modern digital alternatives exist.

Lunch featured an excellent soup followed by beef and mash, before an opportunity to step out and actually explore the stadium itself, which was a real pleasure.

The final talk of the day, from Dr Fred Mayall (Consultant Histopathologist, Taunton and Somerset NHS), provided a refreshing perspective. He described the evolution of AI over the years with the comfort of a computer scientist, highlighting tools already embedded in day-to-day workflows, such as speech-to-text systems like Dragon, and proofreading assistants like Copilot. But he also boldly asked 鈥渨here are these new image-based AI tools? They don鈥檛 seem to be helping me.鈥� For those of us working in computational pathology, it was a slightly uncomfortable but important truth. It underscored the gap between research innovation and clinical impact, a gap that the field must work to close if it is to realise its full potential.

Soon after, it was time to head back home, taking a tram to Edinburgh Waverley, then the final (and rather slow) 4.5-hour train to Birmingham New Street, during which this article was written. Having arrived in Edinburgh only the night before, it made for a busy 24 hours but was certainly worth it.

By Adam Shephard