Solving Unpopular Problems: The QuPath Story

Meet the developer of the open-source digital pathology platform that's transforming image analysis across the globe. An interview with QuPath creator Peter Bankhead (University of Edinburgh) on the s

Summary

An interview (by Helen Bristow, The Pathologist, 29 July 2026) with Peter Bankhead, Reader at the Institute of Genetics and Cancer, University of Edinburgh, and creator of QuPath — one of the world's most widely used open-source image-analysis platforms for digital pathology research. The QuPath team was recently given a Royal College of Pathologists team award for innovation. The piece frames QuPath's success against a road "beset with false starts, administrative battles, rejection, and frustration."

"When I started in digital pathology, I felt the lack of a pathology-friendly open-source platform was a big problem." — Peter Bankhead

Origins — an accidental platform

• Bankhead entered digital pathology as a postdoc in 2012. His background was retinal image analysis (PhD) and three years as an image analyst in a microscopy facility; he hadn't appreciated how large and complex whole-slide images could be, and existing tools weren't designed for them.

• He spent ~2 years trying to adapt image-analysis tools to score IHC biomarkers in tissue microarrays — largely unsuccessful because the plugins/scripts he wrote were hard to use.

• He eventually set out to write a whole-slide image viewer "mostly to prove to myself it wouldn't work." With the help of OpenSlide (an open-source library for reading pathology image formats), the viewer quickly became useful; adding analysis features on top of it grew into QuPath.

"It wasn't planned, and it only happened because not writing QuPath hadn't worked very well."

Setbacks

• Bankhead had to fight to make QuPath open-source — permission to release it was granted only after he resigned his postdoc, during his notice period. When a later job stopped him continuing the work (even in his spare time), he left that too, and spent some months unemployed.

• The main QuPath paper was rejected by at least five journals without being sent for peer review, considered unlikely to have much impact. These years, though exhausting, made him less concerned with conventional measures of success.

Reach and impact

• The main paper was eventually published in Scientific Reports and has been cited >6,000 times (>1,500 new citations last year alone) — an undercount, since many studies use QuPath without citing it.

• The software has been downloaded over a million times, used across academia and industry and worldwide across many diseases; one major tech company even featured QuPath in product launch videos.

How it is used

• Many projects involve detecting and classifying cells in WSIs by morphology, staining, or both. Its application to Ki67 is well established (identify tumour vs non-tumour cells by morphology, then compute the % of tumour cells positive by staining).

• Crucially, QuPath has no dedicated Ki67 algorithm — it provides image-processing and machine-learning building blocks for custom algorithms, which is what makes it flexible. It's not limited to cells or to WSIs (projects include fluorescence confocal z-stacks and electron microscopy). Bankhead has personally used it to digitise a family photo album; he's heard of it being used on fossils and to quantify the fracture behaviour of cheese.

Community and ecosystem

• A small core team (never more than a few people) with a much larger user community — tens of thousands of posts across >5,500 topics on the Scientific Community Image Forum.

• The team keeps writing the core software themselves and encourages others to build extensions rather than fork:

  • Collaboration with Joel Saltz's group at Stony Brook (esp. Jakub Kaczmarzyk) to run AI models interactively.

  • An extension for InstanSeg, a fast, accurate AI model for nuclei/cell detection created by Thibaut Goldsborough (a PhD student in Bankhead's group); other developers have integrated it into their own software.

  • OpenMicroanatomy and its QuPath Edu component, an open-source teaching platform built by medical student Aaron Yli-Hallila (University of Oulu, Finland), used to teach medical students in Finland for years, with pilots in South Africa and Namibia.

Future

• Two developers currently work on QuPath's code — Bankhead and Alan O'Callaghan (research software engineer/postdoc). They have a backlog of ideas for the era of AI, multiplexed, and multidimensional imaging, and a year or two of funding to implement them.

Why open-source

• Bankhead's career was built on open-source software (ImageJ, about which he wrote an open handbook). He saw the lack of a pathology-friendly open platform as "incredibly — even unethically — inefficient": in-house/proprietary tools made it impossible to verify claims, reproduce results, or reuse methods.

• He understands why companies and academics are reluctant to open their code (business case; fear of exposed bugs; the time cost of documentation; the career disincentive to polish rather than publish) — which is precisely why no one had built one, and why he was determined his own software would be open.

On clinical use (a deliberate "no")

• Bankhead does not want QuPath or a derivative approved for clinical use. Adapting it for the clinic would make it less flexible for research and introduce legal/regulatory burdens better handled by companies. He argues the work helps patients and pathologists more effectively by keeping its research priorities.

"If a problem already gets a lot of attention, then I'd rather spend my time on something else."

On impact and AI caution

• He believes QuPath has helped labs make substantial cost/efficiency savings — making some studies possible and others unnecessary — and hopes it improved research culture around openness and reproducibility. Because everything is open, "it makes little sense for anyone to publish something new and worse"; open software raises the baseline rather than competing for users.

• On AI-augmented image analysis, he urges caution: the important questions are about how the technology is used and who benefits (he welcomes studies of automation bias), and there's a "huge difference between a proof-of-concept published in a journal and a genuinely useful software tool."

"If you're a pathologist, I hope you won't trust what computationally minded people like me claim our tools can do. Rather, I hope you'll try them out where you can, engage critically, ask awkward questions, and help shape how the field continues to develop."

Source

• Solving Unpopular Problems: The QuPath Story — Helen Bristow, The Pathologist, 29 July 2026 (Interview, ~8 min read).