In a pathology lab at University College London Hospitals, an AI model called Kraken is quietly performing a task that would be impossible for a human. Every day, it analyzes roughly 200 anonymised tissue slides, identifying and classifying millions of individual cells in under two minutes [UCL Engineering, July 2026] [UCL Innovation & Enterprise, November 2026]. This is the clinical reality for Octopath AI, a London-based spinout from UCL that is betting its AI foundation models can turn the routine microscope slide into a source of precise, quantitative data for cancer care.
For pathologists, the promise is not replacement but augmentation. The sheer volume of cellular information on a single slide is overwhelming. Manually counting mitotic figures, assessing immune cell infiltration, or calculating tumour cellularity across an entire sample is a repetitive, time-intensive process. Octopath’s software, built on models like Kraken for H&E slides and Nemo for immunohistochemistry (IHC), automates that quantitative grunt work. It provides cell counts, classifications, densities, and spatial relationships,measurements that could inform everything from a patient’s prognosis to their likely response to a specific drug [UCL Innovation & Enterprise, November 2026].
The wedge of single-cell quantification
The company’s technical wedge is its claim of single-cell resolution across millions of annotated data points. The Kraken model, released in June 2026, was trained on 14.1 million labelled cells spanning 14 tissue and cancer types, and can detect up to 23 different cell types in a single pass [UCL Engineering, July 2026]. This allows it to generate what the company calls “clinical-grade spatial metrics” from a standard H&E slide, the most common stain used in diagnosis.
The output is not a binary “cancer or not” diagnosis, but a rich dataset. For a research pathologist, it could quantify how immune cells are arranged around a tumour. For a pharmaceutical trial team, it could provide an objective, reproducible endpoint, like the density of tumour cells before and after treatment. The company says its platform is already assessing samples from four clinical trials and is deployed across five sites, with UCLH as the flagship deployment [UCL Innovation & Enterprise, November 2026].
An academic engine with clinical contact
Octopath’s founding team reads like a page from a UCL research directory, which is precisely its strength. The trio,Dr Charles-Antoine Collins-Fekete, Dr Zhuoyan Shen, and Dr Esther Bär,are all embedded in the university’s medical physics and biomedical engineering departments [UCL Innovation & Enterprise, November 2026]. This academic pedigree is more than a credential; it’s a distribution channel. The company’s stated clinical-contact model involves pathologists at UCL and the Royal National Orthopaedic Hospital reviewing the system’s outputs on their own real cases, providing a direct feedback loop from lab to clinic [UCL Innovation & Enterprise, November 2026].
This research-to-clinic bridge is evident in the team’s published work. Dr Shen was the lead author on a recent study that used AI-quantified tumour cell density to identify which patients with advanced rectal cancer would benefit from adding a specific chemotherapy [News-Medical.Net, July 2026]. It’s a clear example of the translational potential the company is commercializing.
The company’s early traction is anchored in this academic-hospital network.
| Metric | Value |
|---|---|
| Sites Deployed | 5 sites |
| Slides Analyzed (Total) | 24000 slides |
| Clinical Trials Using Platform | 4 trials |
| Daily Slides at UCLH | 200 slides |
The path to a regulated product
In November 2026, Octopath AI secured £1.5 million in combined seed investment and grant funding, led by TwinPath Ventures and matched by an Innovate UK grant [UCL Innovation & Enterprise, November 2026]. The capital is earmarked for clinical evaluation, product development, and the long, expensive work toward regulatory approval. The company is hiring for roles in product, research, and software engineering, signaling a move from pure research toward a scalable product organization [Octopath AI].
The regulatory pathway is the most significant hurdle. The software performs a “measurement” function, which may allow it to navigate a potentially clearer, if still rigorous, path to market compared to AI tools making diagnostic claims. The company’s focus on providing data to support pathologists, rather than replacing them, is a prudent clinical and regulatory positioning.
The competitive landscape and the scale challenge
The field of AI-powered digital pathology is crowded with well-funded players, though Octopath’s specific foundation-model approach and deep academic integration differentiate it. The primary competitive pressure, however, may be less about technology and more about commercial scale and clinical utility. Moving from five academic hospital sites to dozens of independent labs requires a different kind of sales motion and evidence package.
The company’s near-term risks are familiar in clinical AI:
- Regulatory timeline. Progress toward FDA or CE Mark approval is costly and uncertain, with timelines that can stretch for years.
- Commercial transition. Success in a friendly, research-oriented environment like UCLH does not guarantee adoption in community hospitals or private labs with different workflows and incentives.
- Clinical validation. While early research is promising, broader studies are needed to prove that the metrics provided consistently lead to better patient outcomes across diverse populations.
The company’s most plausible answer lies in its dual-track strategy. By serving both clinical researchers and trial sponsors in pharma, it can generate revenue and evidence through research contracts while building the validation needed for broader hospital adoption.
What precision oncology looks like today
For patients with solid tumours, from breast cancer to sarcoma, the current standard of care often involves a pathologist examining stained tissue under a microscope. Key metrics, like the density of dividing cells or the presence of immune cells at the tumour border, are often estimated visually or sampled in small, representative fields. This process is subjective, variable between observers, and can miss subtle spatial patterns that carry prognostic information. Octopath AI is betting that by quantifying every cell, it can bring a new level of objectivity and depth to this foundational step in oncology. The next twelve months will be about turning the data from those 200 daily slides at UCLH into the clinical evidence that convinces regulators and a wider market that this isn't just research,it's the future of routine care.
Sources
- [UCL Innovation & Enterprise, November 2026] Octopath AI secures £1.5 million to support cancer pathology | http://www.uclventures.com/spinout-portfolio-news/2026/octopath-ai-secures-ps15-million-support-cancer-pathology
- [UCL Engineering, July 2026] New AI foundation model ‘Kraken-1’: from biopsies to precision medicine insights under 2 minutes | https://www.ucl.ac.uk/engineering/news/2026/jul/new-ai-foundation-model-kraken-1-biopsies-precision-medicine-insights-under-2-minutes
- [News-Medical.Net, July 2026] AI reveals effective combination therapy for advanced rectal cancer | https://www.news-medical.net/news/20260727/AI-reveals-effective-combination-therapy-for-advanced-rectal-cancer.aspx
- [Octopath AI] Careers page | https://octopath.ai/career
- [UCL Engineering, December 2024] New AI cancer pathology research platform launched | https://www.ucl.ac.uk/engineering/news/2024/dec/new-ai-cancer-pathology-research-platform-launched