Cells are the basic units of life and disease, and modern microscopy makes it possible to image them in their native tissue context across thousands of patients. The bottleneck is no longer image acquisition — it is the reliable, reproducible, quantitative analysis of those images. This challenge places students in the middle of that bottleneck and asks them to help break through it.

The challenge is offered jointly by [PioneeBio AB](https://pioneerbio.se/), and it builds directly on the ECIU micromodule “AI-based biomedical image analysis for cell segmentation”. While the micromodule introduces the methodology (classical segmentation, U-Net, Cellpose, and quantitative evaluation) on curated benchmark data, the challenge confronts students with a real biomedical imaging dataset and an open clinical question.

Working with anonymised specular-microscopy images of the corneal endothelium (provided by PioneerBio AB in collaboration with the Ophthalmology clinic in Linköping, student teams segment endothelial cells and compare quantitative readouts — endothelial cell density (ECD), cell area, polymegethism (CV of area), and hexagonality (pleomorphism) — between distinct patient groups. The biological question framing the work: does AI-derived morphometry detect subclinical stress signatures of the corneal endothelium that are not visible to the eye?

The challenge is deliberately scoped to be tractable in five weeks: the heavy AI engineering (training infrastructure, pre-trained models, etc) is already provided by PioneerBio AB on a dedicated NVIDIA GPU server at the LiU Valla data centre. Students focus on the scientific reasoning — choosing models, designing the comparison, validating the output, and interpreting the result — which is exactly the skill set the next generation of biomedical scientists and clinicians will need.

Teams work online during weeks 1–2 and 4–5, and meet face-to-face for an intensive on-site week at LiU Campus US in Linköping (week 3). The on-site week includes guest lectures from the clinical provider, a microscopy lab visit, hands-on debugging, and team coaching from the teachers. The challenge concludes with a written report and a public oral presentation to the challenge provider and an invited audience.

**Challenge provider:**

[PioneerBio AB](https://pioneerbio.se/) — a Linköping University AI spin-out (registered 2024) that develops a GPU-accelerated platform for code-free quantitative biomedical image analysis. PioneerBio AB provides the data, the segmentation infrastructure, and the clinical framing for the challenge, in collaboration with research and clinical partners at LiU.

Transcript of Records is an official document issued by your home university. It specifies the name of the institution, the academic programme you are enrolled in, the list of courses taken up until the current period, the exams or tests you passed, the grades and the credits you earned. 

Transcript of Records

Linkoping University

a close-up of a poker chip and a poker chips

AI-based biomedical image analysis for cell segmentation

Associate partner

PioneerBio AB

Associate partner

PioneerBio AB