Case studies · Healthcare & medical research
AI-assisted digital pathology for severe asthma (SputumAI / Cytomine)
In short
A Cytomine-based medical-imaging and AI environment for sputum cytology in severe asthma: expert annotation at scale, AI-assisted labelling, cell detection and classification, morphology and inflammatory biomarkers — on dedicated GPU infrastructure.
- ~500,000
- expert annotations across the programme
- 20+
- cell types and categories
- ~100
- experts and contributors annotating
- Client
- Belgian digital-health consortium — hospital/university research environment and technology partners; 15+ project contributors and ~100 annotation experts
- Timeline
- May 2023 – April 2027
- Status
- 48-month R&D and validation programme — not a commercially deployed medical device.
- Runs on
- Dedicated private infrastructure: 4 GPUs, 50-TB NAS, dedicated Cytomine instance
The challenge
Manual analysis of sputum cytology is expert-intensive and hard to scale. The programme needed reproducible, AI-assisted cell identification, characterisation and clinical analysis — with an annotation workflow a large expert community could actually use.
What we built
Mentis built the imaging and AI environment and the infrastructure around it:
01
Annotation & AI-assisted labelling
A dedicated Cytomine instance for expert annotation, with AI-assisted labelling to speed up and standardise the work of ~100 contributors.
02
Cell detection & classification
Detection and classification models across 20+ cell types and categories, plus morphology analysis and inflammatory biomarkers.
03
MLOps & model evaluation
Training, evaluation and versioning pipelines so models can be compared and validated reproducibly.
04
Dedicated infrastructure & clinical integration
4 GPUs, a 50-TB NAS and large-image storage, with integration work toward hospital and clinical workflows.
Results
- ✓ ~500,000 expert annotations across the programme
- ✓ 20+ cell types and categories
- ✓ ~100 experts and contributors annotating
Figures are programme-level counts, not clinical performance claims.
Stack & deployment
Related services
Client names are withheld under NDA; figures are those established with the client — nothing is estimated.
Want a similar outcome?
Tell us the problem in a 30-minute readiness debrief. You get an honest read on feasibility, data, compliance exposure and a first step.
Mohamed Ben Haddou
Founder & CEO · Independent AI Expert for the European Commission
You talk to Mohamed, not a sales team — and the person who scopes your work is the one who delivers it.