☰

BioCam

From laboratory validation to regulated capsule endoscopy

Industry
Medical Devices and Diagnostics
Headquarters
Wrocław, Poland
Public information as of
January 2026

A4BEE prepared this analysis from publicly available sources. It reflects our own reading of BioCam's published strategy and is not endorsed by, or produced in cooperation with, BioCam. Company website

Strategic priorities

BioCam was founded in Wrocław in 2019 to replace conventional gastrointestinal endoscopy with a swallowable capsule, a wearable receiver, and a cloud platform that runs computer-aided detection (CADe, automatically flagging abnormalities) and computer-aided diagnosis (CADx, classifying them) on the captured imagery. The capsule itself measures 11 mm by 23 mm, captures roughly 74,000 images per examination, runs on more than 10 hours of battery, and transmits through a custom wireless stack developed in-house because off-the-shelf radio modules were not viable across the digestive tract.

The near-term focus sits in two places at once. The veterinary PetCam product generates early revenue with shorter regulatory lead time, while the human programme advances through clinical studies that will feed MDR 2017/745 certification in the European Union and FDA 510(k) or Breakthrough Device review in the United States. As of early 2026 the company had raised approximately $4.05 million across seed, grant and angel rounds, with a team of over 40 staff and a medical board responsible for image labelling.

Hardware design and AI are already strong: published accuracy is reported at up to 97 percent on the AI platform, and the OVHcloud partnership with NVIDIA H100 GPUs has shortened an experimental processing cycle from 55 days to 5 days. The next layer is the one that turns a working prototype into a regulated product: a documented validation evidence chain, a managed cloud operating model, integration with hospital LIMS (Laboratory Information Management Systems), EHR (Electronic Health Record) and PACS (Picture Archiving and Communication Systems) environments, and a clinical-trial data model that can be queried for regulatory submission and for AI retraining.

Capacity is also expanding along the hardware roadmap. Future capsule generations are expected to add narrow-band imaging, contrast studies and biopsy capability, each of which moves the device up the MDR risk classification and adds to the validation evidence that must be produced and re-produced through the product's life.

Challenges we see

  • Compliance Regulatory

    Producing regulatory evidence for an AI medical device under MDR and FDA

    BioCam is preparing for EU MDR 2017/745 certification and an FDA 510(k) or Breakthrough Device pathway in parallel, with limited Notified Body capacity across the industry and additional rules from the EU AI Act applying to AI-driven devices.

    Where the AI component is itself a regulated element, the evidence chain has to be assembled and refreshed continuously across the device life rather than produced once at submission, which puts ongoing documentation on the critical path.

  • Digital Integration

    Generalising AI across diverse patient populations and pathologies

    Each examination yields roughly 74,000 images, and the algorithm has to keep its diagnostic accuracy across Crohn's disease, celiac disease, small-bowel tumours, unexplained anaemia and the long tail of incidental findings.

    As clinical studies expand to broader demographics, the labelled image set becomes the bottleneck, and the rate at which new labelled data enters training drives the rate at which the model's claimed accuracy holds up.

  • Operations Manufacturing

    Sourcing specialised components for capsule manufacturing

    The pandemic-era semiconductor shortage forced BioCam to develop a more proprietary hardware platform around specialised CMOS (complementary metal-oxide-semiconductor) image sensors and STMicroelectronics microcontrollers.

    Designing for component availability as well as performance means qualification work has to be redone when a part changes, so lifecycle and supply-chain visibility become part of the validation evidence rather than a procurement detail.

  • Digital Operations

    Running the cloud platform economically as it scales

    The OVHcloud partnership provides NVIDIA H100 GPUs for AI training, and an experimental processing cycle that previously took 55 days now completes in 5 days, but commercial scale multiplies compute and storage demand without a proportional revenue line in place yet.

    Where compute and storage costs scale with usage rather than revenue, the platform's operating model rather than its peak performance increasingly sets the unit economics of every examination.

  • Operations Integration

    Coordinating image labelling across the medical board and clinical partners

    A medical board of more than 40 practitioners produces the labelled image set that trains the algorithms, and expansion of the training set depends on coordinated work across multiple clinical partners under consistent annotation protocols.

    When labelling throughput and consistency are the binding constraint, the tooling that supports the medical board drives the trajectory of the AI as much as the model architecture does.

Opportunities, by urgency and business impact

Each bubble is one opportunity, numbered to match the list below. Further right means it bites sooner; higher means a bigger effect on the business. A bigger bubble means a bigger implementation effort.

Source: A4BEE analysis of public sources

  1. Prioritising physician review on the segments the AI flags

    Conventional capsule endoscopy asks the clinician to scan thousands of images per examination, which is slow and exposes the review to fatigue.

    An AI-prioritised review surface that surfaces flagged segments first, attaches the diagnostic reasoning and presents biological KPIs (key performance indicators, here quantitative metrics such as lesion density or transit time) alongside the image lets the physician spend their time on the cases that need it.

    • BioCam AI platform documentation
    • NIH, Integration of Artificial Intelligence-Enhanced Capsule Endoscopy in Clinical Practice

  2. Generating first drafts of regulatory and clinical documents from source data

    MDR technical files, 510(k) submissions, clinical study reports and post-market surveillance summaries are produced on a recurring basis, and the source material already lives in the engineering, quality and clinical systems.

    Narrow agents that draft from those source records, check completeness against the relevant template and trace each statement back to the data behind it shorten the document cycle without removing the named reviewer from the path.

    • BioCam research project record, NCBR (Polish National Centre for Research and Development)
    • BioCam 2025 funding profile, PitchBook

  3. Running the telemedicine platform as a managed cloud service

    Compute and storage demand grow with each new clinical trial and commercial deployment, and current infrastructure is already running AI training on NVIDIA H100 GPUs at OVHcloud.

    Treating the platform as a managed cloud service with container orchestration, autoscaling and cost telemetry lets BioCam keep clinical and research workloads on the same estate while making the unit cost of an examination legible to the finance team.

    • OVHcloud BioCam case study
    • BioCam 2025 funding profile, PitchBook

  4. Connecting the platform to hospital LIMS, EHR and PACS systems

    Hospitals run their own LIMS, EHR and PACS estates, and adopting a new diagnostic device usually means integration work at every site.

    An HL7 FHIR (Health Level Seven Fast Healthcare Interoperability Resources, the standard for exchanging clinical data) gateway with DICOM (Digital Imaging and Communications in Medicine, the standard for medical image storage and transfer) interfaces and a vendor-agnostic abstraction layer makes the BioCam platform attachable to a typical hospital IT landscape with a known interface rather than a project.

    • BioCam Products and About Us pages
    • HL7 FHIR R4 specification, DICOM standard

  5. Harmonising multi-site clinical trial data into one queryable model

    Clinical studies run across multiple sites, each producing data in its own structure, and the same dataset has to serve both regulatory submission and ongoing AI retraining.

    An ontology-driven clinical data model with GxP (Good 'x' Practice, the umbrella of quality regulations covering laboratory, manufacturing and clinical practice) controls and audit-grade lineage lets the trial data be queried once for both purposes and lets new sites attach to the model rather than to a custom extract.

    • BioCam company documentation, data science team and medical board
    • EU MDR 2017/745, Annex XV on clinical investigations

What we'd propose

  • Digital Lab

    AI-prioritised capsule endoscopy review surface

    An image review application that presents AI-flagged segments first, attaches the diagnostic reasoning and renders biological KPIs alongside the image so the physician works down a prioritised queue rather than through raw footage.

    • CADe and CADx pipeline integration

      Detection and classification at the surface

      Surface the existing computer-aided detection (flagging what looks abnormal) and computer-aided diagnosis (classifying what the flag means) outputs as the primary review path, with confidence attached to each finding so the physician can see what the model is sure of and where to look closely.

    • Anatomy-aware review navigation

      Move by tract, not by frame number

      Lay the review interface over an anatomical map of the gastrointestinal tract so navigation moves by organ and region rather than by frame, putting each flagged segment in the context of the surrounding tissue.

    • Biological KPI overlay

      Quantitative metrics on the same screen

      Render quantitative metrics such as lesion density, transit time and pathology severity alongside the image so the review surface carries diagnostic context, not only visual evidence.

    • Review time per examination falls because the physician works the prioritised queue first.
    • The AI's confidence travels with each finding, making the human review faster to interpret.
    • Quantitative context is on the same screen as the image, so the report is built rather than assembled.

    We've done this before

    See how we do Digital Lab
  • Digital Lab

    Healthcare integration platform for LIMS, EHR and PACS

    An HL7 FHIR gateway with DICOM interfaces and a vendor-agnostic abstraction layer so the BioCam platform attaches to a typical hospital IT landscape through a known interface rather than a site-specific project.

    • HL7 FHIR gateway

      Clinical data exchange in a standard form

      Implement HL7 FHIR R4 resources for patient demographics, diagnostic reports and clinical observations so the platform exchanges clinical data with hospital EHR systems in a documented, standards-based way.

    • DICOM worklist and storage

      Imaging flows into the radiology path

      Add DICOM worklist and storage interfaces so capsule endoscopy images and findings reach the radiology workflow and the institution's PACS without bespoke image-format work at each site.

    • Vendor-agnostic integration layer

      One integration per hospital pattern, not per vendor

      Build an abstraction layer over the integration so a single deployment connects to multiple EHR and LIMS vendors, which keeps each new hospital site on a known engagement rather than a greenfield integration.

    • Hospital adoption follows a known interface pattern instead of a per-site integration project.
    • Diagnostic reports and images arrive in the systems the clinical team already uses.
    • Future sites attach to the same gateway rather than commissioning their own.

    We've done this before

    See how we do Digital Lab
  • Enterprise AI

    Managed cloud operating model for the telemedicine platform

    A managed cloud operating model for the BioCam telemedicine stack: container orchestration, autoscaling, cost telemetry and a change-management discipline that lets research and clinical workloads share the same estate.

    • Container orchestration for AI and clinical workloads

      Independent scaling per workload

      Run AI inference, image storage and the user-facing application as independent containerised services so each scales on its own demand curve rather than the platform being sized to the busiest workload.

    • Cost and health telemetry

      Unit cost of an examination, on the dashboard

      Stream GPU utilisation, storage growth and API latency into a single operations view so engineering, clinical operations and finance all see the same numbers on the same wall.

    • Change management with rollback

      Updates that can be undone

      Apply a structured change-management process to AI model updates and platform changes, with versioned rollouts and documented rollback paths so a clinical deployment is never the first place a change is exercised.

    • Research and clinical workloads share infrastructure without competing for capacity.
    • The unit cost of an examination becomes legible, not a quarterly surprise.
    • AI model updates reach production on a known process rather than on a one-off script.

    We've done this before

    See how we do Enterprise AI
  • Enterprise AI

    Clinical trial data platform with regulatory-grade lineage

    An ontology-driven clinical data platform that harmonises data from multiple trial sites, holds GxP controls around it and exposes it both for regulatory submission and for AI retraining.

    • Clinical data ontology

      One schema for every site

      Define the clinical entities (patient, site, visit, sample, finding, image) once, in a documented ontology, so data from different sites loads against the same model rather than being reconciled per report.

    • GxP validation and audit lineage

      A regulated path through the data

      Apply GAMP5-aligned (Good Automated Manufacturing Practice, the ISPE framework for validating computerised systems) validation and 21 CFR Part 11 controls to the data path so every record has the provenance a regulatory submission needs.

    • AI retraining dataset extraction

      Submission data and training data, the same data

      Expose the harmonised data set to AI retraining pipelines under version control and provenance, so a model trained on it carries the same audit evidence the regulators would see.

    • One data set serves both regulatory submission and AI retraining, with consistent lineage.
    • New clinical sites attach to the ontology instead of producing yet another schema.
    • GxP controls travel with the data, so audits are answered from the platform rather than reconstructed.

    We've done this before

    See how we do Enterprise AI
  • Agents

    AI agents for regulatory and clinical document work

    Narrow, reviewable agents that draft MDR technical files, 510(k) sections and clinical study documents from the underlying engineering, quality and clinical systems, check each draft against its template and leave a named reviewer in the approval path.

    • First drafts from source records

      Documents that start from system data

      Generate the first draft of a regulatory or clinical document directly from the source records in the engineering, quality and clinical systems, so the author edits and judges rather than assembles.

    • Template and completeness checking

      Gaps caught before review

      Check a submitted document against the relevant template and the organisation's checklist, returning missing or inconsistent sections before the document enters the human review queue.

    • Change impact search across the document set

      Which documents a change touches

      When a standard, method or specification changes, retrieve every controlled document that references it and rank them by how directly they are affected, so the scope of the update is known on day one.

    • Document cycles shorten because the first draft is produced from the source data.
    • Review queues receive documents that already match the template.
    • Every output is traceable to the source records it came from and signed off by a named reviewer.

    We've done this before

    See how we do Agents

Digital maturity: today and target

Scored out of 100 across six dimensions. The target is what BioCam's own published ambition implies — not a perfect score.

Source: A4BEE analysis of public sources
AI / ML operations 65 → 90
Algorithm development is the strongest part of the platform today, with reported accuracy of up to 97 percent. The next layer is MLOps (machine learning operations, the discipline of running ML models in production with versioning, monitoring and retraining discipline) discipline: model versioning, monitoring and a continuous retraining loop that ties new labelled data to model updates.
Cloud infrastructure 55 → 85
The OVHcloud partnership with NVIDIA H100 GPUs is in place and has produced a measurable speed-up in experimental processing. Commercial scale adds the need for cost telemetry, autoscaling and a documented operating model that covers both research and clinical workloads.
Device connectivity 70 → 90
Proprietary wireless transmission is functional and validated in environments mimicking real conditions. Standardised integration with hospital LIMS, EHR and PACS estates is the gap to close before wider clinical adoption.
Data platform 50 → 85
Clinical trial data is held across multiple sites and systems, and the medical board's labelled image set is managed outside a unified model. A single ontology-backed platform for both trial data and training data is the next step.
Quality and compliance 45 → 80
MDR 2017/745 certification in the EU and FDA review in the United States are in progress. AI components will be regulated elements under both frameworks, which raises the bar for validation evidence generated and refreshed continuously through the device life.
Cybersecurity 40 → 75
Healthcare data handling brings the usual production security requirements on top of the development focus that has brought the platform this far. Cloud-hosted patient data and connected device software both require the controls to be in place before commercial scale.

Check this yourself

Our Service Portal has free self-assessments and market comparisons. These are the ones that line up with what we've read above — no sales call required.

Think we've read this right?

Talk to us

Related reading

This is an independent analysis prepared by A4BEE from publicly available information as of January 2026. It reflects A4BEE's own interpretation and opinion, is not affiliated with, endorsed by, or verified with BioCam, and may be incomplete or inaccurate. All company names and trademarks are the property of their respective owners. To request a correction or removal, contact [email protected].