Life sciences
Three problems in clinical detection.
Resistance, risk and imaging. Different data, different workflows, different regulatory paths. What they have in common is that all three were built to run where the clinical work actually happens.
AMR responsible-protein detection
Which protein is defeating the drug, read from the sequencing data rather than looked up in a catalogue.
Read more →PredictionEarly cancer risk prediction
Cervical and breast cancer risk, identified ahead of presentation and used to set screening priority.
Read more →ImagingRadiology AI
Detection from diagnostic imaging, built to sit inside the reporting workflow rather than beside it.
Read more →What we will and will not put on a website
Capability here. Evidence under agreement.
All four of our technologies have patent applications pending. Study design, cohort composition, comparator method and model architecture are the parts that matter to an examiner, and publishing them is prior art. So they are not here.
What is here is what each system takes in, what it returns, where it can be deployed and how you would put it through a trial of your own. If you are a hospital, a laboratory or a screening programme, that is usually enough to decide whether the next conversation is worth an hour.
Under agreement we share the validation design, the performance data, the failure modes by subgroup, and the regulatory position in your market. We do not summarise those. You get the whole thing or none of it.
Getting from a model to a clinic
Most of the work is not machine learning.
This is the sequence every one of our diagnostic systems goes through, and the part buyers usually want to talk about first.
Intended use
The population the system is for, the question it answers, and the decision it is allowed to influence. Everything downstream is defined against this one paragraph, including the device classification.
Analytical validation
Performance against a reference method, on data held out from training, with failure modes characterised by subgroup.
Clinical validation
Performance in the population and the workflow where it will really be used, at the prevalence that population actually has.
Quality system
Design history, risk management, change control and post-market surveillance. Procurement normally asks for this before it asks about the clinical evidence, which surprises people.
Regulatory submission
Per market, under the relevant device framework. Classification follows intended use, not the technology.
Deployment and monitoring
Integration into the estate, model versions pinned per site, and performance watched for drift after go-live.
Clinical, laboratory and research enquiries.
Evaluation packages and integration detail are shared under agreement.