AI Medical
AI models for medical work
38 services across imaging, pathology, records, speech, genomics and discovery, run as managed AI pipelines in our GPU clusters — and, where medical governance requires it, on a dedicated private cluster in our cloud or installed on premise.
How these services run
Most customers run their pipeline in our GPU clusters: no hardware to buy, no capacity to plan, and a cost per study that is known before the work starts. For medical work we also offer a dedicated private cluster in our cloud, or an on-premise installation where your governance requires the data to stay in the building.
Tell us the modality, the volume, the accuracy target and any constraint on where the workload may run. We select the models, build the pipeline and size the GPU profile that meets the target at the lowest practical cost.
Every output is clinical decision support for a qualified professional to review and sign. Where clinical use requires regulatory approval in your jurisdiction, that approval remains yours to hold; we provide the infrastructure, the model operations and the audit trail behind it.
AI Medical services
Each service is a pipeline, not a single model: several models and methods are usually combined to hit an accuracy target at a workable cost per study.
Every service page lists the production models behind it — producer, input, output, reported accuracy and three-tier hardware sizing — so your technical team can see exactly what a deployment costs before anyone commits to it.
{{ filterNote }}
Diagnostic imaging
- CT analysis
- Embeddings, findings, triage scores and retrieval from CT volumes.
- MRI analysis
- Multi-sequence MRI tasks, and prostate lesion detection with reader-study evidence.
- Chest X-ray screening
- Finding probabilities, zero-shot search and encoders for your own endpoints.
- Chest X-ray reporting support
- Findings ranked for the worklist and drafted as structured text to sign.
- Breast cancer screening
- Malignancy scores per breast and image-derived one-to-five-year risk.
- Lung cancer screening
- One-to-six-year risk from a single low-dose CT, and nodule detection.
- Musculoskeletal imaging
- Fracture detection scoped to the body parts where accuracy holds, and bone age.
- MRI reconstruction and acceleration
- Diagnostic images from undersampled k-space — a quarter of the table time.
- Ultrasound analysis
- Classification, segmentation and enhancement with models built for the modality.
Specialty imaging
- Echocardiography analysis
- Ejection fraction, wall thickness and chamber function from echo video.
- ECG interpretation
- Rhythm and diagnostic labels from 12-lead and single-lead waveforms.
- Cardiac MRI analysis
- Ventricle and myocardium segmentation with volumes and ejection fraction.
- Brain MRI analysis
- Morphometry in an hour, plus tumour and MS lesion delineation.
- EEG interpretation and sleep staging
- Hypnograms at human-scorer agreement, and EEG embeddings.
- Stroke and haemorrhage detection
- Intracranial haemorrhage detection and volume on head CT.
- Retinal image analysis
- Fundus and OCT disease grading, vessel quantification, oculomics.
- Skin lesion analysis
- Lesion triage with concept-level explanation and label-efficient transfer.
- Digital pathology analysis
- Whole slides classified, searched and scored, with cell-level analysis.
- Endoscopy video analysis
- Polyp detection and segmentation in colonoscopy video.
- Surgical workflow analysis
- Phase, instrument and action recognition in laparoscopic video.
- Health acoustics analysis
- Cough, breath and lung sound screening, and spirometry interpretation.
Cross-modality foundations
- Medical image segmentation
- Organs, lesions and structures contoured across CT, MRI and ultrasound.
- Medical image retrieval
- One image-text index for search, zero-shot scoring and prior comparison.
- Medical visual question answering
- Answers, findings and draft text from images across many modalities.
Records, coding and language
- Clinical note summarization
- Discharge summaries, referrals and handovers drafted from the record.
- Medical record extraction
- Conditions, medications and coding candidates as structured fields.
- Clinical coding support
- Ranked ICD and CPT candidates, and SNOMED CT concept linking.
- Clinical text representation
- Encoders your own classifiers, extractors and indexes are built on.
- Clinical reasoning assistant
- Question answering over your own guidelines, with the passage cited.
- EHR outcome prediction
- Mortality, readmission and next-event forecasting from coded data.
- Clinical de-identification
- PHI removed from notes and DICOM, with a documented removal trail.
- Medical speech to text
- Consultations and dictation transcribed with speaker labels.
- Mental health text analysis
- Indicator detection with explanations, and human review of every risk signal.
Genomics and discovery
- Genomic sequence analysis
- Variant effect, regulatory prediction and phenotype-driven prioritisation.
- Single-cell analysis
- Cell-type annotation, batch integration and perturbation prediction.
- Protein structure prediction
- Structure and complex prediction, embeddings, and de novo design.
- Molecular property prediction
- ADMET and property prediction, and drug interaction mining.
- Binding affinity prediction
- Co-folding, affinity estimates and ranked docking poses.
How the work reaches the models
Work reaches the pipeline the way that suits your systems, over a link you control.
Imaging arrives from a PACS node, a DICOM store or a watched folder.
Records and documents arrive over your API, a message queue or an SCP drop behind your VPN.
Slides and recordings transfer over SCP behind a VPN, or on a NAS or DAS sent to our location.
For medical deployments we can also run the pipeline on a dedicated private cluster in our cloud, or install it on premise.
How results come back
Results return in the format your systems already read: DICOM segmentation objects and structured reports, HL7 or FHIR messages, JSON over your API, or files to a location you nominate.
Every job carries an audit record naming the model version that produced it, which is what makes a result defensible months later.
Other input types
- Pictures OCR, detection, segmentation, captioning, restoration. AI models for Pictures →
- Documents Extraction, classification, redaction, question answering. AI models for Documents →
- Audio files Transcription, diarization, separation, summarization. AI models for Audio files →
- Video files Tracking, transcription, subtitles, summarization. AI models for Video files →
- Live feed Counting, tracking, live transcription, PPE and queues. AI models for Live feed →
One study or a whole archive
Start with a proof of concept on a sample of your own data: real accuracy, real throughput, a real cost per study before anything is committed. The same pipeline then scales from one GPU to a dedicated cluster without changing the integration.