← BACK TO ZEKLAR HEALTH/health/kairos-medtech

Product· pharma

KAIROS

Q: Does KAIROS decide the dose instead of the doctor?

No. KAIROS is a decision-support and research tool: it estimates exposure, toxicity and efficacy on a virtual patient to aid hypothesis exploration. The clinical decision remains entirely with the treating physician.

Q: Is it a certified medical device?

No, and it is not presented as one. KAIROS provides neither diagnosis nor prescription and is not intended for use as a medical device; it is built for clinical research and decision support in a study setting.

Q: How reliable are the results?

Vascular segmentation reaches a 0.91 Dice score and simulations show 92% in-silico concordance against validation; these remain model estimates, to be interpreted in clinical context and not as absolute truth.

Q: Does patient data leave the hospital?

In on-premise and isolated modes, no: imaging and genomics stay within the clinical perimeter, integrated via DICOM to PACS. Only the research cloud configuration uses de-identified data with per-project isolation.

The pharmacokinetic digital twin for safer dosing.

KAIROS is a pharmacokinetic digital twin of the patient: it fuses genomic profile, anatomy and drug model to estimate exposure, toxicity and efficacy before administration. Change the dose or a variant and watch the therapeutic window shift in real time, on the virtual patient instead of the real one. It is a research and decision-support tool, not a medical device.

<4 h END-TO-END SIMULATION PER PATIENT0.91 VASCULAR SEGMENTATION DICE SCORE92% IN-SILICO CONCORDANCEADVISORY MODE · READ-ONLY<4 h END-TO-END SIMULATION PER PATIENT0.91 VASCULAR SEGMENTATION DICE SCORE92% IN-SILICO CONCORDANCEADVISORY MODE · READ-ONLY

01In figuresFIELD DATA

<4 h

end-to-end simulation per patient

0.91

vascular segmentation Dice score

92%

in-silico concordance

02The problem

The therapeutic window is often discovered on the patient.

Many high-risk therapies, chemotherapy first among them, are calibrated by trial and error: a wrong dose means severe toxicity or under-treatment. The data to do better already exists, genomic sequencing, high-resolution imaging, kinetic models, yet it sits in separate silos, never unified into a single predictive model. What's missing is the infrastructure to make it speak on one computational graph.

03What it does04

Therapeutic-window simulation

Estimates exposure, toxicity and efficacy for a given dose and visualises how risk moves as the regimen changes.

Patient-specific PK engine

Maps SNP and CNV variants onto a kinetic model calibrated per individual patient, not a population average.

Anatomy from imaging

A 3D segmentation reconstructs the vascular mesh from DICOM, so the simulation runs on the patient's actual anatomy.

Research-clinic-compatible turnaround

A directed-graph pipeline parallelises ingestion, segmentation and the CFD solver, cutting end-to-end simulation from a day to a few hours.

04How it works04 STEPS

Ingest

Genomic sequence, DICOM imaging and drug parameters enter the pipeline.

Reconstruct

3D segmentation builds the patient's vascular mesh from the imaging.

Simulate

The CFD solver and PK engine compute drug exposure and distribution.

Explore

The clinician varies dose and assumptions and reads the shift in risk and efficacy.

05Architecture

KAIROS is a directed-graph pipeline that brings genomics, imaging and the drug model onto a single computational graph. Ingestion, anatomical reconstruction, simulation and exploration are parallelised stages, so an end-to-end simulation that took a day drops to a few hours. It is a research and decision-support tool, not a medical device.

01Multi-source ingestionGenomic sequence (WGS and panels), DICOM imaging from hospital PACS and drug parameters enter the pipeline, orchestrated by Nextflow with provenance tracked for every input.

02Anatomical reconstructionA PyTorch3D 3D segmentation, accelerated with NVIDIA Clara, reconstructs the patient's vascular mesh from DICOM, so the simulation runs on the actual anatomy rather than an average model.

03PK + CFD solverThe patient-specific PK engine maps SNP and CNV variants onto a per-individual calibrated kinetic model, and a CUDA/A100 GPU CFD solver computes drug exposure and distribution across the mesh.

04Therapeutic-window explorationThe clinician-researcher varies dose and assumptions and reads in real time how the window shifts between toxicity and efficacy on the virtual patient, without ever acting on the real one.

06Specifications

End-to-end simulation per patient<4 h

Vascular segmentation Dice score0.91

In-silico concordance92%

Imaging inputDICOM (PACS)

Solver acceleratorCUDA / A100 GPU

Pipeline orchestrationNextflow

Genomic inputWGS / pannelli

ClassificationRicerca / non-dispositivo

07Deployment

Research cloud

Managed GPU environment (A100) for research centres and trial sponsors, with de-identified data and per-project isolation; suited to retrospective studies and large-scale cohort simulation.

Hospital on-premise

Deployment inside the hospital IT perimeter, integrated to PACS over DICOM, so patient imaging and genomic data never leave the clinical infrastructure. GPU compute can lean on a local MONOLITH.

Isolated / clinical-grade

Fully isolated configuration for highly sensitive clinical data: no external connectivity, model updates via signed media and processing logs kept within the institution's perimeter.

08Security & compliance

Standards and regulatory status

KAIROS is a research and decision-support tool, not a medical device, and it does not provide diagnosis or prescription. Data handling is aligned to ISO 27001 and to health-data practices (GDPR, and HIPAA where applicable).

Data sovereignty

In on-premise and isolated modes, DICOM imaging and genomic sequences stay within the hospital perimeter; the pipeline works on de-identifiable data and keeps provenance at the resource level for compliance.

Access and audit

RBAC separates researchers, clinicians and administrators; every simulation, explored dose variation and access to patient data is recorded in an immutable audit trail, essential for research reproducibility.

Encryption

AES-256 encryption at rest for imaging, genomics and simulation results, and TLS 1.3 in transit to PACS and between pipeline stages; keys remain within the institution's perimeter in isolated modes.

09Integrations

Hospital PACS over DICOMNVIDIA Clara (DICOM-to-tensor)Nextflow pipeline orchestrationPyTorch3D 3D segmentationCUDA / A100 GPU CFD solverWGS sequencing and genomic panels

10FAQ

Does KAIROS decide the dose instead of the doctor?