A Vindicara company · In development

Multimodal AI for acute myeloid leukemia.

Every model decision, signed and independently verifiable.

AML is read from five signals scattered across separate instruments. Axiisium fuses them into one calibrated call, and binds each prediction to a record a regulator and a clinician can verify.

NPM1 from morphology · reproduced
0.894 ±0.066
patient-level AUROC, leakage-free cross-validation
A model reproduces NPM1-from-morphology at literature level on patient-grouped public data. Validated, not a claim.
How it works

Five signals, fused.

AML is diagnosed from many signals at once, and the decisive ones rarely reach a single model together. Some genetics that drive treatment are written in the cells; some are not. Fusing the signals reaches the ones a single view cannot.

1

Pathology

Bone-marrow and blood-smear cell morphology.

2

Flow cytometry

Immunophenotype for classification and residual disease.

3

Cytogenetics

Karyotype and FISH signals morphology cannot see.

4

Molecular

Mutation and fusion status: NPM1, FLT3, the wider panel.

5

Clinical

Age, blast percentage, and labs that frame the prediction.

Two products, one engine

For the people who run AML trials, and the people who treat AML.

For pharma

Axiisium Enrich

Rank which trial candidates to sequence first for a target mutation, so a sponsor fills an NPM1-defined cohort with fewer assays and a signed record of every selection. It ranks, it never decides; sequencing always confirms; a recall floor bounds what it can miss.

Research-use today
For the clinic

Axiisium Decide

A signed assessment that evolves from day zero to final molecular: a provisional read while sequencing runs, an ELN-2022 risk class, and a therapy-eligibility map, each claim tied to the signal that triggered it. Decision support that defers the hard calls to molecular and the clinician.

On a De Novo device pathway
Live model

Drop in a cell. Get a signed call.

This runs the real morphology model on your image and returns a blast call with a calibrated confidence, then binds it to a tamper-evident, independently verifiable record. The molecular confirmation runs separately on the NVIDIA Clara Parabricks pipeline; morphology never calls the hard mutations on its own.

One click, real cells from the dataset:
▲ NVIDIA GPU inference · checking…
Awaiting a cell image…
Whole slide

Blast burden across a patient slide

A blast count over every cell on the slide is the number a clinician acts on. But the threshold that count is judged against is not fixed: it depends on the genetics. Run the slide for the morphology rung, then the four-rung diagnosis below applies the real WHO 2022 and ICC 2022 rules.
Four-rung diagnosis · MICM

Morphology, immunophenotype, cytogenetics, molecular: one signed call

Research use only. Decision support, not a diagnostic device. A qualified clinician makes the diagnosis.
The blast percentage alone does not make the call. Click run: Axiisium takes the live morphology blast burden plus the molecular, flow, and cytogenetic results, applies the WHO 2022 and ICC 2022 rules, and returns one signed, explained diagnosis. The molecular rung can override the blast threshold entirely.
Partner with us

Building in AML?

We are looking for clinical and pharma design partners in hematology. If you run acute myeloid leukemia trials or hold aligned multimodal data, we would like to talk.