The science, stated precisely

In a field full of single-modality models that overclaim, precision is the point. Here is exactly what Axiisium reads, what it defers, and how each output is made verifiable.

Why one modality is not enough

Acute myeloid leukemia is a multimodal disease. The genetics that drive treatment are partly written in the cells and partly invisible to morphology. A model that reads only images, or only flow, leaves the deciding signals on the table. Axiisium fuses pathology, flow cytometry, cytogenetics, molecular, and clinical data into a single assessment, so each signal covers the blind spots of the others.

What morphology carries, and what it does not

The honest boundary is the credibility. Axiisium reads the markers the literature supports and refuses the ones it does not, deferring those to the molecular assay by design.

What Axiisium produces

For a single case, Axiisium fuses the available signals into a calibrated, confidence-bounded output, and signs it. For pharma, that is a ranked sequence-first list. For the clinic, it is a provisional read on day zero, an ELN-2022 risk class once genetics return, and a therapy-eligibility map, every claim tied to the signal that triggered it. The model shows its evidence and its confidence on every line.

The accountability layer

Every output is a signed, tamper-evident record, attributable to a named clinician and independently verifiable, powered by Project AIR. To be precise about what that buys: it makes the evidence auditable across the lifecycle and gives partners and regulators a verifiable chain of custody. It is not a substitute for analytical and clinical validation, which Axiisium pursues on its own track. We never conflate a trustworthy record with a validated prediction.

Built on real data

Axiisium is grounded in expert-labeled public cytomorphology data and validated against published cohorts with known molecular status, then extended toward multi-site, regulatory-grade evidence through clinical design partners. We report the level of evidence behind every claim, and we do not publish a performance number we have not earned.