An AML diagnosis turns on the blast percentage. But "blast" is not "immature cell": count the wrong ones and you push a patient across a treatment line. And the line itself moves. A single defining mutation can make 8% blasts a cancer, or leave 25% as something else. The five signals that settle it are scattered across four instruments and the patient's chart, and rarely reach one model together.
Bone-marrow and blood-smear cell morphology, read by the image model.
Flow cytometry: lineage, the line between a myeloid and a lymphoid leukemia.
Karyotype and FISH: structural fusions morphology cannot see.
Mutation and fusion status: NPM1, FLT3, the defining lesions.
Age, labs, and history that frame the read.
A real workup does not arrive at once. Morphology and flow come back first; the genetics take days. Every other tool guesses in the gap. Axiisium states a provisional call, names what it is waiting for, and defers to the assay, exactly while the assay can still change the answer. Each step is signed, linked to the one before it, and anchored.
Fifteen percent sits in the gap: WHO 2022 needs 20%, ICC 2022 calls 10–19% a distinct MDS/AML category. Neither is a diagnosis yet. Axiisium says so, and names exactly what would settle it: molecular.
Flow confirms myeloid lineage, so this is not ALL. But lineage does not settle AML versus MDS. The call still waits on the genetics.
It does not fill the gap with a guess. It holds the call provisional only while a pending signal could still change it, and hands the decision to the physical assay. That restraint is the product.
NPM1 is AML-defining. WHO 2022 waives the 20% rule; ICC's 10% cutoff is met at 15%. The call turns over to AML with NPM1, and every step that got here is on the record, in order, unaltered.
Axiisium reads the morphology signal from images and takes the other four as structured inputs. It counts blasts the clinically correct way, excluding the maturing precursors most tools miscount, then applies both WHO 2022 and ICC 2022 and shows exactly where a defining mutation overrides the blast threshold.
The blast percentage counts myeloblasts and blast equivalents over nucleated leukocytes, and excludes erythroblasts and precursors like promyelocytes. Counting those inflates the number and can flip a borderline case. Axiisium uses the clinically correct definition.
Under WHO 2022 a defining lesion waives the 20% requirement entirely; ICC 2022 uses 10%. Axiisium returns the call under both systems and shows where they diverge, so the molecular result, not a hard-coded number, decides.
Every other model in this space hands you a number you have to take on faith. Axiisium hands you one you can check, without trusting us.
Each decision, and each change to it, is bound to a tamper-evident record: the exact model version, the inputs, the attributed clinician, cryptographically signed and linked to the one before it. The ledger runs in production and is anchored to a public, append-only transparency log, with a checkpoint signed by a key that never leaves hardware. Change one byte and verification fails.
Powered by Project AIRAML was reclassified in 2022, and WHO and ICC still disagree on where the line sits. Axiisium counts the blasts correctly, applies both rulesets, and shows where a defining mutation overrides the threshold entirely. Most tools still hard-code one number from the old criteria.
WHO 2022 + ICC 2022The production ledger head is anchored on a public transparency log. Read the recorded hash and confirm it matches, with no access to Axiisium and no trust in us.