M&M (Minority & Majority classifier)

par Prinses Máxima Centrum

AI Application

Faible confiance

M&M (Minority & Majority classifier) is a machine-learning tool that suggests the tumour type and subtype of a childhood cancer from RNA-sequencing data of the tumour. It is used in a paediatric oncology diagnostic laboratory, where pathologists and haematologists make the diagnosis. Its developers at the Prinses Máxima Centrum say it compares each new patient's gene-expression profile with all profiles measured before, combining two ensembles of weighted random-forest models, one aimed at rare and one at more frequent tumour (sub)types. It covers 52 tumour types and 96 subtypes; the developers report about 99% precision for confident tumour-type calls on internal test data; calls below a probability threshold are flagged as low confidence. The Máxima describes it as a tool the pathologist or haematologist uses to confirm their own diagnosis or, when in doubt, to reach a precise one; the paper says its output should always be weighed against other information. The R package is on GitHub under GPL-3.0 and the trained models on Zenodo; hosting is not stated.