Sturgeon

von UMC Utrecht

AI Application

Niedrige Vertrauensstufe

Sturgeon is software that classifies tumours of the central nervous system from DNA methylation data while the patient is still in surgery. It is used in a hospital laboratory next to rapid nanopore sequencing of a tumour sample taken in the operating theatre. Its developers at UMC Utrecht say a transfer-learned neural network assigns a molecular tumour class from the sparse methylation profile that a short sequencing run yields, with a score that is updated as more data arrive; in their Nature study it gave a result within 90 minutes in 25 live surgeries, 18 correct and 7 below the confidence threshold. The developers treat scores below 0.8 as inconclusive. The result is read together with the neuropathologist's frozen-section diagnosis, and the neurosurgeon uses it to choose the surgical strategy. Sturgeon runs as a locally installed Python package or binary; no hosting is described. The code is on GitHub under an evaluation licence that bars clinical use without a licence from Cyclomics.