
npj Digital Medicine, Journal Year: 2025, Volume and Issue: 8(1)
Published: March 5, 2025
Molecular subtyping and grading of adult-type diffuse gliomas are essential for treatment decisions patient prognosis. We introduce GlioMT, an interpretable multimodal transformer that integrates imaging clinical data to predict the molecular subtype grade according 2021 WHO classification. GlioMT is trained on multiparametric MRI from institutional set 1053 patients with IDH mutation status, 1p/19q codeletion tumor grade. External validation TCGA (200 patients) UCSF (477 shows outperforms conventional CNNs visual transformers, achieving AUCs 0.915 (TCGA) 0.981 (UCSF) mutation, 0.854 0.806 codeletion, 0.862 0.960 prediction. enhances reliability decision-making by offering interpretability through attention maps contributions data.
Language: Английский