Riskformer: Survival prediction from MR imaging in patients with IDH-wildtype glioblastoma.

• Deep learning-based features can provide additional information beyond the scope of human visual perception. • We developed and validated a vision transformer-based model, named Riskformer, from preoperative MRI. • Riskformer score could provide independent and incremental prognostic value over existing clinical factors in OS prediction. Glioblastoma is the most common malignant brain tumor in adults. Despite aggressive treatment, its prognosis is still poor, with a median overall survival of less than 15 months. In this study, we developed and validated a vision transformer-based model, named Riskformer, built from preoperative MRI for predicting OS in patients with newly diagnosed IDH-wildtype glioblastoma patients. Riskformer score was associated with OS as an independent prognostic factor in patients with IDH-wildtype glioblastoma. In the training dataset (hazard ratio [HR]: 5.930, 95% confidence interval [CI]: 3.933–8.942, P < 0.001), validation dataset (HR: 2.436, 95% CI: 1.334–4.447, P < 0.001) and testing dataset (HR: 4.651, 95% CI: 2.233–9.688, P < 0.001). The Riskformer score could provide independent and incremental prognostic value over existing clinical factors in OS prediction in patients with IDH-wildtype glioblastoma. [ABSTRACT FROM AUTHOR]