Glioblastoma is the most frequent and aggressive primary brain tumor. Its diagnosis is based on resection or biopsy that could be especially difficult and dangerous in the case of deep location or patient comorbidities. Monitoring disease evolution and progression also requires repeated biopsies that are often not feasible. Therefore, there is an urgent need to develop biomarkers to diagnose and follow glioblastoma evolution in a minimally invasive way. In the present study, we described a novel cancer detection method based on plasma denaturation profiles obtained by a non-conventional use of differential scanning fluorimetry. Using blood samples from 84 glioma patients and 63 healthy controls, we showed that their denaturation profiles can be automatically distinguished with the help of machine learning algorithms with 92% accuracy. Proposed high throughput workflow can be applied to any type of cancer and could become a powerful pan-cancer diagnostic and monitoring tool requiring only a simple blood test.