Diagnostic Accuracy of Ambient Mass Spectrometry with Blood Plasma in a Murine Glioma Model Using Machine Learning

Malignant glioma progresses rapidly and shows poor prognosis, but clinically applicable blood plasma-based biochemical tumor markers remain lacking. This study aimed to develop a diagnostic system using probe electrospray ionization mass spectrometry (PESI-MS) and a machine-learning logistic regression model to detect plasma changes at various time points in a murine glioma model. We used a syngeneic intracranial orthotopic murine model with GL261 glioma cells. Blood plasmas were collected before and 3, 7, and 14 days after intracranial transplantation of glioma cells (tumor group, n = 7) or injection of phosphate-buffered saline (control group, n = 8). Mass spectra from those samples were obtained using PESI-MS and compared between control and tumor groups. We explored changes in mass spectra at the 3 time points (3, 7, and 14 days) after transplantation. The performance of machine-learning logistic regression-based diagnosis algorithm was evaluated to clarify the potential utility for early diagnosis. Sixteen significant mass spectrum peaks were identified between the tumor and control groups. Multiple logistic regression analysis revealed 5 key mass spectra, achieving sensitivity of 0.875 and specificity of 0.943 for tumor discrimination. The area under the receiver operating characteristic curve was 0.981, outperforming analyses of individual spectra. These results indicate that PESI-MS combined with machine learning-based diagnostics in blood plasma could be a promising approach to accurate detection of malignant glioma.