Predictive Value of Blood Markers in Pediatric Brain Tumors Using Machine Learning

Abstract

<b><i>Background:</i></b> The brain tumor is the most common solid tumor in children. Blood markers in most malignancies are altered due to the effect of inflammatory mediators on the bone marrow. <b><i>Objective:</i></b> This study aimed to predict the malignancy of pediatric brain tumors using blood markers. <b><i>Methods:</i></b> The pediatric brain tumors were divided into benign and malignant groups. Blood markers, including RBC, WBC, neutrophil, lymphocyte, monocyte, platelet, neutrophil to lymphocyte ratio, lymphocyte to monocyte ratio (LMR), platelet to lymphocyte ratio, and derived neutrophil to lymphocyte ratio were extracted. Differences in blood markers between two groups were assessed using statistical analysis. The accuracy of machine learning to determine pediatric brain tumors’ malignancy was evaluated using blood markers and demographic information. <b><i>Results:</i></b> Among 113 patients, 55 patients were in the benign tumor group, and 58 patients were in the malignant tumor group. In the statistical study of blood markers in two groups, LMR was significantly different and positively correlated with malignancy. Other blood markers were not significantly different between two groups. This study showed that support-vector machines using blood markers, age, and sex can differentiate benign and malignant pediatric brain tumors with 71.6% accuracy. <b><i>Conclusions:</i></b> Despite the statistically significant differences in blood markers in different grades of brain tumors in adults, their differences in pediatric brain tumors, except LMR, were not significant. Machine learning using blood markers can differentiate between benign and malignant pediatric brain tumors with 71.6% accuracy.