Correlation Between Brain Functional Integrity And Tumor Biological Aggressiveness In Glioma Patients

Accurate and noninvasive preoperative grading of cerebral glioma is essential for the intervention planning.Structural,functional and metabolic information has been extracted from multi modal imaging data and has been demonstrated to be correlated with the malignancy degree of gliomas.However,malignancy degree is likely to be over-or underestimated based on solitary parameters.Multi-dimensional information is advantageous in improving glioma characterization but also brings challenge to researchers on how to make full use of the complex information.The introduction of artificial intelligence and machine learning theory offers an efficient maneuver for the preoperative aggressiveness estimation of glioma based on multimodal data,in which,various histogram and texture attributes of the tumoral region were constructed from MR images.However,the degree of the malignancy may not be fully represented by the local morphology-based features,as it has been widely demonstrated that the effect of the gliomas on the brain has extended beyond the tumor boundaries defined by neuroimaging.Thus,it is important to estimate the functional impairment of glioma at the global scale.Connectivity based approach has been increasingly accepted for the neuroscience researches in compensation of the morphology-based methodology.This study aims to investigate the correlation between the malignancy degree of cerebral glioma and the topological features of resting-state functional network,and estimate the malignancy degree of gliomas based on the topological features and machine learning classifier.Resting state Magnetic Resonance imaging data of 122 right-hand subjects with histologically confirmed glioma was retrieved and preprocessed by DPARSF.GRETNA was employed to construct the functional networks and analyze the topological characteristics of networks.Mann-Whitney U test was conducted to extract the significant features to optimize the support vector machine learning model.Finally,the optimized SVM model achieved 84.31% in accuracy,77.36% in sensitivity and 91.84% in specificity in the malignancy classification of the glioma based on the topological features of the resting state cerebral functional networks.This study proved that the contralesional hemispheric functional networks preserved small worldness and the introduction of the machine learning and graph theory provides a pathway promising for multidimensional imaging data analysis toward a better tumor characterization and treatment response assessment of glioma.