Abstract
Differentiation between grades I and II Meningioma can be challenging in terms of radiologic features. Hence, this study uses radiomic features of 3D T1 enhanced MRI to analyze the difference between grades I and II Meningioma. A pre-operative contrast enhanced T1 images was analyzed retrospectively for 47 patients (16 grade II patients and 31 grades I). The grading of the meningioma was determined by a histo-pathologist who received samples from a neurosurgeon. Life X software which is designed to characterize tumor heterogeneity based on histogram, textural, and shape indices was used to extract the tumor features at a one and five voxel distance from neighbors. The texture indices were generated, and comparisons were made using the Weka software, which was used to calculate the sensitivity, specificity, F-score, AUROC and SPSS software was used to differentiate between WHO grades I and II of meningioma. For our dataset which consisted of 47 lesions of meningioma (31 lesions of grade I and 16 cases of grade II), For five voxel distance between neighbors, out of 17 texture features only 11 was statistically significant, namely, conventional-skewness, discretized-mean, discretized-q1, discretized-q2, discretized-q3, discretized-Skewness, GLRLM-LRHGE, GLRLM-HGRE, GLRLM-SRHGE, GLZLMHGZE, and GLZLM-SZHGE (p-value <= 0.05). A random forest classifier was built after choosing the top 16 highly correlated features had a weighted sensitivity, specificity, F-score, and AUROC of 74.5%, 65.7%, 74%, and 79% respectively. For one voxel distance between neighbors, out of 17 texture features only 7 were statistically significant , namely, discretized mean, discretized_q1, discretized_q2, discretized_q3, GLRLM_LRHGE, GLRLM_HGRE, GLRLM_SRHGE were significant (p-value <= 0.05). The classifier had a weighted sensitivity, specificity, F-score, and AUROC of 78.7%, 70.4%, 78.7%, and 80%, respectively. Our results suggest that the analysis of texture and shape is potentially useful preoperative tool for WHO grades I and II differentiation of meningioma. Out of the several hundred radiomic texture characteristics extracted from the image, few features can be used for the clear discrimination between WHO grades I and II preoperatively.