Automatic Differentiation of Grade I and II Meningioma on Magnetic Resonance Image Using an Asymmetric Convolutional Neural Network.

Purpose/Objective(s) The grade of meningioma has significant implications on the selection of treatment regimens ranging from observation to surgical resection with adjuvant radiation. For most patients, meningiomas are diagnosed radiologically and grade is not determined unless a surgical procedure is completed. We hypothesize that the use of a 3D convolutional neural network (CNN) architecture with two encoding paths will provide auto-classification of grade I and II meningioma using T1-contrast enhancing (T1-CE) and T2-Fluid attenuated inversion recovery (FLAIR) magnetic resonance (MR) images. Materials/Methods Ninety-six consecutive treatment naive patients with pre-operative T1-CE and T2-FLAIR MR images from 2010 to 2019 were evaluated. The patients underwent surgical resection resulting in a diagnosis of grade I or II meningioma. Histopathology was confirmed by a neuropathologist and neuropathology fellow based on current classification guidelines described in the 2016 World Health Organization (WHO) Bluebook. The delineation of meningiomas was completed on both T1-CE and T2-FLAIR MR with the inclusion of surrounding T2-FLAIR hyperintensity. A novel asymmetric 3D CNN architecture was constructed with two encoding paths based on T1-CE and T2-FLAIR MR. Each path used the same 3 × 3 × 3 kernel with different filters to weight the spatial features of each sequence separately. Final model performance was assessed by 10-fold cross-validation. Results Of the 96 patients, 55 (57%) were pathologically classified as grade I and 41 (43%) as grade II meningiomas. Optimization of our model led to a filter weighting of 18:2 between the T1-CE and T2-FLAIR MR image paths with alternative weighting shown in Table 1. Eighty-six (90%) patients were classified correctly and 10 (10%) were misclassified based on their pre-operative MRs with a model sensitivity of 0.85 and specificity of 0.93. Among the misclassified, 4 were grade I, and 6 were grade II. The model was robust to tumor locations and sizes. Conclusion A novel asymmetric CNN with two differently weighted encoding paths was developed for successful automated meningioma grade classification. Our model outperforms CNN using a single path for single or multimodal MR-based classification.