Multi-view fusion segmentation for brain glioma on CT images

摘要

Computed tomography (CT) images of the brain aid the radiotherapy of glioma. The identification and contouring of the gross tumor volume (GTV) are important for radiotherapy. However, manual segmenting GTV is time-consuming, exhausting, and subjective, and automated methods are desired. To overcome these shortcomings, a novel neural network framework based on multi-view fusion is proposed to segment GTV in brain glioma automatically. The multi-view image that includes the previous image, current image, and following image is inputted in this framework to abstract extra spatial features and then aggregated to segment the GTV. Compared with the 2D segmentation framework, the proposed framework retains more spatial information due to the multi-view image. Meanwhile, compared with the 3D segmentation framework, the proposed framework considers fewer images, which means the model has fewer parameters and is easier to train while retaining much useful spatial information. Moreover, the GliomaCT dataset, a large CT dataset collected from West China Hospital, is used to train, validate, and test the proposed method. The performance of the proposed method and other state-of-the-art methods are compared on this dataset. The high dice similarity coefficient achieved in the experiments demonstrates the effectiveness of the proposed method for segmenting the GTV in brain glioma.