Attention-Guided Brain Tumor Segmentation Using Multimodal MRI

Abstract

The segmentation of brain tumors in multimodal magnetic resonance imaging (MRI) is an important task in the clinical diagnosis, treatment planning and disease monitoring processes. The accurate delineation is however difficult because of the heterogeneity of tumors, irregularity of their borders and the different intensity distributions on various imaging modalities. In this work an Attention-Guided Brain Tumor Segmentation framework is presented, that relies on complementary information extracted from T1, T1ce, T2 and FLAIR MRI sequences for accurate tumour localization and segmentation. The architecture proposed combines various modalities, and uses attention guided encoder to enhance the representation of discriminative features and to remove irrelevant background information. Data quality is enhanced and the model generalizes better with a comprehensive preprocessing pipeline, which involves skull stripping, intensity normalization, image resizing and data augmentation. The experimental results demonstrate that the proposed attention-guided model offers more accurate segmentation, preservation of boundaries and robustness compared with the traditional deep learning model. Besides, whole tumor segmentation, segmentation of tumor core and enhancing tumor regions are realized with good results by combining multimodal MRI features. In summary, the proposed method is clinically applicable and is a reliable, effective method for automatic brain tumor segmentation, which aids in the better diagnosis and treatment planning.