<scp>ME?Net</scp> : <scp>Multi?encoder</scp> net framework for brain tumor segmentation

MRI plays a vital role to evaluate brain tumor diagnosis and treatment planning. However, the manual segmentation of image is strenuous. With development deep learning, large number automatic methods have been developed, but most them stay in 2D images, which leads subpar performance. Aiming at segmenting 3D MRI, we propose model for with multiple encoders. Our reduces difficulty feature extraction greatly improves We also introduced new loss function named “Categorical Dice,” set different weights segmented regions same time, solved problem voxel imbalance. evaluated our approach using online BraTS 2020 Challenge verification. proposed method can achieve promising results compared state-of-the-art approaches Dice scores 0.70249, 0.88267, 0.73864 intact tumor, core, enhancing tumor.