Medical images are one of the important sources of information describing lesions in patients and have been widely used in clinical practice. 3D medical images can comprehensively and finely describe lesion information. Medical image segmentation can effectively assist clinicians in diagnosing and managing lesions in patients. Traditional CNNs have insufficient contextual feature capture ability for 3D medical images, while Transformers have inadequate feature fusion capabilities for different types of 3D medical images. To address these issues, we propose Dual Encoder U-Net (DEU-Net), which can simultaneously use CNNs and Transformers to extract and fuse features from different types of 3D medical images. We propose a Dual Feature Fusion Module (DFFM) to fuse the different features extracted by the two encoders from different 3D medical images. We train and validate the model on the BraTS 2021 dataset, and by comparing it with other SoTA methods, we find that our proposed model improves the accuracy of 3D medical image segmentation through the feature fusion of 3D medical images.