Retinal vessel segmentation plays a crucial role in the early diagnosis of diseases such as diabetic retinopathy and glaucoma. Methods based on convolutional neural networks have achieved excellent results in this field. However, standard convolutional neural networks can only capture local features without sensitivity to directional information and local feature differences. Meanwhile, simple skip connections cannot effectively fuse multi-scale features, making it challenging to capture the complex structure of retinal vessels. This paper proposes a differential feature fusion network called DiFFuse-Net to address these issues. Specifically, we introduce a Hybrid Differential Convolution Block that combines standard convolution with four different differential convolutions. This block can comprehensively capture various directional features and edge information of ocular vessels, enhancing the contrast between vessels and background. Furthermore, we design an Integrated Feature Fusion Module (IFFM) to integrate multi-scale information effectively. IFFM incorporates high-level and low-level features from different scales simply based on grouped aggregation, achieving robust feature representation. Quantitative and qualitative experimental results demonstrate that the proposed method outperforms the state-of-the-art methods in retinal vessel segmentation.