Automatic Tumor Segmentation Based on Convolutional Neural Network and Image Gradient Enhancement

In this paper, we propose an automatic tumor segmentation method combining convolutional neural networks with image gradient Enhancement-Enhancement Characteristic U-Net (ECU-Net). This approach integrates the strengths of UNet, generative adversarial networks (GANs), and Image Gradient Enhancement (IGE) to enhance the accuracy of tumor segmentation. First, U-Net extracts image features through its encoder-decoder structure to achieve preliminary segmentation of tumor regions. Secondly, the GAN generator produces synthetic images that closely resemble real tumor images, while the discriminator is employed to differentiate between real and synthetic images. This process thereby enhances the model's capability to learn tumor features. Finally, the image gradient enhancement technology is used to further optimize the segmentation performance and improve the processing ability of the model for complex background and low contrast images. The experimental results show that ECU-Net performs well on multiple tumor image datasets, especially when dealing with complex backgrounds and low-contrast images. This result proves the reliability of ECU-Net framework in automatic tumor segmentation, and also provides a new technical idea for the field of automatic tumor segmentation.