Research on Brain Tumor Segmentation and Identification Technology Based on Improved Unet

Brain tumors are common neurological diseases that severely threaten human health and even life. The accurate extraction of brain tumor information through Magnetic Resonance Imaging (MRI) is crucial for assisting diagnosis. Manual segmentation of brain tumor subregions is time-consuming, prone to error, and highly subjective. Therefore, it is necessary to design accurate, efficient, and reproducible automatic segmentation methods. This paper proposes an improved Unet model for three-dimensional brain tumor image segmentation. First, given the variability of three-dimensional brain tumor images, we downsample the Unet model three times to extract image features, which not only reduces the number of model parameters but also retains the target edge features to the maximum extent. Second, to address the class ambiguity caused by the reduction of deep features, we introduce a Multiple-Attention based Feature Pyramid Network (MA-FPN) module in the encoding and decoding parts of the model to extract multi-scale information from images and enhance feature expression. Finally, we add an edge extraction module based on the basic encoder. This module combines the semantic feature information extracted by the backbone network and the edge feature information obtained from the input image through Canny operator and dilation mathematical morphology operation, thereby better learning the edges of three-dimensional brain tumor images. The comparative experimental results show that our proposed model significantly improves the accuracy of brain tumor segmentation compared to traditional Unet, PSPNet, ICNet, and Fast-SCNN, with increases of 3.9%, 1.3%, 5%, and 3.9% respectively.