Research on the magnetic resonance imaging brain tumor segmentation algorithm based on DO-UNet

摘要

With the social and economic development and the improvement of people'sliving standards, smart medical care is booming, and medical image processingis becoming more and more popular in research, of which brain tumor segmentationis an important branch of medical image processing. However, themanual segmentation method of brain tumors requires a lot of time and effortfrom the doctor and has a great impact on the treatment of patients. In orderto solve this problem, we propose a DO-UNet model for magnetic resonanceimaging brain tumor image segmentation based on attention mechanism andmulti-scale feature fusion to realize fully automatic segmentation of braintumors. Firstly, we replace the convolution blocks in the original U-Net modelwith the residual modules to prevent the gradient disappearing. Secondly, themulti-scale feature fusion is added to the skip connection of U-Net to fusethe low-level features and high-level features more effectively. In addition, inthe decoding stage, we add an attention mechanism to increase the weight ofeffective information and avoid information redundancy. Finally, we replacethe traditional convolution in the model with DO-Conv to speed up the networktraining and improve the segmentation accuracy. In order to evaluatethe model, we used the BraTS2018, BraTS2019, and BraTS2020 datasets totrain the improved model and validate it online, respectively. Experimentalresults show that the DO-UNet model can effectively improve the accuracy ofbrain tumor segmentation and has good segmentation performance.