OrganVMUNet: A Novel Vision Mamba UNet-Based Network for Multi-Organ Segmentation in Abdominal CT Images
Year: 2025
Type: conference-paper
Abstract: The paper mainly focuses on how to use CT scan images to identify and divide abdominal organs, which is particularly important for doctors to diagnose the condition, develop treatment options, and surgical navigation. The commonly used segmentation method now uses a convolutional neural network or a transformer model, each with its own problems. Although the convolutional neural network can extract local characteristics well, its sensory field is relatively limited; while the transformer model can capture global information, the amount of computation is too large. This allows for better extraction of different scales of features; followed by the addition of spatial and channel attention bridges where the jump connection is made, so that the feature fusion effect between the encoder and decoder is better. We did experiments on the abdominal CT data set, and the results showed that the segmentation effect of this method was very good, with DSC reaching 84.86% and HD95 being 16.78 mm. These data show that the method we proposed is very promising in clinically using multiple organs in the abdominal CT image.
Authors: Junhao Wang, Menghan Yang, Yawen Wang, Qiumei Pu
Institution: Minzu University of China
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4414288759