SIA-Unet: A Unet with Sequence Information for Gastrointestinal Tract Segmentation

Rongguang Ye, Ranmin Wang, Yantong Guo, Lei Chen

Abstract: Magnetic resonance imaging (MRI) has been widely applied to the medical imaging diagnosis of various human body systems. Deep network-based medical image segmentation techniques for MRI can help patients receive more accurate and effective diagnoses. However, multiple consecutive two-dimensional MRIs have sequence information in reality. Different organs may appear specifically in a sequence of MRI data for the body part. Therefore, MRI sequence information is the key to improving the segmentation effect in deep network architecture design. In this paper, we propose the SIA-Unet, an improved Unet network that incorporates MRI sequence information. SIA-Unet also has an attention mechanism to filter the feature map’s spatial data to extract valuable data. Extensive experiments on the UW-Madison dataset have been conducted to evaluate the performance of SIA-Unet. Experimental results have shown that with a coherent end-to-end training pipeline, SIA-Unet significantly outperforms other baselines. Our implementation is available at https://github.com/min121101/SIA-Unet.