Keywords: nnU-Net ; SE attention ; Medical image segmentation
Abstract: Medical image segmentation plays a pivotal role in clinical
diagnosis and treatment. In particular, automatic segmentation of abdominal organs is crucial for computer-aided diagnosis, surgical navigation, and various medical applications. This article introduces an improved approach based on nnU-Net, a state-of-the-art neural network
model for biomedical segmentation, with the integration of attention
mechanisms. The proposed method enhances segmentation accuracy by
introducing attention modules while preserving nnU-Net’s established
workflow. Three key preprocessing steps—image cropping, resampling,
and normalization—are detailed to prepare abdominal CT images. Additionally, SE (Squeeze-and-Excitation) attention modules are incorporated into nnU-Net to improve feature representation and semantic segmentation accuracy. Experimental results on the FLARE 2023 dataset
demonstrate the effectiveness of the proposed method, achieving a mean
Dice Similarity Coefficient of 0.411, even with limited training data.
Submission Number: 31
Loading