🤖 AI Summary
To address the challenge of balancing segmentation accuracy and computational efficiency for organ segmentation in resource-constrained clinical settings (e.g., standard hospital PCs), this paper proposes LSU-Net, a lightweight deep learning architecture. Methodologically, it introduces two key innovations: (i) a novel collaborative block design—Light Conv Block and Tokenized Shift Block—that integrates depthwise separable convolutions with spatial shift mechanisms to drastically reduce parameter count; and (ii) a dynamic weighted multi-task loss function to enhance multi-scale feature representation. Evaluated on the UWMGI and MSD Colon datasets, LSU-Net achieves competitive or superior Dice scores (1.3–2.1% improvement) with only ~1.2M parameters—over 50% fewer than mainstream state-of-the-art models—while accelerating inference by 2.4×. The model thus achieves an optimal trade-off among lightness, accuracy, and deployability, offering a practical paradigm for AI deployment in resource-limited healthcare environments.
📝 Abstract
UNet and its variants have widespread applications in medical image segmentation. However, the substantial number of parameters and computational complexity of these models make them less suitable for use in clinical settings with limited computational resources. To address this limitation, we propose a novel Lightweight Shift U-Net (LSU-Net). We integrate the Light Conv Block and the Tokenized Shift Block in a lightweight manner, combining them with a dynamic weight multi-loss design for efficient dynamic weight allocation. The Light Conv Block effectively captures features with a low parameter count by combining standard convolutions with depthwise separable convolutions. The Tokenized Shift Block optimizes feature representation by shifting and capturing deep features through a combination of the Spatial Shift Block and depthwise separable convolutions. Dynamic adjustment of the loss weights at each layer approaches the optimal solution and enhances training stability. We validated LSU-Net on the UWMGI and MSD Colon datasets, and experimental results demonstrate that LSU-Net outperforms most state-of-the-art segmentation architectures.